Health and status API
Health, status, and runtime feature endpoints.
Health & Status
Section titled “Health & Status”GET /health
Section titled “GET /health”No authentication required. Legacy health check retained for backward
compatibility. New integrations should prefer GET /health/live for liveness
and GET /health/ready for readiness.
Response
{ "status": "healthy", "uptime": 3600.5, "pid": 12345, "version": "0.124.5", "port": 3850, "agentsDir": "/home/user/.agents", "db": true, "dbWriter": { "queued": 0, "maxQueue": 64, "oldestWaitMs": null, "lastDurationMs": 2.1, "active": false, "oldestOperation": null, "rejected": 0, "cancelled": 0, "timedOut": 0, "lastQueueWaitMs": 0 }, "dbReader": { "activeLeases": 2, "maxConnections": 16, "queued": 0, "maxQueue": 64, "oldestWaitMs": null, "lastWaitMs": 0.4, "rejected": 0, "syncRejected": 0, "cancelled": 0, "timedOut": 0 }, "dbRuntime": { "operations": { "read": { "count": 10, "p50Ms": 1.2, "p95Ms": 4.5, "p99Ms": 5.1, "maxMs": 5.1 }, "write": { "count": 4, "p50Ms": 2.1, "p95Ms": 8.0, "p99Ms": 8.0, "maxMs": 8.0 } }, "queueWait": { "read": { "count": 10, "p50Ms": 0, "p95Ms": 12, "p99Ms": 12, "maxMs": 12 }, "write": { "count": 4, "p50Ms": 0, "p95Ms": 25, "p99Ms": 25, "maxMs": 25 } }, "cancelled": 0, "rejected": 0, "timedOut": 0, "failed": 0, "completed": 14, "eventLoopLag": { "count": 20, "p50Ms": 1.1, "p95Ms": 3.2, "p99Ms": 8.0, "maxMs": 8.0 }, "queue": { "readDepth": 0, "readMaxDepth": 64, "readOldestAgeMs": null, "readActiveLeases": 2, "writeDepth": 0, "writeMaxDepth": 64, "writeOldestAgeMs": null, "writeActive": false } }, "shuttingDown": false, "updateAvailable": false, "pendingRestart": false, "resources": { "total": -1, "memoryMd": 0, "sockets": 0, "inotify": 0, "pipes": 0, "db": 0, "other": 0, "rss": 169, "heapUsed": 106, "physicalFootprint": 2867, "peakPhysicalFootprint": 3584 }}dbReader reports active read leases, FIFO waiter depth and age, bounded
admission limits, and cancellation/rejection counts. syncRejected counts
legacy synchronous read attempts rejected at the hard connection cap. The
dbRuntime.queue snapshot reports current read/write depth, queue age, and
active leases; eventLoopLag is the bounded independent event-loop sample.
Memory resource values are MiB. On macOS, physicalFootprint and
peakPhysicalFootprint come from proc_pid_rusage and include compressed
and driver-backed memory that RSS can miss. They are null on platforms
where this metric is unavailable.
GET /health/live
Section titled “GET /health/live”No authentication required. Cheap liveness probe: reports that the daemon process is up and serving HTTP. It never touches the database or any subsystem, and always returns 200 while the process is alive.
Response (always 200)
{ "status": "healthy", "uptime": 3600.5, "pid": 12345, "version": "0.124.5", "port": 3850, "shuttingDown": false, "eventLoop": { "status": "healthy", "lagP95Ms": 3, "lagP99Ms": 8 }}status is "healthy", or "shutting_down" once shutdown has begun. The
eventLoop block is diagnostic only and does not make this independent probe
depend on database availability. It reports bounded event-loop lag samples and
may report "degraded" while the process still returns 200.
The @signet/daemon service-management API uses this liveness endpoint for
getDaemonStatus() with a 1.2-second deadline. Its returned status is
"healthy" when the probe succeeds, "degraded" when the service manager
reports a running process but the probe fails or reports shutdown, and
"unavailable" when no service is running.
GET /health/ready
Section titled “GET /health/ready”No authentication required. Readiness probe: reports whether the daemon can
actually serve work. Returns 200 only when every gate passes, otherwise 503
with a human-readable reasons list. Gates:
db— a readonly database connection answersSELECT 1.migrations— no pending database migrations.embedding— the configured embedding provider is reachable. Passes withnote: "disabled"when the provider is intentionally"none".inference— the extraction route is not fullyblocked; adegradedroute still passes readiness.queue— durable queue depth, dead-letter rate, and oldest pending job age are within thresholds. Becomes{ "error": "database unavailable" }when the database check fails.
Response (200 when ready)
{ "status": "ready", "version": "0.124.5", "shuttingDown": false, "checks": { "db": true, "dbReader": { "activeLeases": 1, "maxConnections": 16, "queued": 0, "maxQueue": 64, "oldestWaitMs": null, "lastWaitMs": 0.4, "rejected": 0, "syncRejected": 0, "cancelled": 0, "timedOut": 0 }, "dbRuntime": { "queue": { "readDepth": 0, "readMaxDepth": 64, "readOldestAgeMs": null, "readActiveLeases": 1, "writeDepth": 0, "writeMaxDepth": 64, "writeOldestAgeMs": null, "writeActive": false } }, "migrations": true, "embedding": { "provider": "ollama", "available": true, "checkedAt": "2026-02-21T10:00:00.000Z" }, "inference": { "status": "active", "configured": "ollama", "effective": "ollama", "reason": null }, "queue": { "score": 1, "status": "healthy", "depth": 0, "oldestAgeSec": 0, "deadRate": 0, "leaseAnomalies": 0 } }, "reasons": []}Response (503 when not ready) — same shape, with status: "not_ready"
and one entry per failing gate in reasons:
{ "status": "not_ready", "version": "0.124.5", "shuttingDown": false, "checks": { "db": true, "migrations": false, "embedding": { "provider": "none", "available": true, "note": "disabled" }, "inference": { "status": "active", "configured": "ollama", "effective": "ollama", "reason": null }, "queue": { "score": 1, "status": "healthy", "depth": 0, "oldestAgeSec": 0, "deadRate": 0, "leaseAnomalies": 0 } }, "reasons": ["pending migrations"]}signet doctor consumes this probe as its runtime contract: healthy means
/health/ready returned ready and local installation checks passed (ok: true,
exit 0); degraded means the daemon is reachable but returned not_ready
(ok: false, exit 2); unavailable means the required daemon or local
installation is unavailable or invalid (ok: false, exit 1). A queue-caused
periodic scheduler decision deferred for queue pressure includes an explicit
Automatic Dreaming deferral finding with the queue age and last error. A queue
readiness failure alone does not claim that Dreaming was deferred by the queue:
the scheduler can instead defer for higher-priority system pressure.
GET /api/status
Section titled “GET /api/status”Full daemon status including pipeline config, embedding provider, and a composite health score derived from diagnostics. Extraction provider runtime resolution persists startup degradation so operators can detect silent fallback or hard-blocked extraction after boot.
Response
{ "status": "running", "version": "0.124.5", "pid": 12345, "uptime": 3600.5, "startedAt": "2026-02-21T10:00:00.000Z", "port": 3850, "host": "127.0.0.1", "bindHost": "127.0.0.1", "networkMode": "localhost", "agentId": "default", "agentsDir": "/home/user/.agents", "memoryDb": true, "resources": { "rss": 169, "heapUsed": 106, "physicalFootprint": 2867, "peakPhysicalFootprint": 3584 }, "pipelineV2": { "enabled": true, "paused": false, "shadowMode": false, "mutationsFrozen": false, "graph": { "enabled": true }, "autonomous": { "enabled": true, "allowUpdateDelete": true }, "extraction": { "provider": "llama-cpp", "model": "qwen3:4b" } }, "pipeline": { "queue": { "memory": { "pending": 0, "leased": 0, "completed": 0, "failed": 0, "dead": 0, "oldestAgeSec": 0, "oldestDeadAgeSec": 0, "lastError": null, "completeness": "exact" }, "summary": { "pending": 0, "leased": 0, "completed": 0, "failed": 0, "dead": 0, "oldestAgeSec": 0, "oldestDeadAgeSec": 0, "lastError": null, "completeness": "exact" } } }, "providerResolution": { "extraction": { "configured": "llama-cpp", "resolved": "llama-cpp", "effective": "llama-cpp", "fallbackProvider": "llama-cpp", "status": "active", "degraded": false, "fallbackApplied": false, "reason": null, "blockedBy": [], "since": null, "enabled": true, "paused": false, "workerRunning": false, "ready": true, "blockedReason": null } }, "logging": { "logDir": "/home/user/.agents/.daemon/logs", "logFile": "/home/user/.agents/.daemon/logs/signet-2026-04-29.log" }, "activeSessions": 1, "bypassedSessions": 1, "agentCreatedAt": "2026-02-21T10:00:00.000Z", "transcripts": { "capture": { "pending": 0, "processing": 0, "failed": 0, "dead": 0 } }, "health": { "score": 0.97, "status": "healthy" }, "update": { "currentVersion": "0.124.5", "latestVersion": null, "updateAvailable": false, "pendingRestart": null, "autoInstall": false, "checkInterval": 21600, "lastCheckAt": null, "lastError": null, "timerActive": true }, "embedding": { "provider": "ollama", "model": "nomic-embed-text", "available": true, "usage": { "total": { "requests": 2084, "tokens": 812345 }, "today": { "requests": 12, "tokens": 4567 }, "bySource": [ { "source": "artifact-index", "requests": 2050, "tokens": 800000 }, { "source": "memory-capture", "requests": 30, "tokens": 12000 }, { "source": "recall", "requests": 4, "tokens": 345 } ], "byProvider": [ { "provider": "ollama", "requests": 2080, "tokens": 812000 }, { "provider": "llama-cpp", "requests": 4, "tokens": 345 } ] } }}The embedding.usage block reports embedding token consumption recorded at
the shared embedding-fetch boundary (migration 108). Counts come from the
real tokenizer (countTokens) applied to the text actually sent to the
provider — never provider-reported usage, since Ollama’s /api/embeddings
returns none and the native ONNX path reports none either. requests counts
successful embedding fetches; tokens sums their input token counts.
bySource breaks totals down by memory-capture, artifact-index,
recall, dreaming, and other; byProvider breaks them down by the
provider that actually served (ollama, llama-cpp, native, openai —
the native fallback chain reports the real serving provider). The block is
omitted when the table does not exist (pre-migration database).
The bypassedSessions field reports how many active sessions currently have
bypass enabled (see Sessions and hooks API).
providerResolution.extraction is the canonical workload-state object. Its
configured, resolved, and effective labels describe provider selection;
they do not imply that jobs are being serviced. Use enabled, paused, and
ready to determine whether extraction is actually available for work. The
standalone extraction worker was retired under the Dreaming cutover (#946),
so workerRunning is always false and ready reflects route resolution
alone (active or degraded). blockedReason is populated only for a
blocked route.
Monitor status for degraded or blocked states when the configured
extraction provider is unavailable or routed to a fallback target.
When extraction is blocked, providerResolution.extraction.blockedBy contains
the first routing candidate’s policy and runtime gate reasons in evaluation
order. The array is empty for non-blocked states.
pipeline.queue exposes per-queue counts (memory / summary).
pipeline.dreaming records the latest periodic Dreaming scheduler decision.
It is null before the worker starts. A deferred result with
reason: "queue_pressure" means the scheduler itself yielded that sweep to
queue health. It does not infer a queue deferral from another readiness gate.
The retired worker’s load/overload telemetry is no longer reported.
transcripts.capture exposes compact durable transcript-capture queue counts;
use GET /api/diagnostics/transcripts for detailed artifact/audit diagnostics.
Use GET /api/inference/status for the shared inference control plane status.
GET /api/diagnostics/workloads
Section titled “GET /api/diagnostics/workloads”Returns bounded in-flight workload pressure for one resolved agent. The
inference snapshot covers background routed work and Pi agent sessions; the
MCP snapshot covers stateless Streamable HTTP requests; the provider semaphore
snapshot covers global LLM permit occupancy; and the Dreaming snapshot covers
running passes plus pending attention. Counts are held in memory except for
the Dreaming snapshot, which reads scoped durable rows. Ages are calculated
from admission or persisted start/creation time, so this endpoint does not
scan general-purpose durable queue tables. agentId may be supplied as a
query parameter or x-signet-agent-id header; scoped deployments reject
requests for another agent.
Response
{ "agentId": "default", "inference": { "active": 1, "agentSessions": 1, "oldestAgeMs": 3200, "oldestAgentSessionAgeMs": 3200, "byOperation": { "memory_extraction": 1 } }, "mcp": { "inFlight": 0, "oldestAgeMs": null, "maxInFlight": 8 }, "pi": { "active": 1, "oldestAgeMs": 3200 }, "providerSemaphore": { "running": 1, "pending": 0, "limit": 2, "oldestPendingAgeMs": null }, "dreaming": { "activePasses": 0, "oldestPassAgeMs": null, "pendingAttention": 0, "oldestAttentionAgeMs": null }}GET /api/diagnostics/queue
Section titled “GET /api/diagnostics/queue”Per-queue counts (memory / summary), oldest-dead job
references, and threshold metadata. Counts are bounded at 1,000 rows per
status; completeness is truncated when a counter is capped and unknown
when the queue schema cannot be observed. Backend path uses the same shared
threshold constants that GET /api/status and /health/ready consume.
Admin permission required.
Response
{ "timestamp": "2026-07-19T00:00:00.000Z", "queues": { "memory": { "pending": 0, "leased": 0, "completed": 1, "failed": 0, "dead": 0, "oldestAgeSec": 0, "oldestDeadAgeSec": 0, "lastError": null, "completeness": "exact" }, "summary": { "pending": 0, "leased": 0, "completed": 0, "failed": 0, "dead": 0, "oldestAgeSec": 0, "oldestDeadAgeSec": 0, "lastError": null, "completeness": "exact" } }, "oldestDeadSummaryJob": null, "oldestDeadMemoryJob": { "...": "..." }, "thresholds": { "memoryDeadWarn": 50, "memoryDeadFail": 500, "memoryOldestPendingWarnSec": 300, "memoryOldestPendingFailSec": 1800, "summaryDeadWarn": 50, "summaryDeadFail": 500, "summaryOldestPendingWarnSec": 300, "summaryOldestPendingFailSec": 1800, "summaryOldestDeadWarnSec": 86400 }}Counts default to 0 and null if a table does not exist on the running
database. Historical summary_jobs rows are retired by migration 117 and are
not reported as live work. The summary queue and summary thresholds remain
zero/compatibility fields for older clients.
POST /api/diagnostics/queue/repair
Section titled “POST /api/diagnostics/queue/repair”Dispatches a queue repair action. Admin permission required. The body
shape covers requeue (extending requeueDeadJobs), cancel
(audit-preserving soft cancel into job_cancellations), and prune
(archive-preserving hard delete into job_archive).
Request body
{ "action": "cancel", "dryRun": true, "tables": ["memory"], "olderThanMs": 2592000000, "errorPattern": "timeout"}action— one ofrequeue,cancel,prune.dryRun— boolean; defaults totrue(safe preview).ids— optional array of row ids; bypasses filter for max precision.tables— optional array containingmemory(default:memory). The historicalsummarytarget is retired and returns HTTP410.olderThanMs— only match rows whosecreated_atis older thannow - olderThanMs.errorPattern— optionalLIKE %pattern%over theerrorcolumn.retentionMs— optional override forprune’s default 90-day window.maxBatch— optional hard cap on rows touched (default: 50 for requeue; 1000 for cancel/prune).
Response
{ "action": "cancelObsoleteJobs", "success": true, "affected": 0, "message": "dry-run: 1667 job(s) match cancel filter; preview shows 100", "preview": ["memory_jobs:abc", "memory_jobs:def"], "totalMatching": 1667}Both queue endpoints require the admin permission in authenticated modes.
When the policy gate denies an action, the response carries success: false
and HTTP 429 (cooldown active / hourly budget exhausted / agents without
autonomous.enabled). Wrong action values or malformed JSON return 400.
Cancel and prune apply requests require migrations 089 and 090; neither daemon
creates audit tables from the request path, and a missing table is reported as
a migration error.
GET /api/features
Section titled “GET /api/features”Returns all runtime feature flags.
Response
{ "featureName": true, "anotherFeature": false}GET /api/mode
Section titled “GET /api/mode”Environment probe (issue #1001). Deliberately lightweight and unauthenticated — the dashboard uses it to distinguish “talking to a real daemon” (any hostname: localhost, Tailscale, .local, tunnel, LAN IP) from the marketing site or the cloud app. If this endpoint responds, there is a real daemon behind the URL.
Response
{ "mode": "local", "requiresAuth": false}mode: the daemon’s auth mode (local,team, orhybrid).requiresAuth:falseinlocalmode,trueinteamandhybridmodes. The endpoint itself is always unauthenticated and carries no data beyond this documented shape. Authenticated data endpoints use Bearer tokens only, with no cookies.