docs: add Claude AI assistant rules and configuration

Add comprehensive rule documentation for AI-assisted development covering
authentication surfaces, outbound-call safety invariants, and other project
conventions to guide Claude's understanding of critical system behaviors.

.claude/rules/auth-surfaces.md

@@ -0,0 +1,69 @@
+---
+description: Casdoor SSO for the browser + owner-minted PATs for MCP/CLI; owner-only on every surface; one resolver; ?token= fallback; dev-owner on loopback
+paths:
+  - "auth.py"
+  - "api/auth.py"
+  - "api/tokens.py"
+  - "api/deps.py"
+  - "api/websocket.py"
+  - "mcp_server/server.py"
+  - "main.py"
+---
+
+# Authentication across the four surfaces
+
+Auth is **Casdoor SSO for the browser + owner-minted Personal Access Tokens for
+MCP/CLI**, and the gateway is **owner-only**: exactly one operator (the Casdoor
+user whose name matches `OWNER_NAME`) may use any surface; every other identity
+gets 403. There is one resolver behind all of it — don't add a second auth
+mechanism, a per-surface token, or a bypass.
+
+- **`resolve_bearer(session, raw_token)` in [auth.py](../../auth.py) is the single
+  resolver.** It turns a bearer string into a `User` (or `None`), classifying it
+  as a PAT (`hs_pat_` prefix → hash lookup) or a Casdoor JWT (RS256, validated
+  against the endpoint's JWKS). `resolve_from_header_or_query` wraps it to accept
+  the token from the `Authorization` header *or* a `?token=` query param. Every
+  surface funnels through these — REST, WebSocket, MCP.
+
+- **Owner gating is `is_owner(user)` + `get_current_owner`.** `is_owner` matches
+  `user.name == OWNER_NAME` (SSO) or the dev-owner sub (dev mode). REST routers
+  carry `dependencies=[Depends(get_current_owner)]` (aliased `_auth` in
+  [main.py](../../main.py)) → 401 if unauthenticated, 403 if not owner. New
+  protected routers get the same dependency. `/auth/me` is the one exception: it
+  resolves the user *without* the owner gate so a signed-in non-owner sees
+  `is_owner:false` (the dashboard's "not authorized" screen) instead of a bare
+  401.
+
+- **The `?token=` query-param fallback is intentional and narrow.** Browsers
+  can't set headers on a WebSocket connect or on an `<audio src>`/`<a href>`
+  recording download, so the current token (Casdoor JWT, or a PAT) rides as
+  `?token=`. It's validated by the same resolver as the header. Don't widen it or
+  remove it without accounting for those two consumers.
+
+- **WebSocket checks ownership itself** ([api/websocket.py](../../api/websocket.py)
+  `_authorize`) rather than via a router dependency, because WS handshakes don't
+  run FastAPI dependencies the same way. It opens a `session_scope`, resolves the
+  bearer (header or `?token=`), and closes with **4401** unless the caller is the
+  owner. Keep the check **before** `websocket.accept()`.
+
+- **MCP is gated by the ASGI `_owner_only_mcp` wrapper in main.py**, not by
+  FastMCP auth. `create_mcp_server` builds `FastMCP(auth=None)`; the wrapper reads
+  the ASGI scope's `Authorization` header, resolves it via the same
+  `resolve_from_header_or_query` + `is_owner`, and short-circuits non-owner
+  requests with 401/403 (plus an RFC 9728 `WWW-Authenticate` header pointing at
+  `/.well-known/oauth-protected-resource/mcp`). This is why PATs *and* JWTs both
+  work on `/mcp` with one code path. The nested `mcp_http_app.lifespan` still runs
+  — the wrapper is pure middleware around the inner app.
+
+- **Dev mode is the loopback bypass, not a token.** `CASDOOR_ENABLED=false` makes
+  every request resolve to the dev owner — permitted **only** on a loopback bind.
+  `_check_startup_config` in [main.py](../../main.py) exits if SSO is disabled and
+  `HOST` is off-loopback (the network would see a dev-owner-open gateway), and
+  exits if SSO is enabled but `CASDOOR_ENDPOINT`/`CLIENT_ID`/`CLIENT_SECRET`/
+  `OWNER_NAME` is missing. **Never weaken these to "warn and continue" — a
+  startup misconfiguration must stop the service.**
+
+- **Never log a token.** `client_secret` is a `SecretStr`; read it via
+  `.get_secret_value()` only where you hand it to the Casdoor SDK. PAT plaintext
+  is shown once at creation and only its SHA-256 hash is stored — never log the
+  plaintext, a JWT, or a hash.

.claude/rules/call-safety.md

@@ -0,0 +1,45 @@
+---
+description: emergency-number guard, concurrent-call cap, dial-path discipline — the outbound safety invariants
+paths:
+  - "core/dial_plan.py"
+  - "core/gateway.py"
+  - "mcp_server/server.py"
+  - "api/calls.py"
+---
+
+# Outbound-call safety
+
+This is the safety core. A defect here means an unwanted real phone call, a
+runaway telephony bill, or — the one that matters most — an AI-initiated
+emergency call that should have been impossible.
+
+- **`is_emergency_number()` is the single source of truth for refusal.** It
+  lives in `core/dial_plan.py`, blocks `911`/`9911`/`112` and their E.164
+  mappings (`_BLOCKED` = keys ∪ values), and normalises the input (strips
+  spaces, dashes, dots) before comparing. If you learn of another dialled form
+  that reaches emergency services, add it to `EMERGENCY_NUMBERS` — never work
+  around the guard.
+
+- **Every outbound path goes through `gateway.make_call`, and the guard is its
+  first check** — before the concurrency cap, before `create_call`, before any
+  SIP action. REST `make_call`, MCP `make_call`, receptionist ring-back, and any
+  transfer to an external number must funnel through it. **Do not introduce a
+  dial path that reaches `sip_engine.make_call` without passing the guard
+  first.** If a new feature needs to place a call, it calls `gateway.make_call`.
+
+- **The concurrency cap is spend control, not decoration.** `max_concurrent_calls`
+  (default 4) is checked in `make_call` after the emergency guard and before
+  call creation, using `len(call_manager.active_calls)`. Keep the ordering:
+  refuse-emergency, then cap, then create. Don't move the count to after
+  creation (it would off-by-one) and don't remove it.
+
+- **Refusals raise `ValueError` at the gateway; surfaces translate it.**
+  `make_call` raises `ValueError` for both refusals; the MCP tool converts it to
+  `ToolError`, and the REST layer maps it to a 4xx. Keep refusals as exceptions
+  from the gateway — a refused call must never look like a placed one.
+
+- **The README's `[!CAUTION]` block is a contract, not decoration.** If you
+  change refusal behaviour, the README caution and this rule must stay true. A
+  system that quietly stops refusing emergency numbers is a serious regression
+  even if every test still passes — add/keep a test that asserts each blocked
+  form is refused.

.claude/rules/concurrency-threads.md

@@ -0,0 +1,55 @@
+---
+description: the Sippy/PJSUA2 OS-thread boundary — two funnels, who owns what state, never cross it directly
+paths:
+  - "core/sippy_engine.py"
+  - "core/sip_engine.py"
+  - "core/media_pipeline.py"
+  - "core/call_manager.py"
+  - "core/gateway.py"
+---
+
+# The thread boundary (the invariant that keeps this app sane)
+
+The README says "single-process async." That's true at the surface, but under
+the SIP engine there are **two execution contexts**: the asyncio event loop, and
+a dedicated **Sippy/PJSUA2 OS thread** running the `ED2` event dispatcher. Almost
+every hard-to-debug class of bug in a telephony gateway comes from touching one
+context's state from the other. This app avoids that with exactly one funnel each
+way. Preserve them.
+
+- **Who owns what:**
+  - *Sippy thread* owns the Sippy UA objects and the `ED2` dispatcher. State:
+    `_ed_ua_to_leg`, `_ed_leg_to_ua` (the "ED-thread-owned state" maps). Only
+    touch these from a Sippy handler or a `_run_on_sippy` closure.
+  - *asyncio loop* owns everything else: `_legs`, `_bridges`,
+    `_registered_devices`, the `EventBus`, the `CallManager`, the media pipeline
+    wiring.
+
+- **Cross thread → loop only via `_post_from_ed`.** It calls
+  `asyncio.run_coroutine_threadsafe(self._on_engine_event(kind, data), self._loop)`.
+  `_on_engine_event` is **the single funnel** where Sippy-thread events mutate
+  loop-owned state, and it runs on the loop. New Sippy-side events post through
+  here with a new `kind`; they do **not** reach into `_legs`/`EventBus` directly
+  from the handler.
+
+- **Cross loop → thread only via `_run_on_sippy`.** It uses `ED2.callFromThread(fn)`
+  so `fn` runs where the Sippy objects live. In simulation mode (no `sippy`
+  import) it runs `fn` inline — keep that fallback so tests and stub mode work
+  without the native library. Anything that manipulates a UA object goes through
+  here.
+
+- **Never:** read/write a Sippy UA object from the loop; never mutate `_legs`,
+  publish an event, or touch the `CallManager` from inside a raw Sippy callback
+  without going through `_post_from_ed`. If you find yourself wanting to, you're
+  about to introduce a data race — add a `kind` to the funnel instead.
+
+- **Background tasks are tracked, both sides.** The gateway's `spawn()` and the
+  engine's `_spawn()` add tasks to a `_tasks` set with a done-callback that
+  discards them, so shutdown can cancel them and the GC can't drop a live
+  coroutine. Launch per-call/background work through these, not a bare
+  `asyncio.create_task` you forget to hold a reference to.
+
+- **`MockSIPEngine` has no thread.** Tests run against it; it satisfies the same
+  `SIPEngine` interface synchronously/async-inline. When you extend the real
+  engine's behaviour, extend the mock to match, or tests will pass against a
+  fiction.

.claude/rules/config-startup.md

@@ -0,0 +1,63 @@
+---
+description: pydantic-settings nested sub-configs + get_settings() singleton, SecretStr discipline, startup refusals, .env hygiene
+paths:
+  - "config.py"
+  - "main.py"
+  - ".env*"
+---
+
+# Config & startup
+
+Config is `Settings` in [config.py](../../config.py): a root `BaseSettings` with
+**nested sub-config models**, each carrying its own `env_prefix`. Read it through
+the `get_settings()` cached singleton.
+
+- **`get_settings()` is the only accessor.** It memoises a single `Settings()`.
+  Don't construct `Settings()` elsewhere, and don't reach for `os.environ.get`
+  for Hold Slayer config — the whole point of the sub-config layout is that every
+  knob has one typed home.
+
+- **Sub-configs own their prefixes.** `SIP_TRUNK_*` → `SIPTrunkSettings`, `LLM_*`
+  → `LLMSettings`, `TTS_*` → `TTSSettings`, `RECEPTIONIST_*` →
+  `ReceptionistSettings`, `CASDOOR_*` → `CasdoorSettings`, `CLASSIFIER_*`,
+  `SPEACHES_*`, `GATEWAY_SIP_*`. Root vars (`DATABASE_URL`, `HOST`, `PORT`,
+  `MAX_CONCURRENT_CALLS`, `USE_MOCK_SIP`, `NOTIFY_SMS_NUMBER`, `DEBUG`,
+  `LOG_LEVEL`, and the auth cross-cutters `OWNER_NAME` + `PUBLIC_BASE_URL`) are
+  unprefixed on the root model. A new knob goes in the sub-config it belongs to;
+  a genuinely new subsystem gets its own sub-config + prefix, not flat root vars.
+  - `OWNER_NAME` (the owner's Casdoor username) and `PUBLIC_BASE_URL` (OAuth
+    discovery base) live on the root, not under `CASDOOR_`, because they cross-cut
+    every surface — like `DATABASE_URL`. The Casdoor *connection* knobs
+    (`enabled`/`endpoint`/`client_id`/`client_secret`/`org_name`/`app_name`) live
+    under `CASDOOR_`.
+  - `HoldSlayerSettings` uses `env_prefix_allow_empty=True` with explicit
+    `validation_alias`es (`DEFAULT_TRANSFER_DEVICE`, `MAX_HOLD_TIME`,
+    `HOLD_CHECK_INTERVAL`) — i.e. those three are read *unprefixed* by design.
+    Follow that pattern only if you deliberately want an unprefixed name.
+
+- **Secrets are `SecretStr`.** `casdoor.client_secret`, `sip_trunk.password`,
+  `llm.api_key`, `tts.api_key`. Keep new secrets as `SecretStr`; call
+  `.get_secret_value()` only at the point of use (outbound header, SDK
+  construction) — never store the bare string, never log it.
+
+- **Startup refuses bad configs loudly, then exits.** In [main.py](../../main.py):
+  - `_check_startup_config` exits if `DATABASE_URL` is unset; if `CASDOOR_ENABLED`
+    is true but any of `CASDOOR_ENDPOINT`/`CLIENT_ID`/`CLIENT_SECRET`/`OWNER_NAME`
+    is missing; or if `CASDOOR_ENABLED` is false while `HOST` is off-loopback
+    (dev-owner mode would be open to the network). See the
+    [auth-surfaces rule](auth-surfaces.md).
+  - `_handle_db_error` turns raw asyncpg failures into human-readable guidance
+    (wrong password, missing DB, connection refused, bad hostname) and `sys.exit(1)`.
+  - SIP engine build failure and mock-vs-real are surfaced, not swallowed.
+  **Keep the pattern: a misconfiguration stops the service with a message a human
+  can act on — never a silent degrade or a stack trace with no guidance.**
+
+- **`use_mock_sip` is opt-in for a reason.** An unconfigured trunk without
+  `USE_MOCK_SIP=true` must fail startup rather than boot a gateway that silently
+  can't place real calls. Don't default it to `True`.
+
+- **`.env` hygiene:** `.env` is gitignored and holds real secrets — never commit
+  it, never treat the checked-out `.env` as a template. Only `.env.example`
+  (placeholders) is committed, and it must stay in sync with the models here and
+  the README config table. Every new var lands in all three: model,
+  `.env.example`, README.

.claude/rules/lifespan-wiring.md

@@ -0,0 +1,56 @@
+---
+description: composition-root lifespan, nested MCP http_app lifespan, app.state wiring, route/mount ordering, honest /health
+paths:
+  - "main.py"
+  - "api/deps.py"
+  - "core/gateway.py"
+---
+
+# Lifespan, composition root & route ordering
+
+[main.py](../../main.py)'s `lifespan` is the **composition root**: it builds the
+gateway and every service, wires them by constructor/registration, and hangs the
+long-lived ones on `app.state`. This is the one place dependencies are
+assembled.
+
+- **Build services here, inject them — nothing self-constructs its deps.** The
+  gateway, classifier, transcription, TTS, routing, recording, receptionist, and
+  notification services are all constructed in the lifespan and wired together
+  (e.g. the receptionist receives tts/transcription/recording/routing;
+  `launch_hold_slayer` is registered as the `HOLD_SLAYER` mode handler). A new
+  service is built here and passed in, not instantiated deep in a call path.
+
+- **The MCP sub-app's lifespan MUST be nested.** The lifespan opens
+  `async with mcp_http_app.lifespan(app):` around all startup. FastMCP's
+  streamable-HTTP session manager is initialised inside *its* lifespan; mount the
+  app without entering that context and every `/mcp` request 500s
+  ("session manager not initialised" / "Task group is not initialized"). **Keep
+  the nesting.** This is the same landmine across the estate's mounted-MCP
+  services.
+
+- **`app.state` is the handoff to request handlers.** The lifespan sets
+  `app.state.gateway`, `.routing_service`, `.transcription_service`,
+  `.notification_service`, `.recording_service`. Dependencies in
+  [api/deps.py](../../api/deps.py) read these and raise `503` if not yet set. MCP
+  tools reach the gateway via the lazy `_get_gateway_instance` resolver. Don't
+  reach for module-level globals; go through `app.state`.
+
+- **Route/mount registration order is load-bearing:**
+  1. `call_history` router registers **before** `calls` — both live under
+     `/api/v1/calls`, and `calls`' `GET /{call_id}` would otherwise swallow the
+     literal path `history`. Keep history first.
+  2. The `"/mcp"` mount and all API/WS/health routes register **before** the
+     `"/"` static dashboard mount — a root mount matches every path, so anything
+     after it is unreachable. The dashboard mount stays last, and only when
+     `dashboard/build/` exists.
+
+- **`/health` is honest by construction.** `healthy` = real (non-`MockSIPEngine`)
+  engine **and** registered trunk **and** reachable DB; it also reports STT/TTS
+  last-known reachability via `_availability`. Don't relax any of these to make a
+  probe pass — a degraded gateway must read as `degraded`, with the reason
+  visible.
+
+- **Shutdown reverses startup.** Stop notifications, stop the gateway (which
+  cancels tracked tasks, ends active calls, stops SIP then media), close the DB.
+  New long-lived resources get a matching teardown here — don't leak a task or a
+  client across restarts.

.claude/rules/mcp-tools.md

@@ -0,0 +1,56 @@
+---
+description: MCP tools return formatted strings; ToolError-vs-return-string convention; lazy gateway resolution; resources are JSON
+paths:
+  - "mcp_server/server.py"
+---
+
+# MCP tools & resources
+
+The MCP server ([mcp_server/server.py](../../mcp_server/server.py)) is the
+AI-assistant control surface. It's built by `create_mcp_server(get_gateway)` and
+mounted at `/mcp/` before the lifespan runs, so everything is resolved lazily.
+
+- **Auth is the ASGI `_owner_only_mcp` wrapper in [main.py](../../main.py), not
+  FastMCP.** `create_mcp_server` builds `FastMCP(auth=None)`; the wrapper resolves
+  the `Authorization` bearer (Casdoor JWT or owner-minted PAT) via the shared
+  `resolve_from_header_or_query` + `is_owner` and returns 401/403 before the inner
+  app runs. Don't reintroduce a FastMCP verifier here — one resolver gates all
+  four surfaces (see the [auth-surfaces rule](auth-surfaces.md)). MCP clients use
+  a PAT (`hs_pat_…`) minted from the dashboard's Tokens modal.
+
+- **Tools return plain formatted strings; resources return JSON strings.** Tools
+  produce human-readable text an assistant reads back to a user (`"Call abc123
+  initiated. …"`). Resources (`gateway://status`, `gateway://call-flows`,
+  `gateway://active-calls`) return `json.dumps(...)`. Don't blur these — a tool
+  that returns raw JSON, or a resource that returns prose, breaks the contract.
+
+- **Error convention — match the two existing patterns:**
+  - **Raise `ToolError`** when the request is invalid or unsafe: emergency
+    number, bad mode, concurrency cap hit, or "gateway still starting"
+    (`require_gateway`). The assistant should treat these as errors.
+  - **Return an error string** for a lookup that simply found nothing or hit a
+    recoverable snag: `"Call {id} not found."`, `"No stored call flow for …"`,
+    `"Error looking up …: {e}"`. The assistant reads these as content.
+  - Rule of thumb: *"you asked for something invalid/unsafe" → raise; "I looked,
+    here's the (maybe empty/failed) answer" → return.*
+
+- **`require_gateway()` gates every tool that needs the live gateway.** It raises
+  `ToolError("Gateway is still starting up …")` when `get_gateway()` returns
+  `None`. This is why the MCP app can mount before the lifespan builds the
+  gateway. Call it at the top of any tool that touches the gateway; never assume
+  the gateway exists.
+
+- **`make_call` is the one tool that dials.** It maps the string `mode` to
+  `CallMode`, defaults unknown modes to `DIRECT`, and lets `gateway.make_call`'s
+  refusals (`ValueError`) surface as `ToolError`. The emergency guard and
+  concurrency cap live in the gateway, **not** here — don't reimplement or skip
+  them at the tool layer (see the call-safety rule).
+
+- **DB-backed tools use `session_scope()`** and read from `call_persistence`.
+  Completed-call history, summaries, recordings, and stored flows come from the
+  database, not from `active_calls` (those are live only). Keep the
+  `async with session_scope() as session:` pattern; don't open ad-hoc sessions.
+
+- **Keep the tool count and README table in sync.** There are 15 tools + 3
+  resources. If you add/remove one, update the README's MCP table and the
+  `docs/mcp-server.md` reference — a drifting tool list is a documented lie.

.env.compose.example

@@ -0,0 +1,34 @@
+# Compose environment for `docker compose up` — copy to `.env` and fill in.
+#
+#   cp .env.compose.example .env
+#
+# These values are substituted into docker-compose.yaml (${VAR}); they are NOT
+# baked into the image (.env is gitignored and in .dockerignore). Distinct from
+# the app's own .env used for a bare `uvicorn` run.
+
+# --- Database (the bundled postgres:17 service) ---
+HS_DB_USER=holdslayer
+HS_DB_PASSWORD=change-me
+HS_DB_NAME=holdslayer
+
+# --- Published port on the host ---
+HS_APP_PORT=21081
+
+# --- SIP: mock by default (dev/local). Set false + fill SIP_TRUNK_* for a real trunk. ---
+USE_MOCK_SIP=true
+
+# --- Auth: Casdoor SSO (owner-only) ---
+# Required: this stack publishes the port on 0.0.0.0, so dev-owner mode
+# (CASDOOR_ENABLED=false) is refused at startup — it's loopback-only. Register a
+# `hold-slayer` app in Casdoor (org heluca, redirect URI <PUBLIC_BASE_URL>/auth/callback).
+CASDOOR_ENABLED=true
+CASDOOR_ENDPOINT=https://id.ouranos.helu.ca
+CASDOOR_CLIENT_ID=
+CASDOOR_CLIENT_SECRET=
+CASDOOR_ORG_NAME=heluca
+CASDOOR_APP_NAME=hold-slayer
+# The owner's Casdoor username — the only identity allowed on any surface.
+OWNER_NAME=
+# Public base URL the browser reaches (drives OAuth discovery + the Casdoor
+# redirect_uri). E.g. http://localhost:21081 for a local run.
+PUBLIC_BASE_URL=http://localhost:21081

CLAUDE.md

@@ -0,0 +1,240 @@
+# CLAUDE.md — Hold Slayer 🔥🐾
+
+Red Panda Standards for the Hold Slayer telephony gateway. This is an
+**AI-powered PSTN gateway**: it places real phone calls, navigates IVR menus,
+waits on hold, and rings a human's desk phone when a live person answers. It
+also answers inbound calls with an AI receptionist and smart routing.
+
+**It dials real numbers on a real SIP trunk and may incur telephony charges,
+and an AI agent drives it.** Treat every change through that lens: a bug here
+isn't a 500, it's an unwanted phone call — or a *refused emergency call that
+should never have been attempted in the first place*. Read this file before
+touching call placement, auth, the SIP thread boundary, or the emergency guard.
+
+Lead with a paw print in this repo.
+
+---
+
+## The shape of this thing (read once, then it's obvious)
+
+One FastAPI process exposes four surfaces over the same port: **REST** (`/api/v1/*`),
+**WebSocket** (`/ws/*`), an **MCP server** (streamable HTTP at `/mcp/`), and the
+built **SvelteKit dashboard** at `/`. All four are **owner-only**, gated by Casdoor
+SSO (browser JWT) or an owner-minted PAT through one shared resolver.
+
+Under them sits a **composition root** in [main.py](main.py)'s `lifespan`: the
+gateway and every service are constructed and wired there, then hung on
+`app.state`. Nothing constructs its own dependencies — if you need a new
+service, build it in the lifespan and pass it in.
+
+Below the services is the part that makes this app unusual: a **`SippyB2BUAEngine`
+that runs the SIP/PJSUA2 event loop on its own OS thread**, not the asyncio loop.
+This is the highest-leverage invariant in the codebase. See
+[the concurrency rule](.claude/rules/concurrency-threads.md) — the README's
+"single-process async" line is a simplification; there are two execution
+contexts and exactly one funnel between them.
+
+```
+REST / WS / MCP / Dashboard   (asyncio, FastAPI)
+        │
+   composition root (lifespan) → services → gateway
+        │
+   SippyB2BUAEngine ──┬── asyncio side (legs, bridges, event bus)
+                      └── Sippy thread (UA objects, ED2 dispatcher)
+                           ↑ crossed only via _post_from_ed / _run_on_sippy
+```
+
+---
+
+## Red Panda Approval™ — what it means here
+
+1. **Fresh Environment Test** — `cp .env.example .env`, set `DATABASE_URL` and
+   either `CASDOOR_*` + `OWNER_NAME` (SSO) or `CASDOOR_ENABLED=false` with
+   `HOST=127.0.0.1` (dev-owner, loopback only), `pip install -e ".[dev]"`,
+   `uvicorn main:app`. It must boot to a clear log banner or **exit with a
+   human-readable reason** (see `_check_startup_config` / `_handle_db_error` in
+   [main.py](main.py)). It must *never* boot into a state where it silently can't
+   place calls — that's why `use_mock_sip` is opt-in and an unconfigured trunk
+   fails startup.
+2. **Elegant Simplicity** — the composition root wires; services do one job;
+   the thread boundary has exactly one funnel each way. Don't add a second path
+   across the thread line, a second auth mechanism, or a service that reaches
+   into another service's internals.
+3. **Observable & Debuggable** — `/health` is *honest*: it reports `degraded`
+   with the reason (mock engine, unregistered trunk, DB down, STT/TTS
+   unreachable) rather than a green light that lies. Keep it honest. Events flow
+   through the typed `EventBus`; new call-lifecycle facts become typed events,
+   not `print`s.
+4. **Consistent Patterns** — config via pydantic-settings sub-configs; MCP tools
+   return formatted strings; REST returns Pydantic models; DB access via
+   `session_scope()`. Match the neighbours.
+5. **Actually Works** — `pytest tests/ -v` (146 tests across 16 files). A change
+   to call placement, routing, the classifier, or auth needs a test. The suite
+   runs against SQLite (`aiosqlite`) and the mock SIP engine — no trunk, no
+   Postgres required to test.
+
+---
+
+## Invariants that must not be weakened
+
+These are safety- and correctness-critical. Loosening one is never a casual
+refactor — it needs an explicit rationale and, where it deviates from a stated
+standard, a note (there is no `docs/EXCEPTIONS.md` yet; if you start
+accumulating documented deviations, create one rather than letting them go
+unrecorded).
+
+- **The emergency-number guard is absolute.** `is_emergency_number()` in
+  [core/dial_plan.py](core/dial_plan.py) blocks `911`/`9911`/`112` (and their
+  E.164 forms, whitespace/dashes stripped) at `gateway.make_call`, **before**
+  the concurrency check and before any SIP action. Every dialling path — REST
+  `make_call`, MCP `make_call`, receptionist call-back, transfer to an external
+  number — must pass through a guard that refuses these. An AI agent must never
+  place an emergency call, and API calls carry no E911 location. **Do not add a
+  dial path that bypasses `make_call`'s guard.** If you add a new outbound path,
+  it calls the guard first. See [the safety rule](.claude/rules/call-safety.md).
+
+- **The concurrency cap is real spend control.** `max_concurrent_calls` (default
+  4) caps simultaneous outbound calls in `gateway.make_call`. It's checked after
+  the emergency guard, before creating the call. Don't remove it or move it
+  below call creation.
+
+- **The Sippy thread boundary is crossed only through the two funnels.** Sippy
+  UA objects and the `ED2` dispatcher live on the Sippy thread; legs, bridges,
+  and the event bus live on the asyncio loop. Cross **thread→loop** only via
+  `_post_from_ed` (→ `run_coroutine_threadsafe` → the single `_on_engine_event`
+  funnel) and **loop→thread** only via `_run_on_sippy` (→ `ED2.callFromThread`).
+  Never touch a Sippy UA object from the loop; never mutate loop-owned state from
+  a Sippy handler. This is the whole reason the app is thread-safe.
+
+- **Auth is Casdoor SSO + owner-minted PATs, owner-only, one resolver.** The
+  browser signs in via Casdoor (short-lived JWT); MCP/CLI clients use owner-minted
+  PATs (`hs_pat_…`). `resolve_bearer` in [auth.py](auth.py) turns either into a
+  `User`, and `is_owner`/`get_current_owner` gate every surface to the single
+  operator (`OWNER_NAME`) — non-owners get 403. Dev mode (`CASDOOR_ENABLED=false`)
+  resolves the dev owner and is **only** permitted on a loopback bind;
+  `_check_startup_config` refuses SSO-off-loopback and SSO-on-with-missing-config.
+  Keep those refusals — a silent dev-owner-open-on-0.0.0.0 is the failure mode
+  they exist to prevent. See [the auth rule](.claude/rules/auth-surfaces.md).
+
+- **`/health` tells the truth.** `healthy` requires a real (non-mock) engine, a
+  registered trunk, and a reachable DB. Don't relax it to make a probe go green;
+  a gateway that can't place calls is not healthy, and the dashboard/operator
+  needs to see that.
+
+- **The database is the source of truth for history; live state is in memory.**
+  Active calls live in the `CallManager`; completed calls + transcripts are
+  persisted on hangup via `call_persistence`. MCP/REST history and summaries read
+  from the DB through `session_scope()`. Don't confuse the two — a call that
+  ended is gone from `active_calls` and only exists in the DB.
+
+---
+
+## Config — pydantic-settings, nested, singleton
+
+Config is `Settings` in [config.py](config.py): a root `BaseSettings` with
+**nested sub-config models** (`SIPTrunkSettings`, `LLMSettings`, `TTSSettings`,
+`ReceptionistSettings`, …), each with its own `env_prefix`
+(`SIP_TRUNK_`, `LLM_`, `TTS_`, `RECEPTIONIST_`, …). Read config through
+`get_settings()` (a cached singleton) — don't call `os.environ.get` for
+Hold Slayer settings, and don't construct `Settings()` yourself outside that
+accessor.
+
+> **Estate note (not a defect):** unlike the `KERNOS_`/`ARGOS_`/`NIKE_`-style
+> single-prefix services, Hold Slayer has **no one umbrella prefix** — root vars
+> (`DATABASE_URL`, `HOST`, `MAX_CONCURRENT_CALLS`, `OWNER_NAME`) are unprefixed
+> and each subsystem carries its own (`CASDOOR_` for the SSO connection). That's a
+> deliberate readability choice for
+> a config with this many subsystems; keep new vars consistent with the
+> sub-config they belong to, and if you add a new subsystem, give it its own
+> sub-config + prefix rather than piling flat vars onto the root.
+
+Secrets (`casdoor.client_secret`, `sip_trunk.password`, `llm.api_key`,
+`tts.api_key`) are
+`SecretStr` — keep them so, and read via `.get_secret_value()` only at the point
+of use. Every new var also needs a row in [.env.example](.env.example) and the
+README's config table. See [the config rule](.claude/rules/config-startup.md).
+
+**`.env` hygiene:** `.env` is gitignored and holds real secrets — never commit
+it, never treat the local `.env` as a template. Only `.env.example`
+(placeholders) is committed.
+
+---
+
+## MCP tools — formatted strings, and the error convention to know
+
+MCP tools ([mcp_server/server.py](mcp_server/server.py)) return **plain
+human-readable strings** (not JSON, not Pydantic models — that's the REST
+layer's job). Resources (`gateway://status`, `gateway://call-flows`,
+`gateway://active-calls`) return JSON strings.
+
+There's a **deliberate but uneven error convention** worth understanding before
+you add a tool:
+
+- `make_call` raises `ToolError` on a bad request (emergency number, bad mode,
+  cap hit) — a hard failure the assistant should treat as an error.
+- Most read/lookup tools **return** an error *string* (`"Call … not found."`,
+  `"Error looking up …: {e}"`) instead of raising — a soft "here's what
+  happened" the assistant reads as content.
+
+When you add a tool: raise `ToolError` for "you asked for something invalid or
+unsafe"; return a plain string for "I looked and here's the (possibly empty)
+answer." The gateway is resolved lazily per call via `require_gateway()` (it
+raises `ToolError` while the gateway is still starting) — keep that, because the
+MCP app is mounted before the lifespan builds the gateway. See
+[the MCP rule](.claude/rules/mcp-tools.md).
+
+---
+
+## What's real vs. stubbed (don't mistake one for the other)
+
+- **PJSUA2 media pipeline runs in stub mode** unless the `pjsua2` bindings are
+  built from pjproject (not pip-installable). Signaling works; audio
+  routing/recording is a no-op stub without them. Code that assumes real audio
+  must degrade honestly, and `/health`/engine-mode must reflect stub vs real.
+- **The mock SIP engine (`MockSIPEngine`) must be asked for** (`USE_MOCK_SIP=true`).
+  It exists for tests and local dev. Production must not silently run on it —
+  `/health` reports `engine: mock` and refuses `healthy`.
+
+---
+
+## Known gaps (flag, don't fold — these are not your task unless asked)
+
+Surfaced honestly so you don't rediscover them as surprises. The README's
+Phase 4/5/6 checklists track most of these; **don't fold fixes into unrelated
+work** — raise them.
+
+- **No structured JSON logging.** Logging is plain `logging.basicConfig` in
+  [main.py](main.py); there's no `LOG_FORMAT`/JSON path (README Phase 4 has this
+  unchecked). If Heluca observability wants JSON logs shipped to a collector,
+  that's a deliberate piece of work, not a drive-by.
+- **No `/metrics` endpoint and no Prometheus.** Unlike the metrics-bearing
+  estate services, there's no exposition endpoint here yet.
+- **No health-probe access-log filter.** Every `/health` poll hits the access
+  log. Other estate services suppress probe noise; this one doesn't.
+- **No rate limiting** on API endpoints (README Phase 4, unchecked).
+- **Docker: single-image `Dockerfile` + `docker-compose.yaml`** (app +
+  `postgres:17`) ship in-repo; the Gitea CI (`cve-scan-docker-build.yml`) builds
+  the image on push to `main`. The compose stack requires SSO enabled
+  (published port ⇒ 0.0.0.0 ⇒ dev-owner mode refused). No systemd unit in-repo.
+- **A committed `.DS_Store` is *not* present** (good), but do check `git status`
+  stays clean of OS cruft; `.gitignore` already lists it.
+
+---
+
+## Working here
+
+- **Run:** `uvicorn main:app --host 0.0.0.0 --port 8000` (or `python main.py`).
+  The CLI `--port` wins over `settings.port` in the startup banner logic — a real
+  gotcha the banner code already accounts for; don't "fix" it into disagreement.
+- **Test:** `pytest tests/ -v`. Fast, no external services (SQLite + mock SIP).
+- **Lint:** `ruff check .` (line length 100, `py312` target).
+- **Dashboard:** `cd dashboard && npm install && npm run build` → served at `/`
+  when `dashboard/build/` exists. The API/WS/health/MCP routes are registered
+  **before** the `"/"` static mount because a root mount matches everything —
+  keep that ordering.
+- **DB migrations:** Alembic (`db/migrations/`), upgrade-on-boot via `init_db`.
+  Schema changes are migrations, never ad-hoc `CREATE`.
+
+Path-scoped rules in [.claude/rules/](.claude/rules/) load automatically when you
+edit the files they cover. They carry the fine-grained "don't break this"
+detail; this file is the map.

README.md

@@ -172,13 +172,17 @@ pip install -e ".[dev]"
 ```bash
 cp .env.example .env
 # Edit .env with your SIP trunk credentials, LLM endpoint, etc.
-# Required: DATABASE_URL, and API_TOKEN unless HOST=127.0.0.1
-openssl rand -hex 32   # → API_TOKEN
+# Required: DATABASE_URL, plus either the Casdoor SSO settings
+# (CASDOOR_* + OWNER_NAME) or CASDOOR_ENABLED=false with HOST=127.0.0.1.
 ```
 
-All REST, WebSocket, and MCP access requires `Authorization: Bearer
-$API_TOKEN` (WebSocket also accepts `?token=...`). An empty token is only
-permitted when bound to loopback.
+The gateway is **owner-only**. The browser dashboard signs in via **Casdoor
+SSO** (short-lived JWT); MCP and CLI clients use a **Personal Access Token**
+(`hs_pat_…`) minted from the dashboard's *API Tokens* menu. Both are presented as
+`Authorization: Bearer <token>` (WebSocket and `<audio>` recording downloads also
+accept `?token=…`). Only the user whose Casdoor username matches `OWNER_NAME` may
+use any surface — everyone else gets 403. With `CASDOOR_ENABLED=false` the gateway
+runs in dev-owner mode, permitted **only** on a loopback bind.
 
 ### 3. Build the dashboard (optional but recommended)
 
@@ -204,6 +208,27 @@ uvicorn main:app --host 0.0.0.0 --port 8000
 pytest tests/ -v
 ```
 
+## Docker
+
+A single image bundles the FastAPI process and the built dashboard (the node
+stage compiles the SPA; `pjsua2` is deliberately not built, so the media
+pipeline runs in stub mode — see the `Dockerfile` header). `docker-compose.yaml`
+brings up the app plus its own PostgreSQL:
+
+```bash
+cp .env.compose.example .env
+# Fill in HS_DB_PASSWORD, and CASDOOR_CLIENT_ID/SECRET + OWNER_NAME.
+docker compose up --build
+# → http://localhost:21081
+```
+
+Because the published port binds the app to `0.0.0.0`, the compose stack must run
+with **Casdoor SSO enabled** — dev-owner mode (`CASDOOR_ENABLED=false`) is
+loopback-only and is refused at startup here. Register a `hold-slayer` app in
+Casdoor (org `heluca`, redirect URI `<PUBLIC_BASE_URL>/auth/callback`) first. The
+image runs with `USE_MOCK_SIP=true` by default (a real trunk needs the
+`SIP_TRUNK_*` vars and `USE_MOCK_SIP=false`).
+
 ## Usage
 
 ### REST API
@@ -212,7 +237,7 @@ pytest tests/ -v
 
 ```bash
 curl -X POST http://localhost:8000/api/v1/calls/hold-slayer \
-  -H "Authorization: Bearer $API_TOKEN" \
+  -H "Authorization: Bearer $TOKEN" \
   -H "Content-Type: application/json" \
   -d '{
     "number": "+18005551234",
@@ -264,7 +289,7 @@ curl -X PATCH http://localhost:8000/api/v1/routing/devices/dev_abc123/dnd \
 ### WebSocket — Real-Time Events
 
 ```javascript
-const ws = new WebSocket(`ws://localhost:8000/ws/events?token=${API_TOKEN}`);
+const ws = new WebSocket(`ws://localhost:8000/ws/events?token=${token}`);
 ws.onmessage = (msg) => {
   const event = JSON.parse(msg.data);
   // event.type: "human_detected", "hold_detected", "ivr_step", etc.
@@ -276,11 +301,12 @@ ws.onmessage = (msg) => {
 ### MCP — AI Assistant Integration
 
 The MCP server is served over **streamable HTTP at `/mcp/`** (note the
-trailing slash) and authenticates with the same bearer token:
+trailing slash) and authenticates with an owner-minted Personal Access Token
+(mint one from the dashboard's *API Tokens* menu — it starts with `hs_pat_`):
 
 ```bash
 claude mcp add hold-slayer --transport http http://localhost:8000/mcp/ \
-  --header "Authorization: Bearer $API_TOKEN"
+  --header "Authorization: Bearer hs_pat_..."
 ```
 
 It exposes 15 tools and 3 resources (`gateway://status`,
@@ -333,7 +359,14 @@ All configuration is via environment variables (see `.env.example`):
 | Variable | Description | Default |
 |----------|-------------|---------|
 | `DATABASE_URL` | PostgreSQL connection string | — (required) |
-| `API_TOKEN` | Static bearer token for REST/WS/MCP | — (required unless `HOST=127.0.0.1`) |
+| `CASDOOR_ENABLED` | Enable Casdoor SSO (false → dev-owner, loopback only) | `false` |
+| `CASDOOR_ENDPOINT` | Casdoor base URL | `https://id.ouranos.helu.ca` |
+| `CASDOOR_CLIENT_ID` | Casdoor application client ID | — (required if SSO on) |
+| `CASDOOR_CLIENT_SECRET` | Casdoor application client secret | — (required if SSO on) |
+| `CASDOOR_ORG_NAME` | Casdoor organization | `heluca` |
+| `CASDOOR_APP_NAME` | Casdoor application name | — |
+| `OWNER_NAME` | Casdoor username of the single operator (owner) | — (required if SSO on) |
+| `PUBLIC_BASE_URL` | Public base URL for OAuth discovery (else derived from headers) | — |
 | `MAX_CONCURRENT_CALLS` | Cap on simultaneous outbound calls | `4` |
 | `SIP_TRUNK_HOST` | Your SIP provider hostname | — |
 | `SIP_TRUNK_USERNAME` | SIP auth username | — |
@@ -408,13 +441,13 @@ Full documentation is in [`/docs`](docs/README.md):
 ### Phase 4: Production Hardening 🚧
 
 - [x] Alembic database migrations (baseline + upgrade-on-boot)
-- [x] API authentication — static bearer token across REST/WS/MCP
+- [x] API authentication — Casdoor SSO (browser JWT) + owner-minted PATs, owner-only across REST/WS/MCP
 - [x] Emergency-number guard + concurrent-call cap on outbound calls
 - [ ] Rate limiting on API endpoints
 - [ ] Structured JSON logging
 - [x] Honest /health — engine mode, DB ping, trunk registration, STT/TTS availability
 - [ ] Graceful degradation (classifier works without STT, etc.)
-- [ ] Docker Compose (Hold Slayer + PostgreSQL)
+- [x] Docker Compose (Hold Slayer + PostgreSQL)
 
 ### Phase 5: Additional Services 🚧
 

api/auth.py

@@ -0,0 +1,191 @@
+"""
+OIDC authentication endpoints (Casdoor SSO).
+
+GET  /auth/login            → redirect to Casdoor authorization URL
+GET  /auth/callback         → exchange code for tokens, redirect to UI with token
+GET  /auth/me               → return current user info (requires Bearer token)
+GET  /auth/silent-refresh   → hidden-iframe refresh (re-auth with existing session)
+GET  /auth/refresh-callback → post the refreshed token to the parent window
+GET  /auth/logout           → redirect to Casdoor logout URL
+
+The dashboard is owner-only; ``/auth/me`` returns ``is_owner`` so a signed-in
+non-owner sees an "access denied" screen instead of a bare 401.
+"""
+
+import secrets
+from urllib.parse import urlencode
+
+from fastapi import APIRouter, HTTPException, Query, Request
+from fastapi.responses import HTMLResponse, JSONResponse, RedirectResponse
+
+from auth import get_sdk, is_owner, resolve_from_header_or_query
+from config import get_settings
+from db.database import session_scope
+
+router = APIRouter(prefix="/auth", tags=["auth"])
+
+
+def _build_casdoor_auth_url(
+    callback: str,
+    *,
+    scope: str = "openid profile email",
+    state: str | None = None,
+    prompt: str | None = None,
+) -> str:
+    """Build the Casdoor authorization URL directly.
+
+    The SDK's get_auth_link() doesn't support the ``prompt`` parameter that
+    silent refresh needs, so build the URL manually.
+    """
+    c = get_settings().casdoor
+    params = {
+        "client_id": c.client_id,
+        "response_type": "code",
+        "redirect_uri": callback,
+        "scope": scope,
+        "state": state or secrets.token_urlsafe(16),
+    }
+    if prompt:
+        params["prompt"] = prompt
+    return f"{c.endpoint.rstrip('/')}/login/oauth/authorize?{urlencode(params)}"
+
+
+@router.get("/login")
+async def login(request: Request, redirect_uri: str = Query(None)):
+    """Redirect the browser to the Casdoor authorization page.
+
+    No ``prompt=login`` — an existing Casdoor session auto-redirects back with
+    a code without showing the login form (silent SSO across *.helu.ca).
+    """
+    if not get_settings().casdoor.enabled:
+        raise HTTPException(400, "Casdoor SSO is not enabled")
+
+    callback = redirect_uri or f"{request.base_url}auth/callback"
+    return RedirectResponse(url=_build_casdoor_auth_url(callback))
+
+
+@router.get("/callback")
+async def callback(
+    code: str = Query(...),
+    state: str = Query(None),
+    redirect_uri: str = Query(None),
+):
+    """Exchange the authorization code for tokens.
+
+    Redirects to the dashboard with the access token in the URL *fragment*
+    (``/#token=...``) so the token stays client-side and is stored in
+    localStorage.
+    """
+    if not get_settings().casdoor.enabled:
+        raise HTTPException(400, "Casdoor SSO is not enabled")
+
+    sdk = get_sdk()
+    try:
+        token = await sdk.get_oauth_token(code=code)
+    except Exception as exc:
+        raise HTTPException(400, f"Token exchange failed: {exc}") from exc
+
+    access_token = token.get("access_token", "")
+    return RedirectResponse(url=f"/#token={access_token}")
+
+
+@router.get("/silent-refresh")
+async def silent_refresh(request: Request):
+    """Start a silent token refresh via hidden iframe (``prompt=none``).
+
+    If the Casdoor session is still active, Casdoor redirects back to
+    ``/auth/refresh-callback`` with a fresh code — no login form. Otherwise it
+    returns an error and the iframe tells the parent to show the login overlay.
+    """
+    if not get_settings().casdoor.enabled:
+        raise HTTPException(400, "Casdoor SSO is not enabled")
+
+    callback = f"{request.base_url}auth/refresh-callback"
+    return RedirectResponse(url=_build_casdoor_auth_url(callback, prompt="none"))
+
+
+@router.get("/refresh-callback")
+async def refresh_callback(
+    code: str = Query(None),
+    error: str = Query(None),
+    state: str = Query(None),
+):
+    """Handle the silent-refresh callback inside the hidden iframe.
+
+    On success posts the new token to the parent window; on failure posts an
+    error so the parent shows the login overlay.
+    """
+    if not get_settings().casdoor.enabled:
+        raise HTTPException(400, "Casdoor SSO is not enabled")
+
+    if error or not code:
+        return HTMLResponse(
+            '<script>window.parent.postMessage('
+            '{type:"hold-slayer-refresh",error:true},"*");</script>'
+        )
+
+    sdk = get_sdk()
+    try:
+        token = await sdk.get_oauth_token(code=code)
+        access_token = token.get("access_token", "")
+    except Exception:
+        return HTMLResponse(
+            '<script>window.parent.postMessage('
+            '{type:"hold-slayer-refresh",error:true},"*");</script>'
+        )
+
+    return HTMLResponse(
+        f'<script>window.parent.postMessage('
+        f'{{type:"hold-slayer-refresh",token:"{access_token}"}},"*");</script>'
+    )
+
+
+@router.get("/me")
+async def me(request: Request):
+    """Return the current authenticated user's profile + ``is_owner``.
+
+    Resolved manually (not via the ``OwnerUser`` gate) so a signed-in
+    non-owner gets a 200 with ``is_owner:false`` — the dashboard uses that to
+    show the "not authorized" screen rather than treating it as a hard 401.
+    """
+    auth_header = request.headers.get("authorization")
+    q_token = request.query_params.get("token")
+
+    async with session_scope() as session:
+        user = await resolve_from_header_or_query(session, auth_header, q_token)
+    if user is None:
+        raise HTTPException(
+            status_code=401,
+            detail="Not authenticated",
+            headers={"WWW-Authenticate": "Bearer"},
+        )
+    return JSONResponse(
+        {
+            "id": user.id,
+            "name": user.name,
+            "display_name": user.display_name,
+            "email": user.email,
+            "is_owner": is_owner(user),
+        }
+    )
+
+
+@router.get("/logout")
+async def logout(request: Request):
+    """Clear the Casdoor session and redirect back to the app.
+
+    ``post_logout_redirect_uri`` must be absolute — Casdoor won't follow a
+    relative ``/`` — so it's derived from ``request.base_url`` (works behind
+    HAProxy/nginx with X-Forwarded-Proto/Host).
+    """
+    c = get_settings().casdoor
+    if not c.enabled:
+        return RedirectResponse(url="/")
+
+    app_url = str(request.base_url).rstrip("/")
+    logout_url = (
+        f"{c.endpoint.rstrip('/')}/login/oauth/logout"
+        f"?client_id={c.client_id}"
+        f"&post_logout_redirect_uri={app_url}/auth/login"
+    )
+    return RedirectResponse(url=logout_url)

api/deps.py

@@ -1,12 +1,12 @@
 """
 API Dependencies — Shared dependency injection for all routes.
+
+Auth is not here: the owner gate lives in `auth.py` (`get_current_owner` /
+`OwnerUser`), applied as a router-level dependency in main.py.
 """
 
-import secrets
+from fastapi import HTTPException, Request
 
-from fastapi import Header, HTTPException, Query, Request
-
-from config import get_settings
 from core.gateway import AIPSTNGateway
 
 
@@ -24,31 +24,3 @@ def get_routing_service(request: Request):
     if routing is None:
         raise HTTPException(status_code=503, detail="Routing service not ready")
     return routing
-
-
-def require_token(
-    authorization: str | None = Header(default=None),
-    token: str | None = Query(default=None),
-) -> None:
-    """
-    Enforce the static bearer token (API_TOKEN) on REST routes.
-
-    A `token` query parameter is accepted alongside the Authorization
-    header for clients that can't set headers — <audio>/<a> elements
-    fetching recordings — matching the WebSocket convention.
-
-    An empty configured token disables auth; startup refuses that
-    combination unless the server is bound to loopback.
-    """
-    expected = get_settings().api_token.get_secret_value()
-    if not expected:
-        return
-    supplied = token or ""
-    if authorization and authorization.lower().startswith("bearer "):
-        supplied = authorization[7:]
-    if not secrets.compare_digest(supplied, expected):
-        raise HTTPException(
-            status_code=401,
-            detail="Missing or invalid bearer token",
-            headers={"WWW-Authenticate": "Bearer"},
-        )

api/tokens.py

@@ -0,0 +1,107 @@
+"""Owner-only CRUD for personal access tokens (PATs).
+
+PATs are long-lived bearer tokens for MCP/CLI clients (Claude Desktop, Cline)
+and scripted API consumers that can't refresh a short-lived Casdoor JWT. The
+plaintext is shown to the caller exactly once at creation; only its SHA-256
+hash is stored.
+"""
+
+import secrets
+import uuid
+from datetime import UTC, datetime
+
+from fastapi import APIRouter, Depends, HTTPException
+from pydantic import BaseModel, Field
+from sqlalchemy import select
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from auth import PAT_PREFIX, OwnerUser, hash_token
+from db.database import PersonalAccessToken, get_db
+
+router = APIRouter(prefix="/api/v1/tokens", tags=["tokens"])
+
+
+class TokenCreate(BaseModel):
+    name: str = Field(..., min_length=1, max_length=200)
+
+
+class TokenOut(BaseModel):
+    id: str
+    name: str
+    token_prefix: str
+    created_at: str | None = None
+    last_used_at: str | None = None
+    expires_at: str | None = None
+    revoked_at: str | None = None
+
+
+class TokenCreated(TokenOut):
+    token: str = Field(..., description="Plaintext token — shown only once. Store it now.")
+
+
+def _serialize(pat: PersonalAccessToken) -> dict:
+    return {
+        "id": pat.id,
+        "name": pat.name,
+        "token_prefix": pat.token_prefix,
+        "created_at": pat.created_at.isoformat() if pat.created_at else None,
+        "last_used_at": pat.last_used_at.isoformat() if pat.last_used_at else None,
+        "expires_at": pat.expires_at.isoformat() if pat.expires_at else None,
+        "revoked_at": pat.revoked_at.isoformat() if pat.revoked_at else None,
+    }
+
+
+@router.get("", response_model=list[TokenOut])
+async def list_tokens(
+    user: OwnerUser,
+    session: AsyncSession = Depends(get_db),
+) -> list[dict]:
+    """List the owner's personal access tokens (no plaintext)."""
+    result = await session.execute(
+        select(PersonalAccessToken)
+        .where(PersonalAccessToken.user_id == user.id)
+        .order_by(PersonalAccessToken.created_at.desc())
+    )
+    return [_serialize(pat) for pat in result.scalars().all()]
+
+
+@router.post("", response_model=TokenCreated, status_code=201)
+async def create_token(
+    payload: TokenCreate,
+    user: OwnerUser,
+    session: AsyncSession = Depends(get_db),
+) -> dict:
+    """Mint a new PAT. The plaintext is returned ONCE in the response."""
+    plaintext = PAT_PREFIX + secrets.token_urlsafe(32)
+    pat = PersonalAccessToken(
+        id=uuid.uuid4().hex,
+        user_id=user.id,
+        name=payload.name,
+        token_hash=hash_token(plaintext),
+        token_prefix=plaintext[: len(PAT_PREFIX) + 4],
+    )
+    session.add(pat)
+    await session.commit()
+    await session.refresh(pat)
+    return {**_serialize(pat), "token": plaintext}
+
+
+@router.delete("/{token_id}", status_code=204)
+async def revoke_token(
+    token_id: str,
+    user: OwnerUser,
+    session: AsyncSession = Depends(get_db),
+) -> None:
+    """Soft-revoke a PAT (sets revoked_at)."""
+    result = await session.execute(
+        select(PersonalAccessToken).where(
+            PersonalAccessToken.id == token_id,
+            PersonalAccessToken.user_id == user.id,
+        )
+    )
+    pat = result.scalar_one_or_none()
+    if pat is None:
+        raise HTTPException(status_code=404, detail="Token not found")
+    if pat.revoked_at is None:
+        pat.revoked_at = datetime.now(UTC)
+        await session.commit()

api/websocket.py

@@ -1,13 +1,11 @@
 """WebSocket API — Real-time call events and audio classification stream."""
 
-import asyncio
 import logging
-import secrets
 
 from fastapi import APIRouter, WebSocket, WebSocketDisconnect
 
-from api.deps import get_gateway
-from config import get_settings
+from auth import is_owner, resolve_from_header_or_query
+from db.database import session_scope
 from models.events import EventType, GatewayEvent
 
 logger = logging.getLogger(__name__)
@@ -17,21 +15,21 @@ router = APIRouter()
 
 async def _authorize(websocket: WebSocket) -> bool:
     """
-    Check the static bearer token before accepting the socket.
+    Require the owner before accepting the socket.
 
-    Browsers can't set headers on WebSocket connects, so a `token`
-    query parameter is accepted alongside the Authorization header.
+    Browsers can't set headers on WebSocket connects, so the Casdoor JWT (or
+    a PAT) is accepted on the `?token=` query param alongside the Authorization
+    header — the same narrow fallback the recording download uses. In dev mode
+    the owner resolves tokenlessly. A non-owner or absent credential closes the
+    socket with code 4401.
     """
-    token = get_settings().api_token.get_secret_value()
-    if not token:
+    q_token = websocket.query_params.get("token")
+    auth_header = websocket.headers.get("authorization")
+    async with session_scope() as session:
+        user = await resolve_from_header_or_query(session, auth_header, q_token)
+    if user is not None and is_owner(user):
         return True
-    supplied = websocket.query_params.get("token", "")
-    auth = websocket.headers.get("authorization", "")
-    if auth.lower().startswith("bearer "):
-        supplied = auth[7:]
-    if secrets.compare_digest(supplied, token):
-        return True
-    await websocket.close(code=4401, reason="Missing or invalid bearer token")
+    await websocket.close(code=4401, reason="Owner authentication required")
     return False
 
 

auth.py

@@ -0,0 +1,385 @@
+"""
+Authentication and authorisation for Hold Slayer.
+
+This gateway is **owner-only**. It dials real phones and spends money, so
+there are no guest/shared resources: exactly one operator (the Casdoor user
+whose name matches ``OWNER_NAME``) may use any surface; every other identity
+gets 403.
+
+Two bearer-token kinds are accepted on ``Authorization: Bearer <token>``
+(or, for the two browser consumers that can't set headers — the WebSocket
+connect and ``<audio>`` recording downloads — on a ``?token=`` query param):
+
+1. **Casdoor JWT** — short-lived, signed by Casdoor. Validated against the
+   public keys served at ``${CASDOOR_ENDPOINT}/.well-known/jwks`` (PyJWKClient
+   cache, RS256). Used by the browser dashboard after OIDC login.
+2. **Personal Access Token** — long-lived ``hs_pat_<random>`` token, minted
+   from the owner-only dashboard and stored hashed in
+   ``personal_access_tokens``. Used by MCP/CLI clients (Claude Desktop, Cline)
+   that can't refresh a JWT.
+
+When ``CASDOOR_ENABLED=false`` (dev, loopback only) every request resolves to
+the dev owner — no token required.
+"""
+
+from __future__ import annotations
+
+import hashlib
+import logging
+import uuid
+from datetime import UTC, datetime
+from typing import Annotated
+
+import jwt
+from casdoor import AsyncCasdoorSDK
+from fastapi import Depends, HTTPException, Query
+from fastapi.security import HTTPAuthorizationCredentials, HTTPBearer
+from sqlalchemy import select
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from config import get_settings
+from db.database import PersonalAccessToken, User, get_db
+
+logger = logging.getLogger(__name__)
+
+# ── Constants ────────────────────────────────────────────────────────────────
+
+_DEV_OWNER_SUB = "dev-owner"
+PAT_PREFIX = "hs_pat_"
+
+# ── Casdoor SDK singleton (for OAuth code exchange in /auth/callback) ─────────
+
+_sdk: AsyncCasdoorSDK | None = None
+
+
+def get_sdk() -> AsyncCasdoorSDK:
+    """Build the Casdoor SDK lazily.
+
+    Used only for the OAuth2 code-exchange step in ``/auth/callback`` — JWT
+    validation happens via PyJWKClient below. The certificate parameter is
+    unused for code exchange but the constructor requires *something*; we pass
+    an empty bytestring.
+    """
+    global _sdk
+    if _sdk is None:
+        c = get_settings().casdoor
+        _sdk = AsyncCasdoorSDK(
+            endpoint=c.endpoint,
+            client_id=c.client_id,
+            client_secret=c.client_secret.get_secret_value(),
+            certificate=b"",
+            org_name=c.org_name,
+            application_name=c.app_name,
+        )
+    return _sdk
+
+
+# ── JWKS client (for Casdoor JWT validation) ─────────────────────────────────
+
+_jwks_client: jwt.PyJWKClient | None = None
+
+
+def init_jwks_client() -> None:
+    """Construct the PyJWKClient pointed at Casdoor's JWKS endpoint.
+
+    Called once from the app lifespan before requests are served. Pre-fetches
+    the keys so the network round-trip happens at startup rather than on the
+    first authenticated request. A no-op when SSO is disabled; a failed
+    prefetch is non-fatal (keys are fetched lazily on first use).
+    """
+    global _jwks_client
+    if not get_settings().casdoor.enabled:
+        return
+    endpoint = get_settings().casdoor.endpoint.rstrip("/")
+    jwks_uri = f"{endpoint}/.well-known/jwks"
+    _jwks_client = jwt.PyJWKClient(jwks_uri, cache_keys=True, lifespan=3600)
+    try:
+        _jwks_client.fetch_data()
+        logger.info("Casdoor JWKS prefetched from %s", jwks_uri)
+    except Exception as exc:
+        logger.warning("Casdoor JWKS prefetch failed (%s); will retry on first request", exc)
+
+
+def _decode_casdoor_jwt(token: str) -> dict:
+    """Validate a Casdoor RS256 JWT against the cached JWKS.
+
+    Refreshes the key cache once on unknown-kid before giving up. Audience
+    verification is disabled because Casdoor sets ``aud`` to the application
+    name, which differs from the client_id; the signature check against
+    Casdoor's key is the primary control.
+    """
+    if _jwks_client is None:
+        raise HTTPException(status_code=503, detail="Auth subsystem not ready")
+
+    issuer = get_settings().casdoor.endpoint.rstrip("/")
+
+    def _decode_with_current_keys() -> dict:
+        signing_key = _jwks_client.get_signing_key_from_jwt(token)
+        return jwt.decode(
+            token,
+            signing_key.key,
+            algorithms=["RS256"],
+            issuer=issuer,
+            options={"verify_aud": False},
+        )
+
+    try:
+        return _decode_with_current_keys()
+    except jwt.ExpiredSignatureError as exc:
+        raise HTTPException(status_code=401, detail="Token has expired") from exc
+    except jwt.PyJWKClientError as exc:
+        logger.warning("Unknown JWKS key (%s); refreshing", exc)
+        try:
+            _jwks_client.fetch_data()
+            return _decode_with_current_keys()
+        except Exception as inner:
+            raise HTTPException(status_code=401, detail=f"Invalid token: {inner}") from inner
+    except jwt.InvalidTokenError as exc:
+        raise HTTPException(status_code=401, detail=f"Invalid token: {exc}") from exc
+
+
+# ── PAT helpers ──────────────────────────────────────────────────────────────
+
+
+def hash_token(plaintext: str) -> str:
+    """SHA-256 hex digest of a plaintext PAT."""
+    return hashlib.sha256(plaintext.encode("utf-8")).hexdigest()
+
+
+async def _validate_pat(session: AsyncSession, plaintext: str) -> User:
+    """Look up a PAT by hash, check it's active, return the owning user."""
+    digest = hash_token(plaintext)
+    result = await session.execute(
+        select(PersonalAccessToken).where(PersonalAccessToken.token_hash == digest)
+    )
+    pat = result.scalar_one_or_none()
+    if pat is None or pat.revoked_at is not None:
+        raise HTTPException(status_code=401, detail="Invalid token")
+
+    now = datetime.now(UTC)
+    if pat.expires_at is not None:
+        # DateTime columns come back naive on SQLite (and on a Postgres
+        # TIMESTAMP without tz); treat a naive value as UTC before comparing.
+        expires_at = pat.expires_at
+        if expires_at.tzinfo is None:
+            expires_at = expires_at.replace(tzinfo=UTC)
+        if expires_at <= now:
+            raise HTTPException(status_code=401, detail="Token has expired")
+
+    pat.last_used_at = now
+    try:
+        await session.commit()
+    except Exception:
+        await session.rollback()
+
+    user_result = await session.execute(select(User).where(User.id == pat.user_id))
+    user = user_result.scalar_one_or_none()
+    if user is None:
+        raise HTTPException(status_code=401, detail="Invalid token")
+    return user
+
+
+# ── User provisioning ────────────────────────────────────────────────────────
+
+
+async def _get_or_create_dev_owner(session: AsyncSession) -> User:
+    """Return the dev-mode owner user row, creating it if it doesn't exist."""
+    result = await session.execute(select(User).where(User.casdoor_sub == _DEV_OWNER_SUB))
+    user = result.scalar_one_or_none()
+    if user is None:
+        user = User(id=uuid.uuid4().hex, name="Owner", casdoor_sub=_DEV_OWNER_SUB)
+        session.add(user)
+        await session.commit()
+        await session.refresh(user)
+    return user
+
+
+async def _find_or_create_user(
+    session: AsyncSession,
+    casdoor_sub: str,
+    name: str,
+    display_name: str,
+    email: str | None,
+) -> User:
+    """Look up a user by casdoor_sub; create a new row on first login.
+
+    Lookup priority, so identity survives a Casdoor redeploy:
+    1. casdoor_sub — the OIDC subject claim (primary SSO identity).
+    2. name — the Casdoor username (stable, unique). Relinks a changed sub.
+    3. email — pre-SSO users logging in via Casdoor for the first time.
+
+    Non-owner users are still provisioned (so ``is_owner`` can say "no"), but
+    they reach nothing — every surface is owner-gated.
+    """
+    result = await session.execute(select(User).where(User.casdoor_sub == casdoor_sub))
+    user = result.scalar_one_or_none()
+    if user is not None:
+        changed = False
+        if user.name != name:
+            user.name = name
+            changed = True
+        if user.display_name != display_name:
+            user.display_name = display_name
+            changed = True
+        if changed:
+            await session.commit()
+            await session.refresh(user)
+        return user
+
+    result = await session.execute(select(User).where(User.name == name))
+    user = result.scalar_one_or_none()
+    if user is not None:
+        logger.info(
+            "Linking user %s (id=%s) to new casdoor_sub %s (was %s)",
+            name, user.id, casdoor_sub, user.casdoor_sub,
+        )
+        user.casdoor_sub = casdoor_sub
+        user.display_name = display_name
+        await session.commit()
+        await session.refresh(user)
+        return user
+
+    if email:
+        result = await session.execute(select(User).where(User.email == email))
+        user = result.scalar_one_or_none()
+        if user is not None:
+            user.casdoor_sub = casdoor_sub
+            user.name = name
+            user.display_name = display_name
+            await session.commit()
+            await session.refresh(user)
+            return user
+
+    user = User(
+        id=uuid.uuid4().hex,
+        name=name,
+        display_name=display_name,
+        email=email,
+        casdoor_sub=casdoor_sub,
+    )
+    session.add(user)
+    await session.commit()
+    await session.refresh(user)
+    logger.info("Created new user: %s (id=%s)", name, user.id)
+    return user
+
+
+def _claims_to_identity(claims: dict) -> tuple[str, str, str, str | None]:
+    """Pull (sub, name, display_name, email) out of Casdoor JWT claims."""
+    sub = claims.get("sub") or claims.get("name") or ""
+    name = claims.get("name") or sub
+    display_name = claims.get("displayName") or claims.get("name") or sub
+    email = claims.get("email") or None
+    return sub, name, display_name, email
+
+
+# ── The single resolver — used by every surface ──────────────────────────────
+
+
+async def resolve_bearer(session: AsyncSession, raw_token: str | None) -> User | None:
+    """Resolve a bare bearer-token string to a User, or None on any failure.
+
+    This is the one place a token becomes an identity. It never raises — the
+    caller decides how to respond (401/403 for REST, 4401 close for WS). The
+    header-based REST path and the ``?token=`` query path both funnel here.
+
+    Dev mode (SSO disabled) ignores the token and returns the dev owner.
+    """
+    if not get_settings().casdoor.enabled:
+        return await _get_or_create_dev_owner(session)
+
+    if not raw_token:
+        return None
+
+    try:
+        if raw_token.startswith(PAT_PREFIX):
+            return await _validate_pat(session, raw_token)
+        claims = _decode_casdoor_jwt(raw_token)
+    except HTTPException:
+        return None
+
+    sub, name, display_name, email = _claims_to_identity(claims)
+    if not sub:
+        return None
+    try:
+        return await _find_or_create_user(session, sub, name, display_name, email)
+    except Exception:
+        return None
+
+
+def _token_from_header(authorization_header: str | None) -> str | None:
+    """Extract the bearer token from an Authorization header, or None."""
+    if not authorization_header:
+        return None
+    parts = authorization_header.split(None, 1)
+    if len(parts) != 2 or parts[0].lower() != "bearer":
+        return None
+    token = parts[1].strip()
+    return token or None
+
+
+async def resolve_from_header_or_query(
+    session: AsyncSession,
+    authorization_header: str | None,
+    query_token: str | None,
+) -> User | None:
+    """Resolve a User from an Authorization header, falling back to ``?token=``.
+
+    The query fallback exists only for the two browser consumers that can't
+    set headers — the WebSocket connect and ``<audio>`` recording downloads —
+    matching Hold Slayer's long-standing narrow ``?token=`` convention. The
+    header wins when both are present.
+    """
+    raw = _token_from_header(authorization_header) or query_token
+    return await resolve_bearer(session, raw)
+
+
+# ── Ownership ────────────────────────────────────────────────────────────────
+
+
+def is_owner(user: User) -> bool:
+    """Whether the user owns this gateway.
+
+    Dev mode: the dev-owner sub is always the owner. SSO mode: the owner is
+    the user whose Casdoor username (``user.name``) matches ``OWNER_NAME``.
+    """
+    if not get_settings().casdoor.enabled:
+        return user.casdoor_sub == _DEV_OWNER_SUB
+    owner_name = get_settings().owner_name
+    return bool(owner_name and user.name == owner_name)
+
+
+# ── FastAPI dependencies ─────────────────────────────────────────────────────
+
+_bearer = HTTPBearer(auto_error=False)
+
+
+async def get_current_user(
+    credentials: Annotated[HTTPAuthorizationCredentials | None, Depends(_bearer)] = None,
+    token: str | None = Query(default=None),
+    session: AsyncSession = Depends(get_db),
+) -> User:
+    """Resolve the authenticated user (owner or not) — 401 if unauthenticated.
+
+    Used by ``/auth/me`` so a signed-in non-owner sees ``is_owner:false``
+    rather than a bare 401. Owner-gating is a separate step.
+    """
+    header = f"Bearer {credentials.credentials}" if credentials else None
+    user = await resolve_from_header_or_query(session, header, token)
+    if user is None:
+        raise HTTPException(
+            status_code=401,
+            detail="Not authenticated",
+            headers={"WWW-Authenticate": "Bearer"},
+        )
+    return user
+
+
+async def get_current_owner(user: Annotated[User, Depends(get_current_user)]) -> User:
+    """The one gate for protected surfaces — 401 if unauthenticated, 403 if not owner."""
+    if not is_owner(user):
+        raise HTTPException(status_code=403, detail="Owner access required")
+    return user
+
+
+OwnerUser = Annotated[User, Depends(get_current_owner)]

config.py

@@ -89,6 +89,25 @@ class TTSSettings(BaseSettings):
     sample_rate: int = 16000
 
 
+class CasdoorSettings(BaseSettings):
+    """Casdoor SSO (OIDC) configuration.
+
+    When `enabled` is true the browser authenticates via Casdoor and every
+    surface is gated to the owner; the SDK is used only for the OAuth2 code
+    exchange in the /auth/callback route (JWTs are validated against the
+    endpoint's JWKS). When false, the app runs in dev-owner mode (loopback only).
+    """
+
+    model_config = SettingsConfigDict(env_prefix="CASDOOR_", env_file=".env", extra="ignore")
+
+    enabled: bool = False
+    endpoint: str = "https://id.ouranos.helu.ca"
+    client_id: str = ""
+    client_secret: SecretStr = SecretStr("")
+    org_name: str = "heluca"
+    app_name: str = ""
+
+
 class ReceptionistSettings(BaseSettings):
     """AI Receptionist behavior settings."""
 
@@ -126,9 +145,14 @@ class Settings(BaseSettings):
     debug: bool = False
     log_level: str = "info"
 
-    # Auth — one static bearer token shared by REST, WebSocket, and MCP.
-    # Empty disables auth, which is only permitted on loopback binds.
-    api_token: SecretStr = SecretStr("")
+    # Auth — Casdoor SSO for the browser + owner-minted PATs for MCP/CLI,
+    # gated to a single owner. `owner_name` is the Casdoor username that owns
+    # this gateway (everyone else gets 403). `public_base_url` seeds the OAuth
+    # discovery URLs; blank derives them from request headers. Both cross-cut
+    # every surface, so they live on the root model (like DATABASE_URL); the
+    # Casdoor connection knobs live under the CASDOOR_ prefix.
+    owner_name: str = ""
+    public_base_url: str = ""
 
     # Outbound-call safety cap (REST + MCP make_call)
     max_concurrent_calls: int = 4
@@ -150,6 +174,7 @@ class Settings(BaseSettings):
     hold_slayer: HoldSlayerSettings = Field(default_factory=HoldSlayerSettings)
     tts: TTSSettings = Field(default_factory=TTSSettings)
     receptionist: ReceptionistSettings = Field(default_factory=ReceptionistSettings)
+    casdoor: CasdoorSettings = Field(default_factory=CasdoorSettings)
 
 
 # Singleton

dashboard/package-lock.json

@@ -13,6 +13,7 @@
         "@sveltejs/vite-plugin-svelte": "^5.0.3",
         "@tailwindcss/vite": "^4.1.3",
         "@types/node": "^25.8.0",
+        "daisyui": "^5.0.0",
         "svelte": "^5.25.3",
         "svelte-check": "^4.1.4",
         "tailwindcss": "^4.1.3",
@@ -1262,6 +1263,16 @@
         "node": ">= 0.6"
       }
     },
+    "node_modules/daisyui": {
+      "version": "5.7.0",
+      "resolved": "https://registry.npmjs.org/daisyui/-/daisyui-5.7.0.tgz",
+      "integrity": "sha512-2/kYbxaKtv349lPrTyxMKC9SHsyA7fBULMSabJljDE82D079cjqz+UyAzsogWgy4sTs5NDvD000acfcFqbO1XA==",
+      "dev": true,
+      "license": "MIT",
+      "funding": {
+        "url": "https://github.com/saadeghi/daisyui?sponsor=1"
+      }
+    },
     "node_modules/debug": {
       "version": "4.4.3",
       "resolved": "https://registry.npmjs.org/debug/-/debug-4.4.3.tgz",

dashboard/package.json

@@ -15,6 +15,7 @@
     "@sveltejs/vite-plugin-svelte": "^5.0.3",
     "@tailwindcss/vite": "^4.1.3",
     "@types/node": "^25.8.0",
+    "daisyui": "^5.0.0",
     "svelte": "^5.25.3",
     "svelte-check": "^4.1.4",
     "tailwindcss": "^4.1.3",

dashboard/src/app.css

@@ -1 +1,4 @@
 @import 'tailwindcss';
+@plugin 'daisyui' {
+	themes: light --default, dark --prefersdark;
+}

dashboard/src/hooks.client.ts

@@ -0,0 +1,5 @@
+import { auth } from '$lib/auth.svelte';
+
+// Runs once, before the app mounts: capture the Casdoor callback token from
+// the URL fragment before any /auth/me call or render.
+auth.captureFragmentToken();

dashboard/src/lib/api.ts

@@ -1,4 +1,5 @@
 import type {
+	AccessToken,
 	CallHistoryRow,
 	CallSummary,
 	DeviceStatus,
@@ -10,19 +11,12 @@ import type {
 } from './types';
 
 // ---------------------------------------------------------------
-// Bearer token — one static API_TOKEN shared with REST/WS/MCP.
-// Kept in localStorage; a 401 prompts once and retries.
+// Authed fetch — the browser holds a Casdoor JWT (or, for scripted use,
+// a PAT) in localStorage via the auth store. On a 401 we attempt one
+// silent refresh and retry; a second failure logs out.
 // ---------------------------------------------------------------
 
-const TOKEN_KEY = 'hold-slayer-token';
-
-function getToken(): string {
-	return localStorage.getItem(TOKEN_KEY) ?? '';
-}
-
-export function setToken(value: string): void {
-	localStorage.setItem(TOKEN_KEY, value);
-}
+import { auth, getToken } from './auth.svelte';
 
 function withAuth(init: RequestInit): RequestInit {
 	const token = getToken();
@@ -36,10 +30,11 @@ function withAuth(init: RequestInit): RequestInit {
 async function request(path: string, init: RequestInit = {}): Promise<Response> {
 	let res = await fetch(path, withAuth(init));
 	if (res.status === 401) {
-		const supplied = window.prompt('Hold Slayer API token (API_TOKEN in .env):');
-		if (supplied !== null && supplied.trim()) {
-			setToken(supplied.trim());
+		const refreshed = await auth.trySilentRefresh();
+		if (refreshed) {
 			res = await fetch(path, withAuth(init));
+		} else {
+			auth.setUnauthenticated();
 		}
 	}
 	return res;
@@ -89,8 +84,10 @@ export async function fetchTranscript(callId: string): Promise<TranscriptRow[]>
 }
 
 export function recordingUrl(callId: string): string {
-	// <audio> can't send headers, so the token rides as a query param
-	// (accepted server-side alongside the Authorization header).
+	// <audio> can't send headers, so the current token (Casdoor JWT, or a PAT)
+	// rides as a query param — the same narrow fallback the WebSocket uses,
+	// accepted server-side alongside the Authorization header. The proactive
+	// refresh timer keeps the stored JWT valid, so it's fresh at click time.
 	const token = getToken();
 	const suffix = token ? `?token=${encodeURIComponent(token)}` : '';
 	return `/api/v1/calls/${callId}/recording${suffix}`;
@@ -137,11 +134,35 @@ export async function setDeviceDnd(deviceId: string, enabled: boolean): Promise<
 	if (!res.ok) throw new Error(`${res.status} ${res.statusText}`);
 }
 
+// ---------------------------------------------------------------
+// Personal Access Tokens (owner-only) — for MCP/CLI clients.
+// ---------------------------------------------------------------
+
+export async function fetchTokens(): Promise<AccessToken[]> {
+	return get<AccessToken[]>('/api/v1/tokens');
+}
+
+export async function createToken(name: string): Promise<AccessToken> {
+	const res = await request('/api/v1/tokens', {
+		method: 'POST',
+		headers: { 'Content-Type': 'application/json' },
+		body: JSON.stringify({ name }),
+	});
+	if (!res.ok) throw new Error(`${res.status} ${res.statusText}`);
+	return res.json() as Promise<AccessToken>;
+}
+
+export async function revokeToken(tokenId: string): Promise<void> {
+	const res = await request(`/api/v1/tokens/${tokenId}`, { method: 'DELETE' });
+	if (!res.ok) throw new Error(`${res.status} ${res.statusText}`);
+}
+
 export function connectEventStream(
 	onEvent: (e: GatewayEvent) => void,
 	onClose: () => void,
 ): () => void {
 	const proto = location.protocol === 'https:' ? 'wss:' : 'ws:';
+	// Browsers can't set headers on WS connects — the token rides as ?token=.
 	const token = getToken();
 	const suffix = token ? `?token=${encodeURIComponent(token)}` : '';
 	const ws = new WebSocket(`${proto}//${location.host}/ws/events${suffix}`);

dashboard/src/lib/auth.svelte.ts

@@ -0,0 +1,146 @@
+import type { User } from './types';
+
+const TOKEN_KEY = 'hold-slayer-token';
+
+// 'denied' = authenticated with Casdoor but not the owner of this gateway.
+export type AuthStatus = 'loading' | 'authed' | 'unauthenticated' | 'denied';
+
+export function getToken(): string {
+	return localStorage.getItem(TOKEN_KEY) || '';
+}
+
+function setToken(token: string) {
+	localStorage.setItem(TOKEN_KEY, token);
+}
+
+export function clearToken() {
+	localStorage.removeItem(TOKEN_KEY);
+}
+
+class AuthStore {
+	status = $state<AuthStatus>('loading');
+	user = $state<User | null>(null);
+
+	private refreshTimer: ReturnType<typeof setTimeout> | null = null;
+
+	get isOwner(): boolean {
+		return this.user?.is_owner ?? false;
+	}
+
+	/**
+	 * Capture a `#token=...` fragment left by the Casdoor callback, persist it,
+	 * and scrub it from the URL. Runs before the first /auth/me call.
+	 */
+	captureFragmentToken() {
+		const hash = window.location.hash;
+		if (hash.startsWith('#token=')) {
+			const token = hash.slice(7);
+			if (token) setToken(token);
+			history.replaceState(null, '', window.location.pathname + window.location.search);
+		}
+	}
+
+	/** Determine auth state on boot. */
+	async init(): Promise<void> {
+		const token = getToken();
+		if (!token) {
+			// No token: maybe SSO is disabled (dev mode) — /auth/me succeeds tokenless.
+			try {
+				const res = await fetch('/auth/me');
+				if (res.ok) {
+					this.applyUser(await res.json());
+					return;
+				}
+			} catch {
+				/* fall through to unauthenticated */
+			}
+			this.status = 'unauthenticated';
+			return;
+		}
+
+		try {
+			const res = await fetch('/auth/me', {
+				headers: { Authorization: `Bearer ${token}` }
+			});
+			if (!res.ok) {
+				// Token invalid — try silent refresh once, then retry.
+				const refreshed = await this.trySilentRefresh();
+				if (refreshed) return this.init();
+				clearToken();
+				this.status = 'unauthenticated';
+				return;
+			}
+			this.applyUser(await res.json());
+			this.scheduleTokenRefresh(token);
+		} catch {
+			this.status = 'unauthenticated';
+		}
+	}
+
+	/** Set user + status from an /auth/me payload. Non-owners are denied. */
+	private applyUser(user: User) {
+		this.user = user;
+		this.status = user.is_owner ? 'authed' : 'denied';
+	}
+
+	setUnauthenticated() {
+		clearToken();
+		this.user = null;
+		this.status = 'unauthenticated';
+	}
+
+	/** Silent token refresh via a hidden iframe + postMessage. */
+	trySilentRefresh(): Promise<boolean> {
+		return new Promise((resolve) => {
+			const iframe = document.createElement('iframe');
+			iframe.style.display = 'none';
+			iframe.src = '/auth/silent-refresh';
+			let resolved = false;
+
+			const cleanup = () => {
+				if (resolved) return;
+				resolved = true;
+				window.removeEventListener('message', onMessage);
+				iframe.remove();
+			};
+
+			const onMessage = (event: MessageEvent) => {
+				if (!event.data || event.data.type !== 'hold-slayer-refresh') return;
+				cleanup();
+				if (event.data.token) {
+					setToken(event.data.token);
+					this.scheduleTokenRefresh(event.data.token);
+					resolve(true);
+				} else {
+					resolve(false);
+				}
+			};
+
+			window.addEventListener('message', onMessage);
+			document.body.appendChild(iframe);
+
+			setTimeout(() => {
+				cleanup();
+				resolve(false);
+			}, 10000);
+		});
+	}
+
+	/** Proactively refresh 5 minutes before JWT expiry. PATs are skipped. */
+	scheduleTokenRefresh(token: string) {
+		if (this.refreshTimer) clearTimeout(this.refreshTimer);
+		try {
+			const payload = JSON.parse(atob(token.split('.')[1]));
+			const exp = payload.exp * 1000;
+			const refreshIn = Math.max(exp - Date.now() - 5 * 60 * 1000, 30 * 1000);
+			this.refreshTimer = setTimeout(async () => {
+				const ok = await this.trySilentRefresh();
+				if (!ok) this.setUnauthenticated();
+			}, refreshIn);
+		} catch {
+			// Not a JWT (e.g. a PAT) — no refresh needed.
+		}
+	}
+}
+
+export const auth = new AuthStore();

dashboard/src/lib/components/DeniedScreen.svelte

@@ -0,0 +1,17 @@
+<script lang="ts">
+	import { auth } from '$lib/auth.svelte';
+</script>
+
+<div class="bg-base-100 fixed inset-0 z-[9999] flex items-center justify-center p-4">
+	<div class="card bg-base-200 w-96 shadow-xl">
+		<div class="card-body items-center gap-4 text-center">
+			<h1 class="text-2xl font-bold">Not authorized</h1>
+			<p class="text-sm opacity-70">
+				Hold Slayer is a single-operator gateway. You're signed in as
+				<span class="font-medium">{auth.user?.display_name ?? auth.user?.name}</span>,
+				but this gateway is reserved for its owner.
+			</p>
+			<a href="/auth/logout" class="btn btn-outline w-full">Sign out</a>
+		</div>
+	</div>
+</div>

dashboard/src/lib/components/LoginScreen.svelte

@@ -0,0 +1,17 @@
+<script lang="ts">
+	// A full-screen sign-in gate. The anchor is a real navigation to the
+	// server-side /auth/login route (which 302s to Casdoor), not a fetch.
+</script>
+
+<div class="bg-base-100 fixed inset-0 z-[9999] flex items-center justify-center p-4">
+	<div class="card bg-base-200 w-80 shadow-xl">
+		<div class="card-body items-center gap-4 text-center">
+			<h1 class="flex items-center justify-center gap-2 text-3xl font-bold">
+				<span class="text-orange-500">🔥</span>
+				Hold Slayer
+			</h1>
+			<p class="text-sm opacity-60">Sign in to the gateway</p>
+			<a href="/auth/login" class="btn btn-primary w-full">Sign in with SSO</a>
+		</div>
+	</div>
+</div>

dashboard/src/lib/components/TokensModal.svelte

@@ -0,0 +1,162 @@
+<script lang="ts">
+	import type { AccessToken } from '$lib/types';
+	import { createToken, fetchTokens, revokeToken } from '$lib/api';
+
+	let { open = $bindable(false) }: { open?: boolean } = $props();
+
+	let tokens = $state<AccessToken[]>([]);
+	let loading = $state(false);
+	let error = $state<string | null>(null);
+	let newName = $state('');
+	let creating = $state(false);
+	// The plaintext of a just-created token — shown once, never re-fetchable.
+	let created = $state<AccessToken | null>(null);
+
+	async function load() {
+		loading = true;
+		error = null;
+		try {
+			tokens = await fetchTokens();
+		} catch (e) {
+			error = e instanceof Error ? e.message : String(e);
+		} finally {
+			loading = false;
+		}
+	}
+
+	$effect(() => {
+		if (open) {
+			created = null;
+			void load();
+		}
+	});
+
+	async function create() {
+		if (!newName.trim()) return;
+		creating = true;
+		error = null;
+		try {
+			created = await createToken(newName.trim());
+			newName = '';
+			await load();
+		} catch (e) {
+			error = e instanceof Error ? e.message : String(e);
+		} finally {
+			creating = false;
+		}
+	}
+
+	async function revoke(id: string) {
+		error = null;
+		try {
+			await revokeToken(id);
+			await load();
+		} catch (e) {
+			error = e instanceof Error ? e.message : String(e);
+		}
+	}
+
+	function mcpConfig(plaintext: string): string {
+		const url = `${location.origin}/mcp`;
+		return JSON.stringify(
+			{
+				mcpServers: {
+					'hold-slayer': {
+						type: 'streamable-http',
+						url,
+						headers: { Authorization: `Bearer ${plaintext}` }
+					}
+				}
+			},
+			null,
+			2
+		);
+	}
+
+	function copy(text: string) {
+		void navigator.clipboard.writeText(text);
+	}
+</script>
+
+{#if open}
+	<div class="modal modal-open">
+		<div class="modal-box max-w-2xl">
+			<h3 class="text-lg font-bold">API Tokens</h3>
+			<p class="py-1 text-sm opacity-60">
+				Personal access tokens for MCP/CLI clients (Claude Desktop, Cline). The
+				plaintext is shown once — store it now.
+			</p>
+
+			{#if error}
+				<div class="alert alert-error my-2 text-sm">{error}</div>
+			{/if}
+
+			{#if created?.token}
+				<div class="alert alert-success my-3 flex-col items-start gap-2">
+					<span class="font-medium">Token created — copy it now, it won't be shown again.</span>
+					<code class="bg-base-300 w-full break-all rounded p-2 text-xs">{created.token}</code>
+					<div class="flex gap-2">
+						<button class="btn btn-xs" onclick={() => copy(created!.token!)}>Copy token</button>
+						<button class="btn btn-xs" onclick={() => copy(mcpConfig(created!.token!))}
+							>Copy MCP config</button
+						>
+					</div>
+				</div>
+			{/if}
+
+			<div class="my-3 flex gap-2">
+				<input
+					class="input input-bordered flex-1"
+					placeholder="Token name (e.g. Claude Desktop)"
+					bind:value={newName}
+					onkeydown={(e) => e.key === 'Enter' && create()}
+				/>
+				<button class="btn btn-primary" disabled={creating || !newName.trim()} onclick={create}>
+					{creating ? 'Creating…' : 'Create'}
+				</button>
+			</div>
+
+			{#if loading}
+				<div class="py-4 text-center opacity-60">Loading…</div>
+			{:else if tokens.length === 0}
+				<div class="py-4 text-center opacity-60">No tokens yet.</div>
+			{:else}
+				<div class="overflow-x-auto">
+					<table class="table table-sm">
+						<thead>
+							<tr>
+								<th>Name</th>
+								<th>Prefix</th>
+								<th>Last used</th>
+								<th></th>
+							</tr>
+						</thead>
+						<tbody>
+							{#each tokens as t (t.id)}
+								<tr class:opacity-50={t.revoked_at}>
+									<td>{t.name}</td>
+									<td><code class="text-xs">{t.token_prefix}…</code></td>
+									<td class="text-xs">{t.last_used_at?.slice(0, 10) ?? '—'}</td>
+									<td class="text-right">
+										{#if t.revoked_at}
+											<span class="badge badge-ghost badge-sm">revoked</span>
+										{:else}
+											<button class="btn btn-ghost btn-xs text-error" onclick={() => revoke(t.id)}>
+												Revoke
+											</button>
+										{/if}
+									</td>
+								</tr>
+							{/each}
+						</tbody>
+					</table>
+				</div>
+			{/if}
+
+			<div class="modal-action">
+				<button class="btn" onclick={() => (open = false)}>Close</button>
+			</div>
+		</div>
+		<button class="modal-backdrop" onclick={() => (open = false)} aria-label="Close"></button>
+	</div>
+{/if}

dashboard/src/lib/types.ts

@@ -1,3 +1,23 @@
+export interface User {
+	id: string;
+	name: string;
+	display_name: string | null;
+	email: string | null;
+	is_owner: boolean;
+}
+
+export interface AccessToken {
+	id: string;
+	name: string;
+	token_prefix: string;
+	created_at: string | null;
+	last_used_at: string | null;
+	expires_at: string | null;
+	revoked_at: string | null;
+	// Present only in the create response — the plaintext, shown once.
+	token?: string;
+}
+
 export interface GatewayStatus {
 	name: string;
 	version: string;

dashboard/src/routes/+layout.svelte

@@ -2,9 +2,15 @@
 	import '../app.css';
 	import { page } from '$app/stores';
 	import { onMount } from 'svelte';
+	import { auth } from '$lib/auth.svelte';
+	import LoginScreen from '$lib/components/LoginScreen.svelte';
+	import DeniedScreen from '$lib/components/DeniedScreen.svelte';
+	import TokensModal from '$lib/components/TokensModal.svelte';
 
 	let { children } = $props();
 
+	let tokensOpen = $state(false);
+
 	type ThemeOverride = 'dark' | 'light' | null;
 	let override = $state<ThemeOverride>(null);
 	let systemDark = $state(true);
@@ -12,7 +18,10 @@
 	let isDark = $derived(override !== null ? override === 'dark' : systemDark);
 
 	$effect(() => {
+		// Keep both theming systems in sync: Tailwind `dark:` variant (.dark class)
+		// for the existing pages, and DaisyUI `data-theme` for the SSO components.
 		document.documentElement.classList.toggle('dark', isDark);
+		document.documentElement.setAttribute('data-theme', isDark ? 'dark' : 'light');
 	});
 
 	function toggleTheme() {
@@ -27,6 +36,8 @@
 	}
 
 	onMount(() => {
+		void auth.init();
+
 		const mq = window.matchMedia('(prefers-color-scheme: dark)');
 		systemDark = mq.matches;
 
@@ -49,7 +60,16 @@
 	];
 </script>
 
-<div class="min-h-screen bg-slate-50 dark:bg-gray-950 text-gray-900 dark:text-gray-100">
+{#if auth.status === 'loading'}
+	<div class="fixed inset-0 flex items-center justify-center bg-slate-50 dark:bg-gray-950">
+		<span class="loading loading-spinner loading-lg text-orange-500"></span>
+	</div>
+{:else if auth.status === 'unauthenticated'}
+	<LoginScreen />
+{:else if auth.status === 'denied'}
+	<DeniedScreen />
+{:else}
+	<div class="min-h-screen bg-slate-50 dark:bg-gray-950 text-gray-900 dark:text-gray-100">
 		<header
 			class="border-b border-gray-200 dark:border-gray-800 bg-white/80 dark:bg-gray-900/80 backdrop-blur sticky top-0 z-10"
 		>
@@ -72,17 +92,31 @@
 						</a>
 					{/each}
 				</nav>
+				<div class="ml-auto flex items-center gap-2">
 					<button
 						onclick={toggleTheme}
-				class="ml-auto text-xs px-2.5 py-1 rounded-full bg-gray-100 dark:bg-gray-800 text-gray-600 dark:text-gray-400 border border-gray-200 dark:border-gray-700 hover:bg-gray-200 dark:hover:bg-gray-700 transition-colors"
+						class="text-xs px-2.5 py-1 rounded-full bg-gray-100 dark:bg-gray-800 text-gray-600 dark:text-gray-400 border border-gray-200 dark:border-gray-700 hover:bg-gray-200 dark:hover:bg-gray-700 transition-colors"
 						title={isDark ? 'Switch to light mode' : 'Switch to dark mode'}
 					>
 						{isDark ? 'Light' : 'Dark'}
 					</button>
+					<div class="dropdown dropdown-end">
+						<button tabindex="0" class="btn btn-ghost btn-sm">
+							{auth.user?.display_name ?? auth.user?.name ?? 'Owner'}
+						</button>
+						<ul class="dropdown-content menu bg-base-200 rounded-box z-20 w-48 p-2 shadow">
+							<li><button onclick={() => (tokensOpen = true)}>API Tokens</button></li>
+							<li><a href="/auth/logout">Sign out</a></li>
+						</ul>
+					</div>
+				</div>
 			</div>
 		</header>
 
 		<main class="mx-auto max-w-7xl px-4 py-6">
 			{@render children()}
 		</main>
-</div>
+	</div>
+
+	<TokensModal bind:open={tokensOpen} />
+{/if}

db/database.py

@@ -14,6 +14,7 @@ from sqlalchemy import (
     Column,
     DateTime,
     Float,
+    ForeignKey,
     Integer,
     String,
     Text,
@@ -153,6 +154,48 @@ class RecordingRecord(Base):
         return f"<Recording {self.id} call={self.call_id} {self.path}>"
 
 
+class User(Base):
+    """An SSO-provisioned identity. The gateway is owner-only: the single
+    owner is the user whose `name` matches settings.owner_name; everyone else
+    is created on first login but reaches nothing (403 on every surface)."""
+
+    __tablename__ = "users"
+
+    id = Column(String, primary_key=True)  # uuid4().hex, set in Python
+    name = Column(String, nullable=False)  # Casdoor username — owner-match key
+    display_name = Column(String, nullable=True)  # Casdoor display name (UI only)
+    email = Column(String, nullable=True, unique=True)
+    casdoor_sub = Column(String, nullable=True, unique=True)  # OIDC subject claim
+    created_at = Column(DateTime, default=func.now())
+    updated_at = Column(DateTime, default=func.now(), onupdate=func.now())
+
+    def __repr__(self) -> str:
+        return f"<User {self.id} {self.name}>"
+
+
+class PersonalAccessToken(Base):
+    """Long-lived bearer token for API/MCP clients (Claude Desktop, Cline)
+    that can't refresh a JWT. Plaintext is shown once at creation; only the
+    SHA-256 hash is persisted. Soft-revoked by setting revoked_at."""
+
+    __tablename__ = "personal_access_tokens"
+
+    id = Column(String, primary_key=True)  # uuid4().hex
+    user_id = Column(
+        String, ForeignKey("users.id", ondelete="CASCADE"), nullable=False, index=True
+    )
+    name = Column(String, nullable=False)
+    token_hash = Column(String, nullable=False, unique=True, index=True)
+    token_prefix = Column(String, nullable=False)  # for display, not a secret
+    created_at = Column(DateTime, default=func.now())
+    last_used_at = Column(DateTime, nullable=True)
+    expires_at = Column(DateTime, nullable=True)
+    revoked_at = Column(DateTime, nullable=True)
+
+    def __repr__(self) -> str:
+        return f"<PersonalAccessToken {self.id} user={self.user_id}>"
+
+
 # ============================================================
 # Engine & Session
 # ============================================================

db/migrations/versions/a1b2c3d4e5f6_users_and_pats.py

@@ -0,0 +1,55 @@
+"""users and personal access tokens
+
+Revision ID: a1b2c3d4e5f6
+Revises: 5187577efc23
+Create Date: 2026-07-22 00:00:00.000000
+
+"""
+from typing import Sequence, Union
+
+import sqlalchemy as sa
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision: str = 'a1b2c3d4e5f6'
+down_revision: Union[str, None] = '5187577efc23'
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    op.create_table('users',
+    sa.Column('id', sa.String(), nullable=False),
+    sa.Column('name', sa.String(), nullable=False),
+    sa.Column('display_name', sa.String(), nullable=True),
+    sa.Column('email', sa.String(), nullable=True),
+    sa.Column('casdoor_sub', sa.String(), nullable=True),
+    sa.Column('created_at', sa.DateTime(), nullable=True),
+    sa.Column('updated_at', sa.DateTime(), nullable=True),
+    sa.PrimaryKeyConstraint('id'),
+    sa.UniqueConstraint('email'),
+    sa.UniqueConstraint('casdoor_sub')
+    )
+    op.create_table('personal_access_tokens',
+    sa.Column('id', sa.String(), nullable=False),
+    sa.Column('user_id', sa.String(), nullable=False),
+    sa.Column('name', sa.String(), nullable=False),
+    sa.Column('token_hash', sa.String(), nullable=False),
+    sa.Column('token_prefix', sa.String(), nullable=False),
+    sa.Column('created_at', sa.DateTime(), nullable=True),
+    sa.Column('last_used_at', sa.DateTime(), nullable=True),
+    sa.Column('expires_at', sa.DateTime(), nullable=True),
+    sa.Column('revoked_at', sa.DateTime(), nullable=True),
+    sa.ForeignKeyConstraint(['user_id'], ['users.id'], ondelete='CASCADE'),
+    sa.PrimaryKeyConstraint('id'),
+    sa.UniqueConstraint('token_hash')
+    )
+    op.create_index(op.f('ix_personal_access_tokens_token_hash'), 'personal_access_tokens', ['token_hash'], unique=True)
+    op.create_index(op.f('ix_personal_access_tokens_user_id'), 'personal_access_tokens', ['user_id'], unique=False)
+
+
+def downgrade() -> None:
+    op.drop_index(op.f('ix_personal_access_tokens_user_id'), table_name='personal_access_tokens')
+    op.drop_index(op.f('ix_personal_access_tokens_token_hash'), table_name='personal_access_tokens')
+    op.drop_table('personal_access_tokens')
+    op.drop_table('users')

docker-compose.yaml

@@ -0,0 +1,84 @@
+# Local Hold Slayer stack: the single app image + its own PostgreSQL.
+#
+# Hold Slayer is ONE FastAPI process exposing REST/WS/MCP and serving its built
+# SvelteKit dashboard at "/" — no separate web/nginx service (the Dockerfile's
+# node stage builds the dashboard into the image).
+#
+# Auth is Casdoor SSO (owner-only). Because the published port binds the app to
+# 0.0.0.0, dev-owner mode (CASDOOR_ENABLED=false) is intentionally REFUSED at
+# startup here — that mode is loopback-only. So the stack expects the CASDOOR_*
+# + OWNER_NAME vars set (see .env.compose.example). MCP/CLI clients then use an
+# owner-minted PAT.
+#
+#   cp .env.compose.example .env
+#   # fill in CASDOOR_* + OWNER_NAME (+ HS_DB_PASSWORD)
+#   docker compose up --build
+
+services:
+  db:
+    image: postgres:17
+    environment:
+      POSTGRES_USER: ${HS_DB_USER:-holdslayer}
+      POSTGRES_PASSWORD: ${HS_DB_PASSWORD:?set HS_DB_PASSWORD in .env}
+      POSTGRES_DB: ${HS_DB_NAME:-holdslayer}
+    volumes:
+      - hs_pgdata:/var/lib/postgresql/data
+    # json-file + Alloy docker-socket discovery is the estate pattern; no
+    # syslog driver / 514xx listener (which would block container creation when
+    # the listener is absent). See ouranos Rosalind/Virgo logging convention.
+    logging:
+      driver: json-file
+      options:
+        max-size: "10m"
+        max-file: "3"
+    healthcheck:
+      test: ["CMD-SHELL", "pg_isready -U ${HS_DB_USER:-holdslayer} -d ${HS_DB_NAME:-holdslayer}"]
+      interval: 10s
+      timeout: 5s
+      retries: 5
+    restart: unless-stopped
+
+  app:
+    build: .
+    depends_on:
+      db:
+        condition: service_healthy
+    environment:
+      # Migrations run in the app's own init_db() on boot; it just needs to
+      # reach the db service. asyncpg URL points at the compose service name.
+      DATABASE_URL: postgresql+asyncpg://${HS_DB_USER:-holdslayer}:${HS_DB_PASSWORD}@db:5432/${HS_DB_NAME:-holdslayer}
+      HOST: "0.0.0.0"
+      PORT: "21081"
+      # Mock SIP: this stack is a dev/local deploy, not a real trunk. /health
+      # honestly reports engine=mock as "degraded". Flip to false + fill the
+      # SIP_TRUNK_* vars for a real-trunk deploy.
+      USE_MOCK_SIP: ${USE_MOCK_SIP:-true}
+      # --- Auth: Casdoor SSO (owner-only) ---
+      CASDOOR_ENABLED: ${CASDOOR_ENABLED:-true}
+      CASDOOR_ENDPOINT: ${CASDOOR_ENDPOINT:-https://id.ouranos.helu.ca}
+      CASDOOR_CLIENT_ID: ${CASDOOR_CLIENT_ID}
+      CASDOOR_CLIENT_SECRET: ${CASDOOR_CLIENT_SECRET}
+      CASDOOR_ORG_NAME: ${CASDOOR_ORG_NAME:-heluca}
+      CASDOOR_APP_NAME: ${CASDOOR_APP_NAME:-hold-slayer}
+      OWNER_NAME: ${OWNER_NAME:?set OWNER_NAME (the owner's Casdoor username)}
+      PUBLIC_BASE_URL: ${PUBLIC_BASE_URL:-}
+    ports:
+      - "${HS_APP_PORT:-21081}:21081"
+    logging:
+      driver: json-file
+      options:
+        max-size: "10m"
+        max-file: "3"
+    healthcheck:
+      # /health returns 200 even when "degraded" (mock engine / unregistered
+      # trunk) — so a 200 means the process is up and serving, which is the
+      # right liveness signal for a mock-SIP dev stack.
+      test: ["CMD", "curl", "-f", "http://localhost:21081/health"]
+      interval: 30s
+      timeout: 10s
+      retries: 3
+      start_period: 40s
+    restart: unless-stopped
+
+volumes:
+  hs_pgdata:

docs/configuration.md

@@ -4,6 +4,25 @@ All configuration is via environment variables, loaded through Pydantic Settings
 
 ## Environment Variables
 
+### Auth (Casdoor SSO + owner)
+
+The gateway is **owner-only**: the browser signs in via Casdoor (JWT), MCP/CLI
+clients use owner-minted PATs, and only `OWNER_NAME` may use any surface. With
+`CASDOOR_ENABLED=false` the gateway runs in dev-owner mode — permitted **only** on
+a loopback `HOST`. Startup refuses SSO-enabled-with-missing-config and
+SSO-disabled-off-loopback.
+
+| Variable | Description | Default | Required |
+|----------|-------------|---------|----------|
+| `CASDOOR_ENABLED` | Enable Casdoor SSO | `false` | No |
+| `CASDOOR_ENDPOINT` | Casdoor base URL | `https://id.ouranos.helu.ca` | If SSO on |
+| `CASDOOR_CLIENT_ID` | Casdoor application client ID | — | If SSO on |
+| `CASDOOR_CLIENT_SECRET` | Casdoor application client secret | — | If SSO on |
+| `CASDOOR_ORG_NAME` | Casdoor organization | `heluca` | No |
+| `CASDOOR_APP_NAME` | Casdoor application name | — | No |
+| `OWNER_NAME` | Casdoor username of the single operator | — | If SSO on |
+| `PUBLIC_BASE_URL` | Public base URL for OAuth discovery (else derived) | — | No |
+
 ### SIP Trunk
 
 | Variable | Description | Default | Required |

docs/mcp-server.md

@@ -3,8 +3,11 @@
 The MCP (Model Context Protocol) server lets any MCP-compatible AI assistant
 control the Hold Slayer gateway. Built with [FastMCP](https://github.com/jlowin/fastmcp),
 it is mounted on the FastAPI app at **`/mcp/`** (trailing slash) over
-**streamable HTTP** and authenticates with the same static bearer token as the
-REST API and WebSocket.
+**streamable HTTP** and authenticates with an owner-minted Personal Access Token
+(`hs_pat_…`) — the same owner-only auth as the REST API and WebSocket. Auth is
+enforced by an ASGI guard (`_owner_only_mcp` in `main.py`) that resolves the
+bearer to the owner; a Casdoor JWT also works, but MCP clients can't refresh one,
+so a PAT is the intended credential.
 
 ## Overview
 
@@ -158,7 +161,7 @@ Claude Code:
 
 ```bash
 claude mcp add hold-slayer --transport http http://localhost:8000/mcp/ \
-  --header "Authorization: Bearer $API_TOKEN"
+  --header "Authorization: Bearer hs_pat_..."
 ```
 
 Generic MCP client configuration:
@@ -168,7 +171,7 @@ Generic MCP client configuration:
   "mcpServers": {
     "hold-slayer": {
       "url": "http://localhost:8000/mcp/",
-      "headers": {"Authorization": "Bearer <API_TOKEN>"}
+      "headers": {"Authorization": "Bearer hs_pat_..."}
     }
   }
 }

main.py

@@ -12,17 +12,21 @@ Usage:
 """
 
 import logging
+import secrets
 import sys
+import time
 from contextlib import asynccontextmanager
 
-from fastapi import Depends, FastAPI
+from fastapi import Depends, FastAPI, Request
+from fastapi.responses import JSONResponse
 from fastapi.staticfiles import StaticFiles
 
-from api import call_flows, call_history, calls, devices, routing, websocket
-from api.deps import require_token
+from api import auth as auth_router
+from api import call_flows, call_history, calls, devices, routing, tokens, websocket
+from auth import get_current_owner, init_jwks_client, is_owner, resolve_from_header_or_query
 from config import Settings, get_settings
 from core.gateway import AIPSTNGateway, build_sip_engine
-from db.database import close_db, init_db
+from db.database import close_db, init_db, session_scope
 from mcp_server.server import create_mcp_server
 from models.call import CallMode
 from services.audio_classifier import AudioClassifier
@@ -106,16 +110,38 @@ def _check_startup_config(settings: Settings) -> None:
         )
         sys.exit(1)
 
-    token = settings.api_token.get_secret_value()
-    if not token and settings.host not in ("127.0.0.1", "localhost", "::1"):
+    loopback = settings.host in ("127.0.0.1", "localhost", "::1")
+    if settings.casdoor.enabled:
+        c = settings.casdoor
+        missing = [
+            name
+            for name, val in (
+                ("CASDOOR_ENDPOINT", c.endpoint),
+                ("CASDOOR_CLIENT_ID", c.client_id),
+                ("CASDOOR_CLIENT_SECRET", c.client_secret.get_secret_value()),
+                ("OWNER_NAME", settings.owner_name),
+            )
+            if not val
+        ]
+        if missing:
             logger.critical(
                 "\n"
-            "❌  API_TOKEN is not set but HOST binds beyond loopback "
+                "❌  CASDOOR_ENABLED=true but required settings are missing:\n"
+                f"    {', '.join(missing)}\n"
+                "    Set them in .env (Casdoor app credentials + the owner's "
+                "Casdoor username), or set CASDOOR_ENABLED=false with HOST=127.0.0.1 "
+                "for tokenless local development."
+            )
+            sys.exit(1)
+    elif not loopback:
+        logger.critical(
+            "\n"
+            "❌  CASDOOR_ENABLED=false but HOST binds beyond loopback "
             f"({settings.host}).\n"
-            "    Every surface (REST, WebSocket, MCP make_call) would be open "
-            "to the network.\n"
-            "    Set API_TOKEN in .env (e.g. `openssl rand -hex 32`), or set "
-            "HOST=127.0.0.1 for tokenless local development."
+            "    Every surface (REST, WebSocket, MCP make_call) would resolve "
+            "to the dev owner — open to the network.\n"
+            "    Set CASDOOR_ENABLED=true (with the Casdoor + OWNER_NAME settings), "
+            "or set HOST=127.0.0.1 for tokenless local development."
         )
         sys.exit(1)
 
@@ -126,6 +152,10 @@ async def lifespan(app: FastAPI):
     settings = get_settings()
     _check_startup_config(settings)
 
+    # Prefetch Casdoor's JWKS so the first authenticated request doesn't pay
+    # the network round-trip (no-op when SSO is disabled).
+    init_jwks_client()
+
     # The MCP session manager lives in the mounted sub-app's lifespan;
     # without entering it, every /mcp request 500s.
     async with mcp_http_app.lifespan(app):
@@ -214,7 +244,11 @@ async def lifespan(app: FastAPI):
                     display_port = int(sys.argv[i + 1])
                 except ValueError:
                     pass
-        auth_state = "bearer token required" if settings.api_token.get_secret_value() else "auth disabled (loopback)"
+        auth_state = (
+            f"Casdoor SSO (owner: {settings.owner_name or 'UNSET'})"
+            if settings.casdoor.enabled
+            else "dev-owner (loopback, no auth)"
+        )
         logger.info(f"   API:       http://{display_host}:{display_port}  [{auth_state}]")
         logger.info(f"   API Docs:  http://{display_host}:{display_port}/docs")
         logger.info(f"   WebSocket: ws://{display_host}:{display_port}/ws/events")
@@ -236,12 +270,92 @@ def _get_gateway_instance() -> AIPSTNGateway | None:
     return getattr(app.state, "gateway", None)
 
 
-mcp = create_mcp_server(
-    _get_gateway_instance,
-    api_token=get_settings().api_token.get_secret_value(),
-)
+mcp = create_mcp_server(_get_gateway_instance)
 mcp_http_app = mcp.http_app(path="/")
 
+
+def _public_base_url(scope_or_request) -> str:
+    """Resolve this service's public base URL (scheme + host), no trailing slash.
+
+    Precedence: explicit PUBLIC_BASE_URL override → X-Forwarded-Proto/Host
+    (nginx/HAProxy) → Host header → localhost. Accepts either a FastAPI
+    ``Request`` or a raw ASGI ``scope`` so the ASGI MCP guard and the FastAPI
+    discovery endpoints share one implementation.
+    """
+    settings = get_settings()
+    if settings.public_base_url:
+        return settings.public_base_url.rstrip("/")
+
+    if hasattr(scope_or_request, "headers"):
+        headers = {k.lower(): v for k, v in scope_or_request.headers.items()}
+        default_scheme = getattr(scope_or_request.url, "scheme", None) or "http"
+    else:
+        headers = {
+            k.decode("latin-1").lower(): v.decode("latin-1")
+            for k, v in scope_or_request.get("headers", [])
+        }
+        default_scheme = scope_or_request.get("scheme", "http")
+
+    proto = (headers.get("x-forwarded-proto") or default_scheme).split(",", 1)[0].strip()
+    host = (headers.get("x-forwarded-host") or headers.get("host") or "localhost")
+    host = host.split(",", 1)[0].strip()
+    return f"{proto}://{host}"
+
+
+def _owner_only_mcp(inner_app):
+    """Wrap the mounted MCP ASGI app to require an owner bearer token.
+
+    MCP tools reach state via the FastMCP lifespan context, not FastAPI's
+    dependency system, so ``Depends`` can't gate ``/mcp``. Instead we read the
+    ASGI scope's ``Authorization`` header, resolve it (Casdoor JWT or PAT)
+    against a fresh DB session, and short-circuit non-owner requests with
+    401/403. In dev mode this resolves to the dev owner, so local development
+    keeps working without a token.
+    """
+
+    async def _send_status(send, scope, status: int, body: bytes) -> None:
+        base = _public_base_url(scope)
+        resource_metadata_url = f"{base}/.well-known/oauth-protected-resource/mcp"
+        await send(
+            {
+                "type": "http.response.start",
+                "status": status,
+                "headers": [
+                    (b"content-type", b"application/json"),
+                    (
+                        b"www-authenticate",
+                        f'Bearer realm="hold-slayer-mcp", '
+                        f'resource_metadata="{resource_metadata_url}"'.encode(),
+                    ),
+                ],
+            }
+        )
+        await send({"type": "http.response.body", "body": body})
+
+    async def app(scope, receive, send):
+        if scope["type"] != "http":
+            await inner_app(scope, receive, send)
+            return
+
+        authorization = None
+        for name, value in scope.get("headers", []):
+            if name == b"authorization":
+                authorization = value.decode("latin-1")
+                break
+
+        async with session_scope() as session:
+            user = await resolve_from_header_or_query(session, authorization, None)
+            if user is None:
+                await _send_status(send, scope, 401, b'{"detail":"Not authenticated"}')
+                return
+            if not is_owner(user):
+                await _send_status(send, scope, 403, b'{"detail":"Owner access required"}')
+                return
+
+        await inner_app(scope, receive, send)
+
+    return app
+
 app = FastAPI(
     title="Hold Slayer Gateway",
     description=(
@@ -258,9 +372,13 @@ app = FastAPI(
 )
 
 # === API Routes ===
+# Every protected surface is gated to the owner (Casdoor JWT or PAT). The
+# unauthenticated OIDC endpoints live on the /auth router (login/callback/…).
 # call_history must register before calls: both live under /api/v1/calls and
 # calls' GET /{call_id} would otherwise capture the literal path "history".
-_auth = [Depends(require_token)]
+_auth = [Depends(get_current_owner)]
+app.include_router(auth_router.router)
+app.include_router(tokens.router, dependencies=_auth)
 app.include_router(
     call_history.router, prefix="/api/v1/calls", tags=["Call History"], dependencies=_auth
 )
@@ -270,11 +388,128 @@ app.include_router(
 )
 app.include_router(devices.router, prefix="/api/v1/devices", tags=["Devices"], dependencies=_auth)
 app.include_router(routing.router, prefix="/api/v1/routing", tags=["Routing"], dependencies=_auth)
-# WebSocket endpoints check the token themselves (query param or header)
+# WebSocket endpoints check the owner themselves (query param or header)
 app.include_router(websocket.router, prefix="/ws", tags=["WebSocket"])
 
-# === MCP (streamable HTTP; clients connect to /mcp/ with the bearer token) ===
-app.mount("/mcp", mcp_http_app)
+# === MCP (streamable HTTP; clients connect to /mcp/ with a PAT or JWT) ===
+# The ASGI guard resolves the bearer to the owner before the inner app runs.
+app.mount("/mcp", _owner_only_mcp(mcp_http_app))
+
+# In-memory store of dynamically registered OAuth clients (RFC 7591). MCP
+# clients re-register each session; the real gate is the bearer token.
+_registered_clients: dict[str, dict] = {}
+
+
+@app.get("/.well-known/oauth-protected-resource", include_in_schema=False)
+@app.get("/.well-known/oauth-protected-resource/mcp", include_in_schema=False)
+async def oauth_protected_resource_metadata(request: Request):
+    """RFC 9728 Protected Resource Metadata — points MCP clients at the AS.
+
+    ``resource`` advertises ``{base}/mcp`` (not the bare origin) because recent
+    ``mcp-remote`` versions verify it matches the URL they connected to.
+    """
+    base = _public_base_url(request)
+    return JSONResponse(
+        {
+            "resource": f"{base}/mcp",
+            "authorization_servers": [base],
+            "bearer_methods_supported": ["header"],
+            "resource_documentation": f"{base}/docs",
+        }
+    )
+
+
+@app.get("/.well-known/oauth-authorization-server", include_in_schema=False)
+async def oauth_authorization_server_metadata(request: Request):
+    """RFC 8414 Authorization Server Metadata.
+
+    When Casdoor SSO is on, the real authorization server is Casdoor — advertise
+    its endpoints. In dev mode there's no OAuth server; clients supply a PAT
+    directly in their MCP configuration.
+    """
+    base = _public_base_url(request)
+    settings = get_settings()
+    if settings.casdoor.enabled:
+        casdoor_base = settings.casdoor.endpoint.rstrip("/")
+        return JSONResponse(
+            {
+                "issuer": casdoor_base,
+                "authorization_endpoint": f"{casdoor_base}/login/oauth/authorize",
+                "token_endpoint": f"{casdoor_base}/api/login/oauth/access_token",
+                "jwks_uri": f"{casdoor_base}/.well-known/jwks",
+                "registration_endpoint": f"{base}/register",
+                "response_types_supported": ["code"],
+                "grant_types_supported": ["authorization_code"],
+                "token_endpoint_auth_methods_supported": ["client_secret_post"],
+                "scopes_supported": ["openid", "profile", "email"],
+            }
+        )
+    return JSONResponse(
+        {
+            "issuer": base,
+            "authorization_endpoint": f"{base}/auth/login",
+            "token_endpoint": f"{base}/auth/callback",
+            "registration_endpoint": f"{base}/register",
+            "response_types_supported": ["code"],
+            "grant_types_supported": ["authorization_code"],
+        }
+    )
+
+
+@app.post("/register", include_in_schema=False)
+async def oauth_dynamic_registration(request: Request):
+    """RFC 7591 Dynamic Client Registration — accept any well-formed request.
+
+    Registered clients are held in memory (ephemeral); the real security gate
+    is the bearer token (PAT or Casdoor JWT) on every /mcp request.
+    """
+    try:
+        body = await request.json()
+    except Exception:
+        return JSONResponse(
+            status_code=400,
+            content={
+                "error": "invalid_client_metadata",
+                "error_description": "Request body must be valid JSON.",
+            },
+        )
+
+    redirect_uris = body.get("redirect_uris")
+    if not redirect_uris or not isinstance(redirect_uris, list):
+        return JSONResponse(
+            status_code=400,
+            content={
+                "error": "invalid_redirect_uri",
+                "error_description": "redirect_uris is required and must be a non-empty list.",
+            },
+        )
+
+    client_id = secrets.token_hex(16)
+    now = int(time.time())
+    _registered_clients[client_id] = {
+        "client_id": client_id,
+        "client_id_issued_at": now,
+        "redirect_uris": redirect_uris,
+        "grant_types": body.get("grant_types", ["authorization_code"]),
+        "response_types": body.get("response_types", ["code"]),
+        "token_endpoint_auth_method": body.get("token_endpoint_auth_method", "none"),
+        "client_name": body.get("client_name"),
+        "scope": body.get("scope"),
+    }
+    logger.info("Registered OAuth client %s (name=%s)", client_id, body.get("client_name"))
+    return JSONResponse(
+        status_code=201,
+        content={
+            "client_id": client_id,
+            "client_id_issued_at": now,
+            "redirect_uris": redirect_uris,
+            "grant_types": _registered_clients[client_id]["grant_types"],
+            "response_types": _registered_clients[client_id]["response_types"],
+            "token_endpoint_auth_method": _registered_clients[client_id][
+                "token_endpoint_auth_method"
+            ],
+        },
+    )
 
 
 @app.get("/api/v1/status", tags=["System"], dependencies=_auth)

mcp_server/server.py

@@ -27,7 +27,6 @@ logger = logging.getLogger(__name__)
 
 def create_mcp_server(
     get_gateway: Callable[[], Optional[AIPSTNGateway]],
-    api_token: str = "",
 ) -> FastMCP:
     """
     Create and configure the MCP server with all tools and resources.
@@ -35,14 +34,13 @@ def create_mcp_server(
     The gateway is resolved lazily per request via `get_gateway` so the
     server can be mounted at app construction, before the lifespan has
     started the gateway.
+
+    Auth is **not** configured on the FastMCP instance: the mounted `/mcp`
+    ASGI app is gated by `_owner_only_mcp` in main.py, which resolves a
+    Casdoor JWT or PAT to the owner (one resolver shared with REST/WS) —
+    so PATs and JWTs both work here with a single code path.
     """
-    auth = None
-    if api_token:
-        from fastmcp.server.auth.providers.jwt import StaticTokenVerifier
-
-        auth = StaticTokenVerifier(tokens={api_token: {"client_id": "hold-slayer"}})
-
-    mcp = FastMCP("Hold Slayer Gateway", auth=auth)
+    mcp = FastMCP("Hold Slayer Gateway", auth=None)
 
     def require_gateway() -> AIPSTNGateway:
         gateway = get_gateway()

pyproject.toml

@@ -32,9 +32,13 @@ dependencies = [
     # HTTP client (for Speaches STT)
     "httpx>=0.28.0",
 
-    # MCP server (3.x — http_app + StaticTokenVerifier)
+    # MCP server (3.x — http_app + ASGI owner guard)
     "fastmcp>=3.0.0",
 
+    # Auth: Casdoor SSO (OAuth code exchange) + RS256 JWT validation
+    "casdoor>=1.0",
+    "pyjwt[crypto]>=2.8",
+
     # Utilities
     "python-slugify>=8.0.0",
 ]
@@ -47,6 +51,8 @@ dev = [
     "httpx>=0.28.0",
     "ruff>=0.8.0",
     "aiosqlite>=0.22.0",
+    # Mint RS256 JWTs in tests without a live Casdoor
+    "cryptography>=42.0",
 ]
 
 [tool.setuptools.packages.find]

tests/test_api_security.py

@@ -1,24 +1,53 @@
 """
-API surface tests — bearer-token enforcement and route registration order.
+API surface tests — owner enforcement and route registration order.
 
 The app is exercised without its lifespan: auth runs before any handler,
-so a 503 ("Gateway not initialized") proves the token was accepted.
+so a 503 ("Gateway not initialized") proves the caller was accepted as
+owner. Auth internals (JWT/PAT resolution) are covered in test_auth.py;
+here we assert the routers are gated and the routes register in the right
+order. In dev-owner mode (SSO disabled) a tokenless request is the owner.
 """
 
 import httpx
 import pytest
-from pydantic import SecretStr
+from sqlalchemy.ext.asyncio import async_sessionmaker, create_async_engine
+from sqlalchemy.pool import StaticPool
 from starlette.routing import Match
 
+import db.database as dbmod
 import main
 from config import get_settings
-
-TOKEN = "test-token-for-suite"
+from db.database import Base
 
 
 @pytest.fixture
-def token_enabled(monkeypatch):
-    monkeypatch.setattr(get_settings(), "api_token", SecretStr(TOKEN))
+async def mem_db(monkeypatch):
+    engine = create_async_engine(
+        "sqlite+aiosqlite:///:memory:",
+        poolclass=StaticPool,
+        connect_args={"check_same_thread": False},
+    )
+    async with engine.begin() as conn:
+        await conn.run_sync(Base.metadata.create_all)
+    factory = async_sessionmaker(engine, expire_on_commit=False)
+    monkeypatch.setattr(dbmod, "_engine", engine)
+    monkeypatch.setattr(dbmod, "_session_factory", factory)
+    yield factory
+    await engine.dispose()
+
+
+@pytest.fixture
+def dev_owner(monkeypatch):
+    """SSO disabled — every request resolves to the dev owner."""
+    monkeypatch.setattr(get_settings().casdoor, "enabled", False)
+
+
+@pytest.fixture
+def sso_enabled(monkeypatch):
+    """SSO enabled with no credentials supplied → 401 on protected routes."""
+    monkeypatch.setattr(get_settings().casdoor, "enabled", True)
+    monkeypatch.setattr(get_settings().casdoor, "endpoint", "https://id.example.test")
+    monkeypatch.setattr(get_settings(), "owner_name", "owner@example.test")
 
 
 @pytest.fixture
@@ -28,45 +57,34 @@ async def client():
         yield c
 
 
-class TestBearerToken:
-    async def test_missing_token_rejected(self, token_enabled, client):
+class TestOwnerGate:
+    async def test_dev_owner_reaches_handler(self, dev_owner, mem_db, client):
+        resp = await client.get("/api/v1/calls/active")
+        assert resp.status_code == 503  # dev-owner accepted; handler 503s (no lifespan)
+
+    async def test_sso_missing_credentials_rejected(self, sso_enabled, mem_db, client):
         resp = await client.get("/api/v1/calls/active")
         assert resp.status_code == 401
         assert resp.headers["www-authenticate"] == "Bearer"
 
-    async def test_wrong_token_rejected(self, token_enabled, client):
-        resp = await client.get(
-            "/api/v1/calls/active", headers={"Authorization": "Bearer wrong"}
-        )
-        assert resp.status_code == 401
-
-    async def test_valid_token_reaches_handler(self, token_enabled, client):
-        resp = await client.get(
-            "/api/v1/calls/active", headers={"Authorization": f"Bearer {TOKEN}"}
-        )
-        # No lifespan ran, so the handler itself 503s — auth was accepted
-        assert resp.status_code == 503
-
-    async def test_empty_token_disables_auth(self, monkeypatch, client):
-        monkeypatch.setattr(get_settings(), "api_token", SecretStr(""))
-        resp = await client.get("/api/v1/calls/active")
-        assert resp.status_code == 503
-
-    async def test_query_param_token_accepted(self, token_enabled, client):
-        """<audio>/<a> elements can't set headers — ?token= must work."""
-        resp = await client.get(f"/api/v1/calls/active?token={TOKEN}")
-        assert resp.status_code == 503  # auth accepted, handler 503s (no lifespan)
-
-    async def test_wrong_query_param_token_rejected(self, token_enabled, client):
-        resp = await client.get("/api/v1/calls/active?token=wrong")
-        assert resp.status_code == 401
-
-    async def test_all_api_routers_protected(self, token_enabled, client):
-        for path in ("/api/v1/calls/active", "/api/v1/call-flows/", "/api/v1/devices/",
-                     "/api/v1/routing/rules", "/api/v1/calls/history"):
+    async def test_all_api_routers_protected(self, sso_enabled, mem_db, client):
+        for path in (
+            "/api/v1/calls/active",
+            "/api/v1/call-flows/",
+            "/api/v1/devices/",
+            "/api/v1/routing/rules",
+            "/api/v1/calls/history",
+            "/api/v1/tokens",
+        ):
             resp = await client.get(path)
             assert resp.status_code == 401, path
 
+    async def test_auth_routes_are_public(self, sso_enabled, mem_db, client):
+        """The OIDC endpoints must be reachable without a token."""
+        # /auth/me with no token → 401 (not 403); /auth/login → redirect to Casdoor
+        resp = await client.get("/auth/login", follow_redirects=False)
+        assert resp.status_code in (302, 307)
+
 
 class TestRouteOrder:
     def _resolve(self, path: str):

tests/test_auth.py

@@ -0,0 +1,239 @@
+"""
+Auth tests — Casdoor JWT + PAT resolution, owner gating, dev-owner mode.
+
+No live Casdoor: we generate an RSA keypair, stub the JWKS client so
+`_decode_casdoor_jwt` trusts our public key, and mint RS256 JWTs locally.
+The app is exercised without its lifespan, so a 503 ("Gateway not
+initialized") proves auth was accepted and the request reached a handler.
+
+DB access (resolve_bearer → users/PATs) hits an in-memory SQLite database
+wired in via the `mem_db` fixture, mirroring tests/test_data_layer.py.
+"""
+
+import time
+import uuid
+
+import httpx
+import jwt
+import pytest
+from cryptography.hazmat.primitives import serialization
+from cryptography.hazmat.primitives.asymmetric import rsa
+from sqlalchemy.ext.asyncio import async_sessionmaker, create_async_engine
+from sqlalchemy.pool import StaticPool
+
+import auth as authmod
+import db.database as dbmod
+import main
+from config import get_settings
+from db.database import Base, PersonalAccessToken, User
+
+ENDPOINT = "https://id.example.test"
+OWNER = "owner@example.test"
+
+
+# ── RSA keypair + JWKS stub ──────────────────────────────────────────────────
+
+
+@pytest.fixture(scope="module")
+def keypair():
+    key = rsa.generate_private_key(public_exponent=65537, key_size=2048)
+    private_pem = key.private_bytes(
+        serialization.Encoding.PEM,
+        serialization.PrivateFormat.PKCS8,
+        serialization.NoEncryption(),
+    )
+    return private_pem, key.public_key()
+
+
+def _mint(private_pem, *, sub, name, email=None, exp_delta=3600):
+    claims = {
+        "iss": ENDPOINT,
+        "sub": sub,
+        "name": name,
+        "displayName": name,
+        "exp": int(time.time()) + exp_delta,
+        "iat": int(time.time()),
+    }
+    if email:
+        claims["email"] = email
+    return jwt.encode(claims, private_pem, algorithm="RS256")
+
+
+class _StubJWKS:
+    """Stands in for jwt.PyJWKClient — returns our fixed public key."""
+
+    def __init__(self, public_key):
+        self._key = public_key
+
+    def get_signing_key_from_jwt(self, token):
+        class _K:
+            key = self._key
+
+        return _K()
+
+    def fetch_data(self):
+        pass
+
+
+@pytest.fixture
+def sso_enabled(monkeypatch, keypair):
+    """Enable Casdoor SSO with a known owner and a stubbed JWKS client."""
+    _, public_key = keypair
+    settings = get_settings()
+    monkeypatch.setattr(settings.casdoor, "enabled", True)
+    monkeypatch.setattr(settings.casdoor, "endpoint", ENDPOINT)
+    monkeypatch.setattr(settings, "owner_name", OWNER)
+    monkeypatch.setattr(authmod, "_jwks_client", _StubJWKS(public_key))
+
+
+@pytest.fixture
+async def mem_db(monkeypatch):
+    engine = create_async_engine(
+        "sqlite+aiosqlite:///:memory:",
+        poolclass=StaticPool,
+        connect_args={"check_same_thread": False},
+    )
+    async with engine.begin() as conn:
+        await conn.run_sync(Base.metadata.create_all)
+    factory = async_sessionmaker(engine, expire_on_commit=False)
+    monkeypatch.setattr(dbmod, "_engine", engine)
+    monkeypatch.setattr(dbmod, "_session_factory", factory)
+    yield factory
+    await engine.dispose()
+
+
+@pytest.fixture
+async def client():
+    transport = httpx.ASGITransport(app=main.app)
+    async with httpx.AsyncClient(transport=transport, base_url="http://test") as c:
+        yield c
+
+
+async def _seed_user(factory, *, name, casdoor_sub=None, email=None) -> str:
+    uid = uuid.uuid4().hex
+    async with factory() as session:
+        session.add(
+            User(id=uid, name=name, display_name=name, email=email, casdoor_sub=casdoor_sub)
+        )
+        await session.commit()
+    return uid
+
+
+async def _seed_pat(factory, user_id, *, revoked=False, expires_at=None) -> str:
+    from datetime import UTC, datetime
+
+    plaintext = authmod.PAT_PREFIX + uuid.uuid4().hex
+    async with factory() as session:
+        pat = PersonalAccessToken(
+            id=uuid.uuid4().hex,
+            user_id=user_id,
+            name="test",
+            token_hash=authmod.hash_token(plaintext),
+            token_prefix=plaintext[: len(authmod.PAT_PREFIX) + 4],
+            revoked_at=(datetime.now(UTC) if revoked else None),
+            expires_at=expires_at,
+        )
+        session.add(pat)
+        await session.commit()
+    return plaintext
+
+
+PROTECTED = "/api/v1/calls/active"
+
+
+# ── JWT paths ────────────────────────────────────────────────────────────────
+
+
+class TestCasdoorJWT:
+    async def test_owner_jwt_reaches_handler(self, sso_enabled, mem_db, client, keypair):
+        private_pem, _ = keypair
+        token = _mint(private_pem, sub="s-owner", name=OWNER, email=OWNER)
+        resp = await client.get(PROTECTED, headers={"Authorization": f"Bearer {token}"})
+        assert resp.status_code == 503  # auth accepted; no lifespan → handler 503s
+
+    async def test_non_owner_jwt_forbidden(self, sso_enabled, mem_db, client, keypair):
+        private_pem, _ = keypair
+        token = _mint(private_pem, sub="s-guest", name="guest@example.test")
+        resp = await client.get(PROTECTED, headers={"Authorization": f"Bearer {token}"})
+        assert resp.status_code == 403
+
+    async def test_expired_jwt_unauthenticated(self, sso_enabled, mem_db, client, keypair):
+        private_pem, _ = keypair
+        token = _mint(private_pem, sub="s-owner", name=OWNER, exp_delta=-10)
+        resp = await client.get(PROTECTED, headers={"Authorization": f"Bearer {token}"})
+        assert resp.status_code == 401
+
+    async def test_garbage_token_unauthenticated(self, sso_enabled, mem_db, client):
+        resp = await client.get(PROTECTED, headers={"Authorization": "Bearer not.a.jwt"})
+        assert resp.status_code == 401
+
+    async def test_no_credentials_unauthenticated(self, sso_enabled, mem_db, client):
+        resp = await client.get(PROTECTED)
+        assert resp.status_code == 401
+        assert resp.headers["www-authenticate"] == "Bearer"
+
+
+# ── PAT paths ────────────────────────────────────────────────────────────────
+
+
+class TestPAT:
+    async def test_owner_pat_reaches_handler(self, sso_enabled, mem_db, client):
+        uid = await _seed_user(mem_db, name=OWNER, casdoor_sub="s-owner", email=OWNER)
+        pat = await _seed_pat(mem_db, uid)
+        resp = await client.get(PROTECTED, headers={"Authorization": f"Bearer {pat}"})
+        assert resp.status_code == 503
+
+    async def test_non_owner_pat_forbidden(self, sso_enabled, mem_db, client):
+        uid = await _seed_user(mem_db, name="guest@example.test", casdoor_sub="s-guest")
+        pat = await _seed_pat(mem_db, uid)
+        resp = await client.get(PROTECTED, headers={"Authorization": f"Bearer {pat}"})
+        assert resp.status_code == 403
+
+    async def test_revoked_pat_unauthenticated(self, sso_enabled, mem_db, client):
+        uid = await _seed_user(mem_db, name=OWNER, casdoor_sub="s-owner", email=OWNER)
+        pat = await _seed_pat(mem_db, uid, revoked=True)
+        resp = await client.get(PROTECTED, headers={"Authorization": f"Bearer {pat}"})
+        assert resp.status_code == 401
+
+    async def test_expired_pat_unauthenticated(self, sso_enabled, mem_db, client):
+        from datetime import UTC, datetime, timedelta
+
+        uid = await _seed_user(mem_db, name=OWNER, casdoor_sub="s-owner", email=OWNER)
+        pat = await _seed_pat(mem_db, uid, expires_at=datetime.now(UTC) - timedelta(minutes=1))
+        resp = await client.get(PROTECTED, headers={"Authorization": f"Bearer {pat}"})
+        assert resp.status_code == 401
+
+    async def test_unknown_pat_unauthenticated(self, sso_enabled, mem_db, client):
+        resp = await client.get(
+            PROTECTED, headers={"Authorization": f"Bearer {authmod.PAT_PREFIX}nope"}
+        )
+        assert resp.status_code == 401
+
+
+# ── Dev-owner mode (SSO disabled) ────────────────────────────────────────────
+
+
+class TestDevOwnerMode:
+    async def test_tokenless_request_is_owner(self, monkeypatch, mem_db, client):
+        monkeypatch.setattr(get_settings().casdoor, "enabled", False)
+        resp = await client.get(PROTECTED)
+        assert resp.status_code == 503  # dev-owner resolved; handler 503s (no lifespan)
+
+    async def test_auth_me_reports_owner(self, monkeypatch, mem_db, client):
+        monkeypatch.setattr(get_settings().casdoor, "enabled", False)
+        resp = await client.get("/auth/me")
+        assert resp.status_code == 200
+        body = resp.json()
+        assert body["is_owner"] is True
+
+
+# ── /auth/me for a non-owner (200 + is_owner:false, not a hard 401) ──────────
+
+
+class TestAuthMe:
+    async def test_non_owner_gets_200_not_owner(self, sso_enabled, mem_db, client, keypair):
+        private_pem, _ = keypair
+        token = _mint(private_pem, sub="s-guest", name="guest@example.test")
+        resp = await client.get("/auth/me", headers={"Authorization": f"Bearer {token}"})
+        assert resp.status_code == 200
+        assert resp.json()["is_owner"] is False

tests/test_mcp.py

@@ -6,12 +6,32 @@ Uses the FastMCP in-memory client (no network, no mounted app).
 
 import pytest
 from fastmcp import Client
+from sqlalchemy.ext.asyncio import async_sessionmaker, create_async_engine
+from sqlalchemy.pool import StaticPool
 
-from config import Settings
+import db.database as dbmod
+from config import Settings, get_settings
 from core.dial_plan import is_emergency_number
 from core.gateway import AIPSTNGateway
+from db.database import Base
 from mcp_server.server import create_mcp_server
 
+
+@pytest.fixture
+async def mem_db(monkeypatch):
+    engine = create_async_engine(
+        "sqlite+aiosqlite:///:memory:",
+        poolclass=StaticPool,
+        connect_args={"check_same_thread": False},
+    )
+    async with engine.begin() as conn:
+        await conn.run_sync(Base.metadata.create_all)
+    factory = async_sessionmaker(engine, expire_on_commit=False)
+    monkeypatch.setattr(dbmod, "_engine", engine)
+    monkeypatch.setattr(dbmod, "_session_factory", factory)
+    yield factory
+    await engine.dispose()
+
 EXPECTED_TOOLS = {
     "make_call",
     "get_call_status",
@@ -43,11 +63,64 @@ class TestToolSurface:
             tools = {t.name for t in await client.list_tools()}
         assert tools == EXPECTED_TOOLS
 
-    async def test_auth_configured_when_token_given(self):
-        assert create_mcp_server(lambda: None, api_token="sekrit").auth is not None
+    async def test_no_fastmcp_auth_configured(self):
+        # Auth for /mcp is enforced by the ASGI _owner_only_mcp guard in
+        # main.py, not on the FastMCP instance itself.
         assert create_mcp_server(lambda: None).auth is None
 
 
+class TestMcpOwnerGuard:
+    """The ASGI wrapper gates /mcp before the inner app runs."""
+
+    def _wrapped(self):
+        import main
+
+        calls = {"inner": 0}
+
+        async def inner(scope, receive, send):
+            calls["inner"] += 1
+            await send({"type": "http.response.start", "status": 200, "headers": []})
+            await send({"type": "http.response.body", "body": b"ok"})
+
+        return main._owner_only_mcp(inner), calls
+
+    async def _run(self, app, headers):
+        sent = []
+
+        async def receive():
+            return {"type": "http.request", "body": b"", "more_body": False}
+
+        async def send(msg):
+            sent.append(msg)
+
+        scope = {
+            "type": "http",
+            "method": "POST",
+            "path": "/mcp/",
+            "headers": headers,
+            "query_string": b"",
+        }
+        await app(scope, receive, send)
+        status = next(m["status"] for m in sent if m["type"] == "http.response.start")
+        return status
+
+    async def test_missing_token_401(self, monkeypatch, mem_db):
+        monkeypatch.setattr(get_settings().casdoor, "enabled", True)
+        monkeypatch.setattr(get_settings().casdoor, "endpoint", "https://id.example.test")
+        monkeypatch.setattr(get_settings(), "owner_name", "owner@example.test")
+        app, calls = self._wrapped()
+        status = await self._run(app, headers=[])
+        assert status == 401
+        assert calls["inner"] == 0
+
+    async def test_dev_owner_passes(self, monkeypatch, mem_db):
+        monkeypatch.setattr(get_settings().casdoor, "enabled", False)
+        app, calls = self._wrapped()
+        status = await self._run(app, headers=[])
+        assert status == 200
+        assert calls["inner"] == 1
+
+
 class TestGatewayResolution:
     async def test_tool_errors_cleanly_before_gateway_ready(self):
         mcp = create_mcp_server(lambda: None)

tests/test_oauth_metadata.py

@@ -0,0 +1,77 @@
+"""
+OAuth discovery metadata tests (RFC 9728 / RFC 8414 / RFC 7591).
+
+MCP clients that get a 401 from /mcp perform OAuth discovery. These
+endpoints are unauthenticated and served straight from main.app.
+"""
+
+import httpx
+import pytest
+
+import main
+from config import get_settings
+
+
+@pytest.fixture
+async def client():
+    transport = httpx.ASGITransport(app=main.app)
+    async with httpx.AsyncClient(transport=transport, base_url="http://test") as c:
+        yield c
+
+
+class TestProtectedResourceMetadata:
+    async def test_resource_advertises_mcp_path(self, client):
+        resp = await client.get("/.well-known/oauth-protected-resource")
+        assert resp.status_code == 200
+        body = resp.json()
+        # mcp-remote verifies this matches the URL it connected to.
+        assert body["resource"] == "http://test/mcp"
+        assert body["authorization_servers"] == ["http://test"]
+
+    async def test_mcp_suffixed_variant(self, client):
+        resp = await client.get("/.well-known/oauth-protected-resource/mcp")
+        assert resp.status_code == 200
+        assert resp.json()["resource"] == "http://test/mcp"
+
+
+class TestAuthorizationServerMetadata:
+    async def test_advertises_casdoor_when_enabled(self, monkeypatch, client):
+        monkeypatch.setattr(get_settings().casdoor, "enabled", True)
+        monkeypatch.setattr(get_settings().casdoor, "endpoint", "https://id.example.test")
+        resp = await client.get("/.well-known/oauth-authorization-server")
+        assert resp.status_code == 200
+        body = resp.json()
+        assert body["issuer"] == "https://id.example.test"
+        assert body["jwks_uri"] == "https://id.example.test/.well-known/jwks"
+        assert body["registration_endpoint"] == "http://test/register"
+
+    async def test_dev_mode_advertises_local(self, monkeypatch, client):
+        monkeypatch.setattr(get_settings().casdoor, "enabled", False)
+        resp = await client.get("/.well-known/oauth-authorization-server")
+        assert resp.status_code == 200
+        body = resp.json()
+        assert body["issuer"] == "http://test"
+        assert body["authorization_endpoint"] == "http://test/auth/login"
+
+
+class TestDynamicRegistration:
+    async def test_registers_client(self, client):
+        resp = await client.post(
+            "/register",
+            json={"redirect_uris": ["http://localhost/cb"], "client_name": "test"},
+        )
+        assert resp.status_code == 201
+        body = resp.json()
+        assert "client_id" in body
+        assert body["redirect_uris"] == ["http://localhost/cb"]
+
+    async def test_rejects_missing_redirect_uris(self, client):
+        resp = await client.post("/register", json={"client_name": "test"})
+        assert resp.status_code == 400
+        assert resp.json()["error"] == "invalid_redirect_uri"
+
+    async def test_rejects_non_json(self, client):
+        resp = await client.post(
+            "/register", content=b"not json", headers={"content-type": "application/json"}
+        )
+        assert resp.status_code == 400

tests/test_structure.py

@@ -9,7 +9,6 @@ through the shared data layer in services/call_persistence.py.
 
 import httpx
 import pytest
-from pydantic import SecretStr
 from sqlalchemy.ext.asyncio import async_sessionmaker, create_async_engine
 from sqlalchemy.pool import StaticPool
 
@@ -116,7 +115,9 @@ class TestInboundPolicy:
 
 @pytest.fixture
 async def client(monkeypatch):
-    monkeypatch.setattr(get_settings(), "api_token", SecretStr(""))
+    # Dev-owner mode: tokenless requests resolve to the owner. auth's DB
+    # session comes through the same get_db override below.
+    monkeypatch.setattr(get_settings().casdoor, "enabled", False)
 
     engine = create_async_engine(
         "sqlite+aiosqlite:///:memory:",

tests/test_websocket.py

@@ -1,27 +1,57 @@
 """
 WebSocket event-stream tests.
 
-The socket is refused (4401) without the bearer token, and an
-authorized client immediately receives the synthetic trunk-status
-event followed by the replayed recent history.
+The socket is owner-gated: refused (4401) when SSO is enabled and no
+credential is supplied, and — in dev-owner mode (SSO disabled) — an
+authorized client immediately receives the synthetic trunk-status event
+followed by the replayed recent history.
+
+The WS `_authorize` resolves the owner via a DB session, so an in-memory
+SQLite database is wired in (StaticPool, shared across the TestClient
+thread) mirroring tests/test_data_layer.py.
 """
 
 import asyncio
 
 import pytest
-from pydantic import SecretStr
+from sqlalchemy.ext.asyncio import async_sessionmaker, create_async_engine
+from sqlalchemy.pool import StaticPool
 from starlette.testclient import TestClient
 from starlette.websockets import WebSocketDisconnect
 
+import db.database as dbmod
 import main
 from config import Settings, get_settings
 from core.gateway import AIPSTNGateway
+from db.database import Base
 from models.events import EventType, GatewayEvent
 
 
 @pytest.fixture
-def ws_app(monkeypatch):
-    monkeypatch.setattr(get_settings(), "api_token", SecretStr("tok"))
+def mem_db(monkeypatch):
+    """Synchronous setup of an in-memory SQLite DB shared with the app."""
+    engine = create_async_engine(
+        "sqlite+aiosqlite:///:memory:",
+        poolclass=StaticPool,
+        connect_args={"check_same_thread": False},
+    )
+
+    async def _create():
+        async with engine.begin() as conn:
+            await conn.run_sync(Base.metadata.create_all)
+
+    asyncio.run(_create())
+    factory = async_sessionmaker(engine, expire_on_commit=False)
+    monkeypatch.setattr(dbmod, "_engine", engine)
+    monkeypatch.setattr(dbmod, "_session_factory", factory)
+    yield
+    asyncio.run(engine.dispose())
+
+
+@pytest.fixture
+def ws_app(monkeypatch, mem_db):
+    """Dev-owner mode: a tokenless WS connect resolves the owner."""
+    monkeypatch.setattr(get_settings().casdoor, "enabled", False)
     gateway = AIPSTNGateway(settings=Settings())
     main.app.state.gateway = gateway
     yield gateway
@@ -38,7 +68,11 @@ def _publish(gateway, call_id: str) -> None:
 
 
 class TestEventStream:
-    def test_refused_without_token(self, ws_app):
+    def test_refused_without_credential(self, monkeypatch, mem_db):
+        """SSO enabled + no token → the socket is closed with 4401."""
+        monkeypatch.setattr(get_settings().casdoor, "enabled", True)
+        monkeypatch.setattr(get_settings().casdoor, "endpoint", "https://id.example.test")
+        monkeypatch.setattr(get_settings(), "owner_name", "owner@example.test")
         client = TestClient(main.app)
         with pytest.raises(WebSocketDisconnect) as exc:
             with client.websocket_connect("/ws/events"):
@@ -50,7 +84,7 @@ class TestEventStream:
         _publish(ws_app, "call_ws2")
 
         client = TestClient(main.app)
-        with client.websocket_connect("/ws/events?token=tok") as ws:
+        with client.websocket_connect("/ws/events") as ws:
             first = ws.receive_json()
             assert first["type"] == EventType.SIP_TRUNK_REGISTRATION_FAILED.value
             replayed = [ws.receive_json() for _ in range(2)]
@@ -58,9 +92,7 @@ class TestEventStream:
 
     def test_per_call_stream_filters(self, ws_app):
         client = TestClient(main.app)
-        with client.websocket_connect(
-            "/ws/calls/call_target/events?token=tok"
-        ) as ws:
+        with client.websocket_connect("/ws/calls/call_target/events") as ws:
             _publish(ws_app, "call_other")
             _publish(ws_app, "call_target")
             msg = ws.receive_json()