llm: routes that name a host, a backup and a window, and a profile that reads them ...

`llm.json` replaces `OLLAMA_BACKENDS_FILE`. It has targets (host and key
environment variable) and routes, where a route is an ordered chain of
`model @ target` entries, each with its own `context_window` and options. The
`default` route now starts on `glm-5.3-flash:cloud` and backs it with
`qwen3.6:35b-128k` on the home host, both at 131072 tokens, so a failover no
longer changes the budget the compressor and `num_ctx` are accounted against.
Embeddings get their own `embedding` route instead of a pinned host.

The engine classifies a failure (network, quota, auth, server, missing model),
cools that entry and moves the call down the chain; a background probe watches
only degraded targets and lets the next turn return to the primary once it is
healthy — it never interrupts a running turn. A profile joins a route with
`llm_route`, and the loader now actually reads it: `llm_route` and
`context_window` were declared on `AgentProfile` and written into every
`config.json` while the loader skipped both, the same silent no-op that
`is_admin_only` used to be, so every profile ran the legacy model list at the
global window. The writer emits them too, or an admin edit would erase the
route from disk.

The turn's resolved entry travels to the UI: the first chunk is stamped with
model and target, `model_info` carries `fallback` when a backup answered, and
the chat header shows a chip only then. `serving_primary` compares the target
as well as the model name, because a route may run one model on two hosts.
`/health` gains an `llm` block with the active entry, target states and missing
models.

`deploy/ollama/` keeps the local model's definition honest: MTP off (slower at
every context on this host), `presence_penalty 0` (the base model's 1.5 breaks
agentic tool calls), and `num_gpu 40` — Ollama's automatic fit claims a 128k
f16 slot fits and then dies mid-prefill with CUDA OOM; 40 of 42 layers leaves
~400 MiB and decodes at ~77 tok/s, 38 layers collapses to 48, and 39 OOMs.
q8_0 KV measured 37 tok/s against 70 on f16 here, so the cache stays f16.

mcp keys W2: per-user client clones, KeyResolver wiring, user_id in call path

container: runtime-import PushService — master could not boot since the PWA commit ...

PushService was only imported under TYPE_CHECKING; create_container raised
NameError at startup. Caught by the first real boot (e2e) — the running
production container predates the PWA commit, so master was silently
un-launchable.

agent parallelism: background tools, bg spawn_agent, parallel tool batches, message queue ...

- backgroundable tools (terminal/ssh_exec/peer/spawn_agent/code_exec) detach
  via TaskManager with per-session/global/spawn caps, rate limit and TTL
- completion delivery both ways: task_update event (out-of-band) + pending
  notes injected into the next turn; tasks tool (list/check/wait/cancel)
- parallel tool-call batches (profile-level gate, off by default) with
  ToolStarted up-front, tagged event mux, call-order results, one save,
  plus dangling tool_call repair on session load
- user message queue instead of busy error: message_queued frame,
  back-to-back drain on the same socket, headless fallback on disconnect
- pin mcp<2 (v2 renames FastMCP with breaking API changes)
- docs: tasks.md (new), websocket/api/agent/config updates, tasks manual,
  spawn_agent background param, persona contract section

PWA: installable webclient, offline shell, web push ...

Installability:
- public/manifest.webmanifest (standalone, theme #16161E) + PNG icons
  generated from logo.svg (regular + maskable, served via /images mount)
- index.html: manifest link, theme-color, apple-touch-icon

Offline shell:
- hand-rolled sw.js (no workbox): navigation = network-first (3s race)
  with cached-shell fallback + background refresh (a stale cached shell
  would 404 on entry chunks after a deploy); /assets/* cache-first
  (content-hashed); /images/* cache-first capped; /api,/ws,/auth,/push,
  /content pass-through
- vite closeBundle plugin stamps __NAVI_BUILD_VERSION__ (digest of
  index.html + asset names) into dist/sw.js; sw.js served no-store so
  every deploy reactivates the SW and activation evicts old caches
- SW registration in main.js, PROD only (dev HMR untouched)
- OfflineBanner (useOnline composable) over the app shell

Web push (VAPID, pywebpush):
- navi/push/ package: push_subscriptions table (postgres, boot-time DDL),
  PushSubscriptionStore, PushService (async fan-out, to_thread sends,
  404/410 prunes dead endpoints, per-session cooldown)
- routes: GET /push/vapid-key, POST/DELETE /push/subscribe (auth-gated)
- trigger in orchestrator run_agent + run_recall: push on StreamEnd when
  no WebSocket client watches the session; fire-and-forget, never
  disturbs the run; anonymous fallback only when auth is off
- client: usePush composable + Notifications settings panel (enable/
  disable via PushManager.subscribe with the server VAPID key)
- notification click focuses the app at /#<session_id> (hash routing
  opens the right chat); payload body is a markdown-stripped <=140-char
  preview

NAVIVAPID keys empty = push fully disabled (graceful, like other optional
integrations). dist/ artifacts committed per repo convention.

Tests: pytest push store/service/routes/trigger (+23), vitest usePush
(83 webclient tests green). Full suite 1143 passed.

terminal: session-broadcast sink + lifecycle events (Etap 1) ...

Persistent terminals emitted their stream only into a per-tool-call sink, which
closed once the open action returned — so a background dev server's output
stopped reaching the client immediately after open. TerminalClosed was never
emitted at all (the handler existed, the emitter did not).

Replace the per-tool event_sink with a single callback the orchestrator wires to
the session's WebSocket(s):

- TerminalManager: set_event_callback(cb) + _emit(session_id, event). Reader
  tasks and open/close deliver TerminalOutputDelta / TerminalOpened /
  TerminalClosed through the callback, which lives for the terminal's (and
  session's) lifetime — not the tool call's.
- New TerminalOpened event (terminal_opened: name/description/pid/background).
  TerminalClosed now actually emitted from _close_one (explicit/idle/shutdown/
  session_ended). Background output keeps streaming after open returns.
- terminal.py: drop the per-tool event_sink from _tm.open (no longer needed).
- orchestrator._on_terminal_event → _notify_session (per-session WS fan-out,
  same path recall/mcp updates use). container wires the callback after building
  the orchestrator.
- Events imported lazily inside terminal_manager methods to avoid a circular
  import (navi.core.events → navi.core.__init__ → agent → registry → navi.tools).

Tests: open emits TerminalOpened; background output streams via the callback
after open returns (gap-3 fix); close emits TerminalClosed; no callback does not
break readers. Existing terminal tests unchanged. Full suite: 968 passed, 1 skipped.

Plan: docs/terminal_tool_plan.md (5 etaps, this is Etap 1).

mcp: dedicated runner task per client (fix cross-task cancel-scope on shutdown) ...

MCP client SDK transports (stdio/sse/streamable_http) + ClientSession are
anyio task groups whose cancel scopes require __aenter__/__aexit__ in the
SAME asyncio task. McpClient entered the transport in one task (lifespan
connect / health-check reconnect / request retry) and exited it in another
(lifespan teardown) -> RuntimeError: Attempted to exit cancel scope in a
different task than it was entered in.

Refactor McpClient to own a single long-running runner task that holds the
AsyncExitStack and performs ALL transport enter/exit + list_tools/call_tool.
The public async API (connect/disconnect/list_tools/call_tool/mark_disconnected)
just enqueues a _Cmd and awaits a Future, so callers from any task no longer
cross cancel-scope boundaries. connected/instructions mirror from the runner
onto the instance to stay sync-readable. disconnect() enqueues a stop command
and awaits shield(runner) so teardown isn't interrupted by lifespan cancel.

Also call mcp_manager.stop_health_check() BEFORE disconnect_all() in
AppContainer.shutdown() so the health-check task cannot enqueue onto a
client whose runner is being torn down. mark_disconnected() is now async
(queued) and its manager caller updated.

Regression test: connect in one task, list_tools in a second, disconnect in
a third — the exact scenario that raised the RuntimeError before.

Move terminal_manager to _internal subpackage ...

- terminal_manager is an internal helper, not a tool
- Update imports in terminal.py, container.py, test_terminal.py

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Fix terminal review issues: security, lifecycle, reactivity ...

- TerminalManager.open now accepts exec_tokens to use create_subprocess_exec
  for restricted commands instead of always using shell
- Fix kill/terminate order: SIGTERM first, SIGKILL fallback
- Pop closed sessions from _sessions dict to prevent memory leak
- Add terminal_manager.shutdown() to AppContainer.shutdown()
- Wait for reader tasks in foreground open before returning output
- Add _MAX_TERMINALS_PER_SESSION limit (10)
- Wrap cleanup_idle tasks in _close_one_safe with error logging
- send_input catches BrokenPipeError/ConnectionResetError specifically
- Foreground terminals auto-close after gathering output
- Vue reactivity: replace terminals object immutably instead of mutating
- onTerminalClosed marks matching tool card as no longer pending
- Update tests for new behavior (foreground auto-close, max limit)

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Add persistent multi-session terminal tool with background support ...

- New TerminalManager module: named subprocess sessions per Navi session,
  background readers, event-sink streaming, idle auto-cleanup
- Refactor terminal tool to multi-action: run, open, close, list,
  status, send_input
- Add TerminalOutputDelta and TerminalClosed events for streaming
- Wire TerminalManager into AppContainer, orchestrator, and registry
- Persist session_metadata in Session model and pg_session_store
- Close all session terminals on session delete
- Webclient: handle terminal_output/terminal_closed WS events,
  display live terminal output in tool cards
- Update unit tests for new terminal actions

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Fix MCP tool registration at startup ...

Root cause: _on_mcp_server_connected was defined and registered AFTER
mcp_manager.load_all(), so startup connections never registered their
tools into the ToolRegistry.

Fix: explicitly iterate connected clients after build_default_registries()
creates tool_registry, and register their MCP tools before starting the
health-check loop. The callback stays for future reconnects.

Also adds diagnostic warning in build_tool_list when tools are missing.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Add debug logging to trace MCP tool registration / resolution ...

Logs added to:
- _on_mcp_server_connected: shows how many tools are registered per server
- build_tool_list: lists missing tool names and counts
- _resolve_tool: dumps tool_map keys when a lookup fails

This is diagnostic — will be removed once the root cause is found.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Add MCP health check loop, auto-reconnect, and system toast notifications ...

Backend:
- McpManager: keep all configured servers in the pool even if connect()
  fails at startup; health-check loop (30s) tries to reconnect dead servers
  and verifies live ones with list_tools()
- McpManager: set_on_server_connected callback re-registers tools when a
  dead server comes back online
- McpClient: add mark_disconnected() for silent drop detection
- McpStatusTool: skip list_tools() for disconnected servers
- Orchestrator: broadcast mcp_status_update to all WebSocket sessions
- New event type McpStatusUpdate with server_name, status, tool_count

Webclient:
- useWebSocket: handle mcp_status_update → toast.success/toast.error

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Fix stop button responsiveness and shutdown CancelledError ...

Agent loop (_execute_tools_with_sink):
- Poll stop_event every 1s while draining the event sink via asyncio.wait_for
- When stopped, cancel the tool task, yield a synthetic ToolEvent failure,
  append a cancellation message to session, yield StreamStopped, and return
- Pass stop_event into _execute_tools_with_sink call site

Subagent runner:
- Check stop_event at the start of each tool in turn_tool_calls loop
- Returns early with ("", False) when stopped mid-batch

McpManager.disconnect_all():
- Disconnect clients sequentially instead of asyncio.gather
- Handle asyncio.CancelledError per-client to avoid shutdown traceback

AppContainer.shutdown():
- Catch BaseException instead of Exception for MCP and DB cleanup

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Fix UnboundLocalError: create mcp_manager before build_default_registries ...

The previous commit passed mcp_manager to build_default_registries but
left the instantiation after the call, causing UnboundLocalError at
runtime. Move McpManager() creation before the registry build and
remove the now-obsolete post-hoc _mcp_manager patching loop.

392 passed, 1 skipped

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Eliminate cross-registry patching in registry.py via proper creation order ...

- Reorder build_default_registries so profiles and cp_registry are created
  BEFORE tools that need them
- All tools now receive full dependencies at construction time:
  ListToolsTool(profile_registry=profiles, mcp_manager=mcp_manager)
  SpawnAgentTool(backend_registry=backends, ...)
  ReloadToolsTool(cp_registry=cp_registry, mcp_manager=mcp_manager)
- Remove all post-creation patch lines (_profile_registry, _backend_registry, _cp_registry)
- Add mcp_manager parameter to build_default_registries
- Update create_container to pass mcp_manager

392 passed, 1 skipped

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Extract single shared Database pool, eliminate 4 duplicated pool creations ...

- Create navi/db.py::Database managing one asyncpg pool
- KvStore, PgSessionStore, MemoryStore, RecallScheduler now accept pool in constructor
- AppContainer holds Database, shutdown closes one pool instead of 4
- create_container creates one pool and passes it to all stores
- All tests updated to set _initialized=True on fakes to skip DDL

392 passed, 1 skipped

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Extract WebSocket business logic into AgentSessionOrchestrator ...

- Create navi/core/orchestrator.py with AgentSessionOrchestrator and SessionRun
- Orchestrator owns _runs, _busy_sessions, Agent creation, run_agent(), run_recall()
- Transport-agnostic: accepts notify callback from WebSocket handler
- WebSocket handler (websocket.py) now only does serialization/deserialization
- _fire_recall delegates to orchestrator.run_recall() instead of inline logic
- recall_scheduler_loop now accepts orchestrator parameter
- AppContainer gains .orchestrator field, created in create_container()
- deps.py: add get_orchestrator()
- Update integration tests for scheduler_loop and websocket unit tests

All 392 tests pass.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Fix PgSessionStore import in container.py ...

`PgSessionStore` lives in `navi.core.pg_session_store`, not `navi.core.session`.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Replace global lazy singletons with explicit AppContainer + lifespan ...

- Create navi/core/container.py with AppContainer dataclass and create_container()
- Rewrite navi/api/deps.py: remove module-level singletons, add _container global
  fallback + set_container(), use _resolve_container() for all getters
- Replace @app.on_event with @asynccontextmanager lifespan in main.py
- Update routes to use Depends(get_scheduler) instead of calling get_scheduler()
- Fix FastAPI body parsing bug: remove dataclass parameters from Depends getters
  (FastAPI was treating AppContainer sub-dependencies as Body params, forcing
  embed=True on all endpoint body params and causing 422 errors)
- Update websocket.py to use _resolve_container() instead of get_registries()
- Update integration test fixtures to build AppContainer and call set_container()
- Remove obsolete tests/unit/test_startup.py (tests removed _on_startup)
- Fix test_scheduler_loop.py fixture (get_registries no longer exists)

All 387 tests pass (excluding websocket hang tests).

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>