Outpost Admin API Reference

The Outpost Admin API is a separate FastAPI application that listens on port 8301 (local-only). It provides emergency override capabilities, local status inspection, and operational controls without requiring connectivity to the Arbitex platform.

Air-gap compatible. All admin API calls are local. No data is forwarded to the platform during admin operations.

---

Authentication

All endpoints require a Bearer token in the Authorization header:

Authorization: Bearer <admin-api-key>

Token resolution order:

1. admin_users[].api_key from the loaded policy bundle (HMAC compare — constant time)
2. OUTPOST_EMERGENCY_ADMIN_KEY environment variable as a fallback
3. A valid HS256/EdDSA JWT obtained from POST /admin/api/auth/login

Brute-force protection: 5 failed authentication attempts per IP within a 15-minute sliding window triggers HTTP 429. The window resets automatically.

CORS: The admin API only accepts requests from http://localhost:8301 and http://127.0.0.1:8301. All other origins are rejected.

Replay protection: When REPLAY_PROTECTION_ENABLED=true, state-changing endpoints (POST/PUT/DELETE) require an X-Request-ID header. Duplicate X-Request-ID values return 409 Conflict.

---

Master Endpoint Summary

Method	Path	Section
POST	/admin/api/auth/login	Authentication & Authorization
POST	/admin/api/auth/refresh	Authentication & Authorization
POST	/admin/api/auth/revoke	Authentication & Authorization
GET	/admin/api/auth/revocation-list	Authentication & Authorization
POST	/admin/api/auth/revocation-list/clear	Authentication & Authorization
GET	/admin/api/sessions	Authentication & Authorization
DELETE	/admin/api/sessions/{session_id}	Authentication & Authorization
GET	/admin/api/auth-events	Authentication & Authorization
GET	/admin/api/status	Status & Health
GET	/admin/api/sync-status	Status & Health
GET	/admin/api/airgap-config	Status & Health
GET	/admin/api/health	Status & Health
GET	/admin/api/health/summary	Status & Health
GET	/admin/api/health/components	Status & Health
GET	/admin/api/health/cluster	Status & Health
GET	/admin/api/connection-health	Status & Health
GET	/admin/api/connection-health/history	Status & Health
GET	/admin/api/compat	Status & Health
GET	/admin/api/rules	Policy & DLP
POST	/admin/api/rules/{rule_id}/toggle	Policy & DLP
GET	/admin/api/rulesets	Policy & DLP
POST	/admin/api/rulesets/{ruleset_id}/toggle	Policy & DLP
POST	/admin/api/policy/simulate	Policy & DLP
GET	/admin/api/policy/verify	Policy & DLP
POST	/admin/api/policy/reload	Policy & DLP
GET	/admin/api/custom-patterns	Policy & DLP
POST	/admin/api/custom-patterns	Policy & DLP
DELETE	/admin/api/custom-patterns/{name}	Policy & DLP
GET	/admin/api/providers	Provider Management
POST	/admin/api/providers/{provider}/disable	Provider Management
POST	/admin/api/providers/{provider}/enable	Provider Management
GET	/admin/api/credint/status	Credential Intelligence
POST	/admin/api/credint/reload	Credential Intelligence
POST	/admin/api/credint/kanon	Credential Intelligence
GET	/admin/api/usage	Budget & Usage
GET	/admin/api/budget/status	Budget & Usage
GET	/admin/api/budget/export.csv	Budget & Usage
GET	/admin/api/cert/status	Certificate Management
POST	/admin/api/cert/renew	Certificate Management
GET	/admin/api/certs/status	Certificate Management
POST	/admin/api/certs/rotate	Certificate Management
GET	/admin/api/audit-buffer	Audit & SIEM
GET	/admin/audit/recent	Audit & SIEM
GET	/admin/audit/stats	Audit & SIEM
GET	/admin/audit/search	Audit & SIEM
GET	/admin/audit/verify	Audit & SIEM
GET	/admin/api/audit/verify	Audit & SIEM
GET	/admin/api/audit-sync/checkpoint	Audit & SIEM
GET	/admin/api/siem/status	Audit & SIEM
GET	/admin/audit-queue/status	Audit & SIEM
POST	/admin/audit-queue/flush	Audit & SIEM
DELETE	/admin/audit-queue/purge	Audit & SIEM
GET	/admin/api/config/export	Configuration
GET	/admin/config/export	Configuration
POST	/admin/api/config/backup	Configuration
GET	/admin/api/config/backup/list	Configuration
POST	/admin/api/config/restore	Configuration
POST	/admin/config/diff	Configuration
GET	/admin/config/validate	Configuration
GET	/admin/api/config/reload-status	Configuration
POST	/admin/api/config/reload	Configuration
POST	/admin/api/config/apply	Configuration
GET	/admin/api/config/reload-history	Configuration
GET	/admin/api/orgs	Multi-Org
GET	/admin/api/orgs/{org_id}/policy-status	Multi-Org
GET	/admin/api/cache/stats	Multi-Org
POST	/admin/api/cache/clear	Multi-Org
POST	/admin/api/emergency-kill	Kill Switch & Drain
GET	/admin/api/drain/status	Kill Switch & Drain
GET	/admin/api/prompt-holds	Kill Switch & Drain
POST	/admin/api/prompt-holds/{hold_id}/approve	Kill Switch & Drain
POST	/admin/api/prompt-holds/{hold_id}/deny	Kill Switch & Drain
GET	/admin/api/prompt-holds/events	Kill Switch & Drain
POST	/admin/api/prompt-holds/bulk-approve	Kill Switch & Drain
POST	/admin/api/routing-override	Kill Switch & Drain
GET	/admin/api/plugins	Plugins & Webhooks
POST	/admin/api/plugins/{name}/enable	Plugins & Webhooks
POST	/admin/api/plugins/{name}/disable	Plugins & Webhooks
GET	/admin/api/webhook-emitter/stats	Plugins & Webhooks
GET	/admin/api/updates/status	Software Updates
POST	/admin/api/updates/check	Software Updates
POST	/admin/api/updates/download	Software Updates
GET	/admin/api/upgrade/status	Software Updates
POST	/admin/api/upgrade/run	Software Updates
POST	/admin/api/upgrade/rollback	Software Updates
GET	/admin/api/upgrade/history	Software Updates
GET	/admin/diagnostics	Operational Tools
GET	/admin/api/metrics-summary	Operational Tools
GET	/admin/api/metrics/connections	Operational Tools
GET	/admin/api/circuit-breakers	Operational Tools
GET	/admin/api/bandwidth	Operational Tools
GET	/admin/api/benchmarks	Operational Tools
GET	/admin/api/log-export/status	Operational Tools
GET	/admin/api/otel/status	Operational Tools
GET	/admin/api/pushgateway/status	Operational Tools
GET	/admin/api/ip-allowlist/status	Operational Tools
GET	/admin/api/body-hash/config	Operational Tools
GET	/admin/api/replay-protection/stats	Operational Tools

---

1. Authentication & Authorization

This section covers JWT-based authentication, session management, and authentication event retrieval. Static Bearer token auth is described in the Authentication section above.

POST /admin/api/auth/login

Issue a signed JWT for admin API access.

POST /admin/api/auth/login
Content-Type: application/json

Request body

{"api_key": "your-admin-api-key"}

Response 200 OK

{
  "token": "eyJhbGciOiJFZERTQSIsInR5cCI6IkpXVCJ9...",
  "expires_at": "2026-03-16T12:00:00Z"
}

Use the returned token in the Authorization: Bearer <token> header for subsequent requests.

---

POST /admin/api/auth/refresh

Refresh a JWT that is within 5 minutes of expiry. Pass the current (non-expired) token as the Authorization header.

curl -s -X POST "http://localhost:8301/admin/api/auth/refresh" \
  -H "Authorization: Bearer $CURRENT_JWT"

Response 200 OK

{
  "token": "eyJhbGciOiJFZERTQSIsInR5cCI6IkpXVCJ9...",
  "expires_at": "2026-03-16T13:00:00Z"
}

Returns 400 Bad Request if the token is not within 5 minutes of expiry. Returns 401 Unauthorized if the token is already expired.

---

POST /admin/api/auth/revoke

Revoke a JWT by adding its JTI (JWT ID) to the in-memory revocation list. The revoked token is rejected on all subsequent requests until the revocation list is cleared or the outpost restarts.

curl -s -X POST "http://localhost:8301/admin/api/auth/revoke" \
  -H "Authorization: Bearer $JWT_TO_REVOKE"

Response 200 OK

Caution

The revocation list is in-memory. Revocations are lost on outpost restart. For emergency access revocation that must persist across restarts, rotate the OUTPOST_ADMIN_KEY instead.

---

GET /admin/api/auth/revocation-list

Return the count of revoked JTIs and the timestamp of the last list clear.

Response 200 OK

{
  "revoked_count": 3,
  "last_cleared": "2026-03-15T08:00:00Z"
}

---

POST /admin/api/auth/revocation-list/clear

Clear the entire revocation list. All previously revoked tokens become valid again.

Dangerous operation

Clearing the revocation list re-validates all previously revoked JWTs. Use only after confirming that all revoked tokens have passed their expiry time, or after rotating OUTPOST_ADMIN_KEY.

Response 200 OK

---

GET /admin/api/sessions

List all currently active admin sessions. Expired sessions are cleaned up before the response is returned.

curl -s http://outpost.internal:8301/admin/api/sessions \
  -H "Authorization: Bearer $ADMIN_API_KEY"

Response 200 OK

{
  "sessions": [
    {
      "id": "sess_7a3b9f1c2e4d5a6b",
      "username": "ops-admin",
      "ip": "10.0.1.42",
      "user_agent": "Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36",
      "created_at": "2026-03-15T11:00:00Z",
      "last_activity": "2026-03-15T11:28:45Z"
    }
  ],
  "count": 1
}

Field	Type	Description
id	string	Session identifier
username	string	Admin username associated with the session
ip	string	Source IP address
user_agent	string	HTTP User-Agent string from the login request
created_at	datetime	Session creation timestamp
last_activity	datetime	Most recent request timestamp

---

DELETE /admin/api/sessions/{session_id}

Force-logout an active admin session by ID. Writes an auth.session.revoked audit event.

Path parameters

Parameter	Type	Description
session_id	string	Session ID from GET /admin/api/sessions

curl -s -X DELETE \
  http://outpost.internal:8301/admin/api/sessions/sess_a1b2c3d4 \
  -H "Authorization: Bearer $ADMIN_API_KEY"

Response 200 OK

{
  "revoked": "sess_7a3b9f1c2e4d5a6b",
  "username": "ops-admin"
}

Returns 404 Not Found if the session is not found or already expired.

---

GET /admin/api/auth-events

Return the most recent authentication events (auth.*) from the local audit queue. Limited to the last 100 entries.

Query parameters

Parameter	Type	Default	Description
limit	int	20	Maximum events to return (capped at 100)

Response 200 OK

{
  "events": [
    {
      "event": "auth.login.success",
      "username": "ops-alice",
      "ip": "10.0.1.42",
      "timestamp": "2026-03-16T14:07:33Z"
    },
    {
      "event": "auth.session.revoked",
      "revoked_session_id": "sess_old123",
      "revoked_by": "ops-alice",
      "timestamp": "2026-03-16T14:05:00Z"
    }
  ],
  "count": 2
}

Returns {"events": [], "count": 0} if the audit queue is not initialized.

---

2. Status & Health

These endpoints provide operational visibility into the outpost state, health checks at varying granularity levels, and real-time connection metrics.

GET /admin/api/status

Returns the current outpost operational status: policy version, active override count, uptime, and emergency kill state.

GET /admin/api/status
Authorization: Bearer <token>

Response 200 OK

{
  "outpost_id": "prod-outpost-01",
  "policy_version": "v2026.03.12-4",
  "uptime_seconds": 86400,
  "active_override_count": 2,
  "emergency_kill": false,
  "last_override_modified": "2026-03-12T08:00:00Z",
  "routing_override": null,
  "update_available": true,
  "latest_version": "0.19.2"
}

Field	Description
active_override_count	Sum of all active provider, rule, ruleset, and routing overrides + kill switch
emergency_kill	true if the emergency kill switch is active (all requests blocked)
routing_override	Provider name if routing is pinned; null if using policy-default routing
update_available	true if latest_version differs from the running outpost version

---

GET /admin/api/sync-status

Returns policy sync health: last sync timestamp, ETag, bundle version, heartbeat status, and TLS certificate expiry.

GET /admin/api/sync-status
Authorization: Bearer <token>

Response 200 OK

{
  "last_sync_at": "2026-03-12T08:52:00Z",
  "etag": "\"abc123\"",
  "bundle_version": "v2026.03.12-4",
  "latest_version": "0.19.2",
  "heartbeat": {
    "last_sent_at": "2026-03-12T08:55:00Z",
    "consecutive_failures": 0
  },
  "cert_expiry_days": 82.5
}

cert_expiry_days is null if no certificate path is configured. A value below 30 should trigger a certificate renewal.

---

GET /admin/api/airgap-config

Returns the read-only air-gap configuration from OutpostSettings. Useful for verifying that air-gap mode is configured correctly.

GET /admin/api/airgap-config
Authorization: Bearer <token>

Response 200 OK

{
  "airgap_enabled": true,
  "policy_path": "/etc/arbitex/policy.json",
  "model_path": "/etc/arbitex/models/",
  "siem_output": "file:///var/log/arbitex/siem.jsonl",
  "hmac_enabled": true,
  "outpost_id": "prod-outpost-01"
}

hmac_enabled is false only when OUTPOST_INSECURE_SKIP_HMAC=true is set (development only).

---

GET /admin/api/health

Config validation health check — runs all startup validation checks and reports pass/fail/warn/skip per check. Includes degradation mode status and TLS configuration summary.

curl -s http://outpost.internal:8301/admin/api/health \
  -H "Authorization: Bearer $ADMIN_API_KEY"

Response 200 OK

{
  "overall": "pass",
  "passed": 14,
  "failed": 0,
  "warned": 1,
  "skipped": 2,
  "checks": [
    {
      "name": "admin_api_key",
      "status": "pass",
      "message": "ADMIN_API_KEY is set"
    },
    {
      "name": "tls_cert",
      "status": "warn",
      "message": "Certificate expires in 28 days"
    }
  ],
  "degradation_mode": false,
  "degraded_since": null,
  "tls_config": {
    "min_version": "1.2",
    "max_version": "1.3",
    "verify_client": true
  }
}

overall is "fail" if any check fails, "warn" if any check warns, otherwise "pass".

---

GET /admin/api/health/summary

Condensed health response for fleet monitoring tools and dashboards.

curl -s http://outpost.internal:8301/admin/api/health/summary \
  -H "Authorization: Bearer $ADMIN_API_KEY"

Response 200 OK

{
  "status": "healthy",
  "uptime_seconds": 86412.3,
  "version": "0.14.2",
  "mode": "single",
  "active_orgs": 0,
  "inflight_requests": 3,
  "cache_hit_rate": 94.7,
  "circuit_breaker_open": false
}

Field	Description
status	healthy / degraded / unhealthy — aggregate of proxy, DLP, and policy sync health
mode	single (single-org) or multi (multi-org mode)
active_orgs	Count of active orgs (multi-org mode only; 0 in single-org mode)
inflight_requests	Current in-flight DLP scan count
cache_hit_rate	Policy cache hit rate percentage

---

GET /admin/api/health/components

Component-level health breakdown with latency and detail fields. Suitable for dashboards and alerting.

curl -s http://outpost.internal:8301/admin/api/health/components \
  -H "Authorization: Bearer $ADMIN_API_KEY"

Response 200 OK (or 503 Service Unavailable when overall is unhealthy)

{
  "status": "healthy",
  "components": {
    "platform_connection": {
      "status": "healthy",
      "latency_ms": 42.3,
      "last_check": "2026-03-16T14:07:10Z"
    },
    "model_service": {
      "status": "healthy",
      "last_check": "2026-03-16T14:07:15Z",
      "dlp_enabled": true,
      "ner_enabled": false
    },
    "cert": {
      "status": "healthy",
      "days_remaining": 89
    },
    "config": {
      "status": "healthy",
      "reload_count": 3
    }
  }
}

Component statuses: healthy / degraded / unhealthy. Platform connection is degraded when last sync is > 5 minutes old and unhealthy when > 1 hour old or the circuit breaker is open.

---

GET /admin/api/health/cluster

Aggregate health status across all outpost peers in the cluster. Queries peers configured via CLUSTER_PEERS. Returns 207 when any peer is degraded, 503 when any peer is unhealthy.

curl -s http://outpost.internal:8301/admin/api/health/cluster \
  -H "Authorization: Bearer $ADMIN_API_KEY"

Response 200 OK (all healthy)

{
  "status": "healthy",
  "peers": [
    {"url": "self (outpost-prod-01)", "status": "healthy"},
    {"url": "http://outpost-prod-02:8301", "status": "healthy"}
  ],
  "total": 2,
  "healthy": 2,
  "degraded": 0,
  "unhealthy": 0
}

HTTP Status	Meaning
200	All peers healthy
207	At least one peer degraded
503	At least one peer unhealthy or unreachable

---

GET /admin/api/connection-health

Real-time connection health metrics from the connection health monitor: latency percentiles, packet loss, and jitter.

curl -s http://outpost.internal:8301/admin/api/connection-health \
  -H "Authorization: Bearer $ADMIN_API_KEY"

Response 200 OK

{
  "status": "healthy",
  "enabled": true,
  "total_probes": 1440,
  "avg_latency_ms": 18.4,
  "p95_latency_ms": 42.1,
  "p99_latency_ms": 87.3,
  "packet_loss_pct": 0.0,
  "jitter_ms": 3.2,
  "probe_count": 60,
  "success_count": 60
}

Returns {"status": "unknown", "enabled": false, ...} when the connection health monitor is not configured.

---

GET /admin/api/connection-health/history

Return the raw probe history for the connection health monitor. Each entry represents one probe cycle against each AI provider.

curl -s "http://localhost:8301/admin/api/connection-health/history" \
  -H "Authorization: Bearer $ARBITEX_ADMIN_KEY"

Response 200 OK

[
  {
    "probed_at": "2026-03-16T09:00:00Z",
    "results": [
      {"provider": "openai", "latency_ms": 142, "status": "ok"},
      {"provider": "anthropic", "latency_ms": 198, "status": "ok"}
    ]
  },
  {
    "probed_at": "2026-03-16T08:55:00Z",
    "results": [
      {"provider": "openai", "latency_ms": 3001, "status": "timeout"},
      {"provider": "anthropic", "latency_ms": 201, "status": "ok"}
    ]
  }
]

---

GET /admin/api/compat

Run version compatibility checks across all outpost components and return a structured matrix summary.

curl -s http://outpost.internal:8301/admin/api/compat \
  -H "Authorization: Bearer $ADMIN_API_KEY"

Response 200 OK

{
  "overall": "pass",
  "components": [
    {
      "name": "platform_api",
      "detected_version": "1.14.0",
      "min_version": "1.12.0",
      "recommended_version": "1.14.0",
      "status": "pass",
      "message": "Platform API version compatible"
    },
    {
      "name": "outpost_agent",
      "detected_version": "0.14.2",
      "min_version": "0.12.0",
      "recommended_version": "0.14.x",
      "status": "pass",
      "message": "Outpost version compatible"
    }
  ]
}

Check statuses: pass / warn / fail / skip.

---

3. Policy & DLP

These endpoints manage DLP rules, rulesets, custom redaction patterns, policy simulation, policy verification, and air-gap policy reload.

GET /admin/api/rules

Lists all DLP rules from the loaded policy bundle with their enabled/disabled state.

GET /admin/api/rules
Authorization: Bearer <token>

Response 200 OK

{
  "rules": [
    {
      "id": "pii-ssn",
      "name": "SSN Detection",
      "tier": 1,
      "action": "block",
      "enabled": true
    },
    {
      "id": "pii-ccn",
      "name": "Credit Card Number",
      "tier": 1,
      "action": "redact",
      "enabled": false
    }
  ]
}

---

POST /admin/api/rules/{rule_id}/toggle

Enables or disables a specific DLP rule. The change takes effect immediately for all subsequent requests.

POST /admin/api/rules/{rule_id}/toggle
Authorization: Bearer <token>
Content-Type: application/json

{
  "enabled": false
}

Response 200 OK

{
  "rule_id": "pii-ccn",
  "enabled": false
}

---

GET /admin/api/rulesets

Lists all DLP compliance rulesets (compliance_bundles in the policy bundle) with their enabled/disabled state.

GET /admin/api/rulesets
Authorization: Bearer <token>

Response 200 OK

{
  "rulesets": [
    {
      "id": "hipaa",
      "name": "HIPAA PHI",
      "enabled": true
    },
    {
      "id": "pci-dss",
      "name": "PCI DSS",
      "enabled": false
    }
  ]
}

---

POST /admin/api/rulesets/{ruleset_id}/toggle

Enables or disables an entire compliance ruleset. Disabling a ruleset suppresses all rules within it.

POST /admin/api/rulesets/{ruleset_id}/toggle
Authorization: Bearer <token>
Content-Type: application/json

{
  "enabled": false
}

Response 200 OK

{
  "ruleset_id": "pci-dss",
  "enabled": false
}

---

POST /admin/api/policy/simulate

Runs a synthetic prompt through the local DLP pipeline and returns a full evaluation trace. Read-only — no request is forwarded to any LLM, no usage is recorded, and the prompt text is not written to the audit log (only its length is logged).

POST /admin/api/policy/simulate
Authorization: Bearer <token>
Content-Type: application/json

{
  "prompt": "My SSN is 123-45-6789, please help me...",
  "provider": "anthropic",
  "model": "claude-sonnet-4-6",
  "org_id": "b3e2a1c0-..."
}

Field	Required	Description
prompt	Yes	Text to evaluate
provider	No	Provider to use for rule-provider matching
model	No	Model ID for model-specific rules
org_id	No	Defaults to the outpost’s configured org_id

Returns 503 if no policy bundle is loaded.

Response — see Policy Simulator guide for full response schema.

---

GET /admin/api/policy/verify

Return the last policy rule verification result from the most recent successful policy sync. Populated automatically after each HMAC-verified policy sync. Returns 404 if no bundle has been synced and verified in the current session.

Response 200 OK

{
  "bundle_version": "v2026.03.16-1",
  "verified_at": "2026-03-16T14:00:00Z",
  "rules_count": 47,
  "packs_count": 8,
  "hmac_valid": true,
  "schema_version": "3",
  "warnings": []
}

Response 404 Not Found (no bundle synced yet)

{"detail": "No policy verification result available yet"}

Tip

Use this endpoint after a forced policy reload (POST /admin/api/policy/reload) to confirm the new bundle loaded and verified correctly.

---

POST /admin/api/policy/reload

Reload the policy bundle from the local filesystem. Only available when OUTPOST_AIRGAP=true. Reads policy_bundle.json from AIRGAP_POLICY_PATH, validates structure, and atomically replaces the active bundle in memory.

curl -s -X POST http://outpost.internal:8301/admin/api/policy/reload \
  -H "Authorization: Bearer $ADMIN_API_KEY"

Response 200 OK

{
  "success": true,
  "bundle_version": "2026-03-16T12:00:00Z",
  "rules_loaded": 47,
  "message": "Policy bundle reloaded from /opt/arbitex/policy/policy_bundle.json"
}

Response 400 Bad Request

{
  "detail": "Policy reload only available in air-gap mode"
}

Response 422 Unprocessable Entity — bundle validation failed

{
  "success": false,
  "bundle_version": null,
  "rules_loaded": 0,
  "message": "Policy bundle must contain a 'rules' array"
}

---

GET /admin/api/custom-patterns

List all loaded custom redaction patterns. Custom patterns extend the built-in DLP regex tier with operator-defined patterns. Enabled via CUSTOM_PATTERNS_ENABLED=true.

Response 200 OK

{
  "patterns": [
    {
      "name": "employee-id",
      "pattern": "EMP-[0-9]{6}",
      "replacement": "[EMPLOYEE_ID]",
      "enabled": true
    }
  ]
}

Returns {"patterns": []} if the custom pattern registry is not enabled.

---

POST /admin/api/custom-patterns

Add a new custom redaction pattern.

Request body

{
  "name": "employee-id",
  "pattern": "EMP-[0-9]{6}",
  "replacement": "[EMPLOYEE_ID]",
  "enabled": true
}

Field	Type	Required	Default
name	string	yes	—
pattern	string	yes	—
replacement	string	no	"[REDACTED]"
enabled	boolean	no	true

Response 201 Created

{
  "status": "added",
  "pattern": {
    "name": "employee-id",
    "pattern": "EMP-[0-9]{6}",
    "replacement": "[EMPLOYEE_ID]",
    "enabled": true
  }
}

Response 400 Bad Request: Missing name or pattern, or invalid regex.

Response 503 Service Unavailable: Custom pattern registry not enabled.

---

DELETE /admin/api/custom-patterns/{name}

Delete a custom redaction pattern by name.

Path parameter: name — the exact pattern name as registered.

Response 200 OK

{
  "status": "deleted",
  "name": "employee-id"
}

Response 404 Not Found: Pattern not found.

---

4. Provider Management

Providers can be temporarily disabled for maintenance, incident response, or cost control without modifying the policy bundle. All provider overrides are in-memory and reset on outpost restart.

GET /admin/api/providers

Lists all providers from the loaded policy bundle along with their current override state.

GET /admin/api/providers
Authorization: Bearer <token>

Response 200 OK

{
  "providers": [
    {
      "name": "anthropic",
      "base_url": "https://api.anthropic.com",
      "models": ["claude-sonnet-4-6", "claude-haiku-4-5-20251001"],
      "disabled": false,
      "disabled_until": null,
      "disable_reason": ""
    },
    {
      "name": "openai",
      "base_url": "https://api.openai.com",
      "models": ["gpt-4o", "gpt-4o-mini"],
      "disabled": true,
      "disabled_until": "2026-03-12T12:00:00Z",
      "disable_reason": "Incident response — elevated error rate",
      "override": {
        "disabled_by": "admin",
        "disabled_at": "2026-03-16T08:30:00Z",
        "disabled_until": "2026-03-16T10:30:00Z",
        "reason": "Maintenance window"
      }
    }
  ]
}

disabled_until is null for a permanent disable (no duration specified).

---

POST /admin/api/providers/{provider}/disable

Disables a provider, optionally for a fixed duration. All requests routed to this provider will fail with 503 until re-enabled.

POST /admin/api/providers/{provider}/disable
Authorization: Bearer <token>
Content-Type: application/json

{
  "duration_hours": 4,
  "reason": "Incident response — elevated error rate"
}

Field	Type	Required	Description
duration_hours	float	No	Auto-re-enable after this many hours. Omit for permanent disable
duration_minutes	integer	No	Alternative to duration_hours
reason	string	No	Reason recorded in the audit log

Returns 400 if the provider name is not in the policy bundle.

Response 200 OK

{
  "provider": "openai",
  "disabled": true,
  "disabled_until": "2026-03-16T10:30:00Z"
}

---

POST /admin/api/providers/{provider}/enable

Re-enables a previously disabled provider. Clears the disable override immediately.

POST /admin/api/providers/{provider}/enable
Authorization: Bearer <token>

No request body required.

Response 200 OK

{
  "provider": "openai",
  "enabled": true
}

---

5. Credential Intelligence

CredInt uses a bloom filter to detect leaked credentials in prompts. These endpoints manage the filter state and k-anonymity verification.

GET /admin/api/credint/status

Returns bloom filter status: whether the filter is loaded, entry count, snapshot date, target false-positive rate, and file size.

GET /admin/api/credint/status
Authorization: Bearer <token>

Response 200 OK (filter loaded)

{
  "enabled": true,
  "loaded": true,
  "entry_count": 847291054,
  "loaded_at": "2026-03-15T08:00:00Z",
  "file_size_bytes": 1073741824,
  "snapshot_date": "2026-03-01",
  "fpr_target": 0.001
}

Response 200 OK (filter not configured)

{
  "enabled": false,
  "loaded": false,
  "entry_count": 0,
  "loaded_at": null,
  "file_size_bytes": 0,
  "snapshot_date": "",
  "fpr_target": 0.0
}

Field	Type	Description
entry_count	integer	Number of entries in the bloom filter
file_size_bytes	integer	On-disk size of the bloom filter file
snapshot_date	string	Date of the credential snapshot
fpr_target	float	Target false-positive rate (e.g., 0.001 = 0.1%)

---

POST /admin/api/credint/reload

Reloads the CredInt bloom filter from the path configured in CREDINT_BLOOM_PATH. Hot-swaps the in-memory filter without an outpost restart.

POST /admin/api/credint/reload
Authorization: Bearer <token>

No request body required.

Response 200 OK (success)

{
  "success": true,
  "message": "Bloom filter reloaded",
  "entry_count": 847291054
}

Response 200 OK (not configured)

{
  "success": false,
  "message": "CredInt scanner not configured"
}

If reload fails (corrupt file, I/O error), the existing in-memory filter remains active (fail-open). Audit-logged as admin_credint_reload.

---

POST /admin/api/credint/kanon

Toggles k-anonymity verification on the CredInt scanner at runtime without a restart.

When k-anonymity is enabled (default), the outpost checks candidate credentials against the Have I Been Pwned k-anonymity API before surfacing a match. Disabling it suppresses those external checks — useful for air-gapped deployments.

Requires: Content-Type: application/json (returns 415 if missing).

Request body

{"enabled": true}

Field	Type	Required	Description
enabled	boolean	Yes	Enable or disable k-anonymity verification

Response 200 OK

{"credint_kanon_enabled": true, "message": "K-anonymity enabled"}

Response 503: {"detail": "Settings not initialised"}

Caution

This mutates both settings.credint_kanon_enabled and the live scanner’s _kanon_enabled flag atomically. Changes take effect immediately for all subsequent scans. Not persisted across restart. Audit-logged as admin_credint_kanon_toggle.

---

6. Budget & Usage

These endpoints provide visibility into token usage, estimated costs, and budget enforcement status for the current billing period.

GET /admin/api/usage

Returns the current-period usage summary from the outpost’s local usage tracker.

GET /admin/api/usage
Authorization: Bearer <token>

Response 200 OK (tracking enabled)

{
  "period_start": "2026-03-01T00:00:00Z",
  "period_end": "2026-03-31T23:59:59Z",
  "tokens_input": 1240000,
  "tokens_output": 890000,
  "total_requests": 3412,
  "estimated_cost_usd": 24.80,
  "budget_cap_usd": 500.00,
  "budget_remaining_usd": 475.20,
  "budget_alert_threshold_pct": 80
}

Response 200 OK (tracking disabled)

{
  "usage_tracking": "disabled",
  "totals": null,
  "budget_config": null
}

---

GET /admin/api/budget/status

Current budget enforcement status for the active billing period. Reports spend against configured caps with percentage tracking.

curl -s http://outpost.internal:8301/admin/api/budget/status \
  -H "Authorization: Bearer $ADMIN_API_KEY"

Response 200 OK

{
  "period": "2026-03-01T00:00:00Z",
  "period_type": "monthly",
  "dollar_cap": 500.0,
  "dollar_spent": 312.847291,
  "dollar_percent": 62.57,
  "request_cap": 100000,
  "request_count": 67432,
  "request_percent": 67.43,
  "status": "ok",
  "warning": false
}

Field	Values	Description
status	ok / warning / exceeded	warning when ≥ 80% of either cap; exceeded when ≥ 100%
warning	bool	true when status is warning
period_type	monthly / hourly	Billing period type from policy bundle

If usage tracking is disabled, returns {"usage_tracking": "disabled", "status": "ok", "warning": false}.

---

GET /admin/api/budget/export.csv

Export budget usage data as a downloadable CSV file. Returns up to 100,000 rows, newest-first. Writes an audit log event on each export.

curl -s http://outpost.internal:8301/admin/api/budget/export.csv \
  -H "Authorization: Bearer $ADMIN_API_KEY" \
  -o budget_usage_$(date +%Y%m%d).csv

Response — text/csv with Content-Disposition: attachment; filename=budget_usage.csv

CSV columns: date, model, provider, org_id, input_tokens, output_tokens, estimated_cost, period.

---

7. Certificate Management

The outpost operates two certificate contexts: mTLS certificates for the outpost-to-platform mutual TLS connection (/admin/api/cert/*), and TLS serving certificates for the outpost’s own HTTPS listener (/admin/api/certs/*).

GET /admin/api/cert/status

Returns mTLS certificate metadata and rotation daemon state.

curl -s "http://localhost:8301/admin/api/cert/status" \
  -H "Authorization: Bearer $ARBITEX_ADMIN_KEY"

Response 200 OK

{
  "cert_path": "certs/outpost.pem",
  "subject": "CN=outpost-prod-us-east.arbitex.ai",
  "issuer": "CN=Arbitex Outpost CA",
  "not_before": "2026-01-01T00:00:00Z",
  "not_after": "2026-06-30T00:00:00Z",
  "days_remaining": 106,
  "status": "amber",
  "last_check_at": "2026-03-15T10:00:00Z",
  "last_rotation_at": "2025-12-01T08:30:00Z",
  "consecutive_failures": 0,
  "total_rotations": 3,
  "needs_rotation": false,
  "rotation_state": "idle"
}

Field	Type	Description
status	string	Traffic-light: green (>30 days), amber (7–30 days), red (<7 days or unreadable)
consecutive_failures	integer	Consecutive rotation failures
total_rotations	integer	Lifetime successful rotations

If the cert rotation client is not configured, rotation fields return zero/null.

---

POST /admin/api/cert/renew

Triggers an immediate mTLS certificate rotation cycle via the CertRotationClient (calls back to Platform’s cert issuance endpoint). Installs the new certificate without restarting.

curl -s -X POST "http://localhost:8301/admin/api/cert/renew" \
  -H "Authorization: Bearer $ARBITEX_ADMIN_KEY"

Response 200 OK (rotated)

{"status": "renewed", "valid_until": "2026-09-30T00:00:00Z"}

Response 200 OK (not needed)

{"success": false, "message": "no_renewal_needed"}

Response 503: {"error": "Cert rotation client not configured"} or control plane unreachable.

Audit-logged as admin_cert_renew.

---

GET /admin/api/certs/status

Returns TLS serving certificate CN, expiry, and auto-rotation configuration.

Response 200 OK

{
  "cert_path": "certs/outpost.pem",
  "cn": "outpost-prod.example.com",
  "expiry": "2026-12-31T00:00:00Z",
  "not_after": "2026-12-31T00:00:00Z",
  "days_remaining": 290,
  "needs_rotation": false,
  "threshold_days": 30,
  "auto_rotate_enabled": true
}

Field	Type	Description
needs_rotation	boolean	True when days_remaining < threshold_days
threshold_days	integer	From CERT_ROTATION_THRESHOLD_DAYS (default 30)
auto_rotate_enabled	boolean	From CERT_AUTO_ROTATE setting

---

POST /admin/api/certs/rotate

Rotates the TLS serving certificate. Accepts a PEM payload or generates a self-signed cert.

Requires: CERT_AUTO_ROTATE=true (returns 403 if disabled).

Request body (optional)

{
  "cert_pem": "-----BEGIN CERTIFICATE-----\n...",
  "key_pem": "-----BEGIN PRIVATE KEY-----\n..."
}

If no body is provided, generates an RSA-2048 self-signed certificate (365-day validity). If cert_pem and key_pem are supplied, they are validated before writing.

Response 200 OK

{"status": "rotated", "cn": "outpost-prod.example.com", "expiry": "2027-03-16T00:00:00Z"}

Response 400: {"error": "Invalid PEM data: ..."} — malformed cert or key

Response 403: {"error": "Certificate auto-rotation is disabled. Set CERT_AUTO_ROTATE=true."}

Response 500: {"error": "Failed to write cert files: ..."} — filesystem or dependency error

Writes to OUTPOST_CERT_PATH and OUTPOST_KEY_PATH. Audit-logged as admin_cert_rotate.

Note

This endpoint does not enforce Content-Type: application/json. The body is parsed from raw bytes, making it tolerant of missing Content-Type headers.

---

8. Audit & SIEM

These endpoints provide local audit visibility through multiple interfaces: the JSONL file-backed buffer, the in-memory ring buffer, HMAC chain integrity verification, SIEM direct sink status, audit queue management, and the persistent sync checkpoint.

GET /admin/api/audit-buffer

Returns the last 200 events from the local JSONL audit buffer on disk. Useful for immediate review without exporting to a SIEM.

GET /admin/api/audit-buffer
Authorization: Bearer <token>

Response 200 OK

{
  "events": [
    {
      "timestamp": "2026-03-12T09:00:01Z",
      "action": "proxy_request",
      "user_id": "u-xyz",
      "provider": "anthropic",
      "model": "claude-sonnet-4-6",
      "dlp_result": "pass",
      "latency_ms": 210
    }
  ],
  "total": 47
}

total reflects the number of events in the response (up to 200), not the total events in the file.

---

GET /admin/audit/recent

Returns DLP scan events from the in-memory ring buffer (50-event capacity) with pagination.

Query parameters

Parameter	Type	Default	Constraints
limit	integer	50	max 200
offset	integer	0	—

Response 200 OK

{
  "events": [
    {
      "event_id": "dlp_01HZ9KQBW4J8YQNXT3R6MFD2AE",
      "prev_event_id": "dlp_01HZ9KQBW4J8YQNXT3R6MFD29X",
      "hmac_signature": "a3f7...",
      "timestamp": "2026-03-15T10:23:44Z",
      "org_id": "org_abc",
      "action_taken": "BLOCK",
      "was_blocked": true,
      "was_redacted": false,
      "scan_duration_ms": 2.34,
      "entities_detected": [
        {"type": "aws_access_key_id", "confidence": 0.98, "redacted": true}
      ],
      "entity_types_detected": ["CREDIT_CARD"],
      "request_body_hash": "sha256:ab12...",
      "response_body_hash": "sha256:cd34..."
    }
  ],
  "total": 847,
  "limit": 50,
  "offset": 0
}

Caution

The ring buffer holds a maximum of 50 events in memory. Older events are permanently discarded. For the JSONL file-backed buffer (200 events), use GET /admin/api/audit-buffer. Body hash fields (request_body_hash, response_body_hash) are only present when BODY_HASH_LOGGING_ENABLED=true. HMAC fields are only present when AUDIT_CHAIN_ENABLED=true.

---

GET /admin/audit/stats

Returns aggregated DLP scan statistics since outpost startup.

Response 200 OK

{
  "total_scans": 12450,
  "total_blocked": 34,
  "total_redacted": 127,
  "avg_scan_duration_ms": 8.7,
  "entity_type_counts": {
    "aws_access_key_id": 22,
    "credit_card": 89,
    "ssn": 41
  },
  "action_distribution": {
    "ALLOW": 12289,
    "BLOCK": 34,
    "REDACT": 127
  },
  "uptime_seconds": 86420,
  "scans_per_minute": 14.2
}

scans_per_minute is computed over a rolling 5-minute window. entity_type_counts and action_distribution are cumulative since startup.

---

GET /admin/audit/search

Searches DLP scan events in the ring buffer with filters. Filters are applied cumulatively (AND).

Query parameters

Parameter	Type	Default	Description
action	string	—	Substring filter on action_taken (case-insensitive)
entity_type	string	—	Substring filter on detected entity types (case-insensitive)
min_duration_ms	float	—	Minimum scan_duration_ms
blocked_only	boolean	false	Filter to action_taken == "BLOCK" only (overrides action)
limit	integer	20	max 100
offset	integer	0	—

Response 200 OK

{
  "events": [...],
  "total": 34,
  "limit": 20,
  "offset": 0
}

---

GET /admin/audit/verify

Verifies HMAC chain integrity of the in-memory DLP scan ring buffer. Walks all events, recomputes HMACs, and checks prev_event_id chain linkage.

Response 200 OK (chain valid)

{
  "valid": true,
  "events_checked": 847,
  "first_broken_at": null,
  "error": null
}

Response 200 OK (chain not enabled)

{
  "valid": true,
  "events_checked": 0,
  "first_broken_at": null,
  "error": "Audit chain not enabled (AUDIT_CHAIN_ENABLED=false)"
}

Response 200 OK (chain broken)

{
  "valid": false,
  "events_checked": 50,
  "first_broken_at": "a1b2c3d4-...",
  "error": null
}

first_broken_at is the event_id of the first event with a mismatched HMAC or broken chain link. The first event in the buffer is exempt from prev-chain checks (it may reference an evicted event).

---

GET /admin/api/audit/verify

Extended variant with richer response schema. Also verifies the HMAC chain of the ring buffer, but additionally includes first_event_id and last_event_id for cross-referencing against platform audit records.

Response 200 OK (chain valid)

{
  "verified": true,
  "total_events": 847,
  "first_event_id": "dlp_01HZ9KQBW4J8YQNXT3R6MFD100",
  "last_event_id": "dlp_01HZ9KQBW4J8YQNXT3R6MFD2AE",
  "chain_valid": true,
  "broken_at_event": null
}

Response 200 OK (chain broken)

{
  "verified": false,
  "total_events": 50,
  "first_event_id": "a1b2c3d4-...",
  "last_event_id": "e5f6a7b8-...",
  "chain_valid": false,
  "broken_at_event": "c9d0e1f2-..."
}

When AUDIT_CHAIN_ENABLED=false, returns verified: true with a note field explaining the chain is not enabled.

Note

Use GET /admin/api/audit/verify (API path) for scripted integrity checks — it provides first_event_id and last_event_id for cross-referencing against platform audit records.

---

GET /admin/api/audit-sync/checkpoint

Return the current state of the audit write-ahead log (WAL), persistent sync checkpoint, and live sync worker. Useful for diagnosing audit delivery gaps after network disconnects or outpost restarts.

Response 200 OK

{
  "wal": {
    "pending_entries": 12,
    "oldest_entry_at": "2026-03-16T13:55:00Z",
    "disk_usage_bytes": 24576,
    "wal_path": "/var/outpost/data/audit.wal"
  },
  "checkpoint": {
    "last_synced_event_id": "dlp_01HZ9KQBW4J8YQNXT3R6MFD29X",
    "last_synced_at": "2026-03-16T14:05:00Z",
    "total_synced": 835
  },
  "sync_worker": {
    "running": true,
    "last_flush_at": "2026-03-16T14:05:00Z",
    "flush_interval_seconds": 30,
    "consecutive_failures": 0,
    "degraded": false
  }
}

Field	Description
wal.pending_entries	Events written to WAL but not yet confirmed as synced to platform
checkpoint.last_synced_event_id	Event ID of the last successfully delivered audit event
sync_worker.consecutive_failures	Number of consecutive sync failures; triggers degradation mode

Any of wal, checkpoint, or sync_worker may be null if that component is not configured.

---

GET /admin/api/siem/status

SIEM direct sink status: sink type, enablement state, event counters, and last error.

curl -s http://outpost.internal:8301/admin/api/siem/status \
  -H "Authorization: Bearer $ADMIN_API_KEY"

Response 200 OK (SIEM enabled)

{
  "enabled": true,
  "sink_type": "splunk_hec",
  "events_sent": 84320,
  "events_failed": 3,
  "last_error": "connection timeout after 10s",
  "last_flush_at": "2026-03-16T14:07:01Z"
}

Response 200 OK (SIEM disabled)

{
  "enabled": false,
  "sink_type": null,
  "events_sent": 0,
  "events_failed": 0,
  "last_error": null,
  "last_flush_at": null
}

Field	Type	Description
sink_type	string | null	"splunk", "splunk_hec", or "http" when enabled; null when disabled
events_sent	integer	Cumulative count of successfully forwarded events since process start
events_failed	integer	Cumulative count of events that failed to forward
last_error	string | null	Most recent forwarding error message, or null if no errors

---

GET /admin/audit-queue/status

Return current local audit queue statistics. The outpost writes audit events to a local SQLite queue during platform outages (circuit breaker open / degradation mode).

curl -s http://outpost.internal:8301/admin/audit-queue/status \
  -H "Authorization: Bearer $ADMIN_API_KEY"

Response 200 OK

{
  "count": 247,
  "oldest_queued_at": "2026-03-16T08:14:22Z",
  "disk_usage_bytes": 1048576,
  "degradation_mode": true
}

---

POST /admin/audit-queue/flush

Trigger an immediate flush of all queued audit events to the Platform, bypassing the regular sync interval.

curl -s -X POST http://outpost.internal:8301/admin/audit-queue/flush \
  -H "Authorization: Bearer $ADMIN_API_KEY"

Response 200 OK

{
  "flushed": 247,
  "failed": 0,
  "duration_ms": 382.1
}

Returns 409 Conflict if the queue is empty.

---

DELETE /admin/audit-queue/purge

Permanently discard all events in the local audit queue. Requires the X-Purge-Confirm: yes header to prevent accidental data loss.

Caution

This operation is irreversible. Purged audit events cannot be recovered and will not be synced to the Platform.

curl -s -X DELETE http://outpost.internal:8301/admin/audit-queue/purge \
  -H "Authorization: Bearer $ADMIN_API_KEY" \
  -H "X-Purge-Confirm: yes"

Response 200 OK

{"purged": 247}

Returns 400 Bad Request if the X-Purge-Confirm: yes header is missing.

---

9. Configuration

These endpoints manage configuration export, backup/restore, diff/validate, hot-reload, and live key apply. Two config export variants are available: a structured JSON diagnostic export and a browser-downloadable backup file.

GET /admin/api/config/export

Diagnostic config export — returns a structured JSON snapshot of the outpost’s effective runtime configuration. All secret and credential fields are replaced with "[REDACTED]". An admin_config_export entry is written to the local audit log.

curl -s http://outpost.internal:8301/admin/api/config/export \
  -H "Authorization: Bearer $ADMIN_API_KEY"

Response 200 OK

{
  "exported_at": "2026-03-15T12:10:00Z",
  "exported_by": "ops-admin",
  "runtime": {
    "log_level": "info",
    "listen_port": 8300,
    "admin_port": 8301,
    "admin_api_key": "[REDACTED]",
    "platform_url": "https://platform.arbitex.ai",
    "platform_mtls_cert": "/etc/outpost/certs/client.crt",
    "platform_mtls_key": "[REDACTED]"
  },
  "airgap": {
    "airgap_enabled": false,
    "policy_path": "/etc/outpost/policy",
    "model_path": "/etc/outpost/models"
  },
  "dlp": {
    "tier1_enabled": true,
    "tier2_enabled": true,
    "tier3_enabled": false,
    "ner_enabled": true,
    "deberta_enabled": false,
    "credint_enabled": true
  },
  "siem": {
    "type": "splunk",
    "url": "https://splunk.corp.example.com:8088/services/collector",
    "token": "[REDACTED]",
    "buffer_capacity": 10000,
    "siem_output": true
  },
  "geoip": {
    "city_mmdb_path": "/etc/outpost/geoip/GeoLite2-City.mmdb",
    "asn_mmdb_path": "/etc/outpost/geoip/GeoLite2-ASN.mmdb"
  },
  "oauth": {
    "jwks_url": "https://platform.arbitex.ai/.well-known/jwks.json",
    "cache_ttl_seconds": 300,
    "scope_enforcement": true,
    "oauth_jwt_public_key": "[REDACTED]"
  },
  "policy_bundle": {
    "loaded": true,
    "version": "2026.03.14-r1",
    "etag": "\"a3f9c2b1d4e5\"",
    "updated_at": "2026-03-14T06:00:00Z",
    "provider_count": 12,
    "dlp_rule_count": 34,
    "compliance_bundle_count": 4,
    "signature_valid": true
  },
  "budget": {
    "daily_limit_usd": 500.00,
    "monthly_limit_usd": 12000.00,
    "enforcement": "block"
  }
}

Section	Contents
runtime	All top-level config keys; secrets replaced with "[REDACTED]"
airgap	airgap_enabled, policy_path, model_path
dlp	Tier enablement flags and NER/DeBERTa/CredInt settings
siem	Type, URL, buffer capacity, output flag; SIEM token redacted
geoip	City and ASN MMDB file paths
oauth	JWKS URL, cache TTL, scope enforcement flag; JWT public key redacted
policy_bundle	Loaded state, version, etag, updated_at, provider/rule/compliance counts, signature validity
budget	Budget limits and enforcement mode from the loaded policy bundle (omitted if bundle not loaded)

---

GET /admin/config/export

Browser-download variant. Returns all non-secret configuration fields as a downloadable JSON file. The following fields are omitted entirely (rather than redacted): admin_api_key, provider_key_encryption_key, siem_direct_token, oauth_jwt_public_key.

curl -s http://outpost.internal:8301/admin/config/export \
  -H "Authorization: Bearer $ADMIN_API_KEY"

Response headers

Content-Type: application/json
Content-Disposition: attachment; filename="outpost-config-backup-2026-03-15T12-10-00Z.json"

The response body is the same shape as /admin/api/config/export but with secret fields omitted rather than redacted, and without the exported_by field.

---

POST /admin/api/config/backup

Create a JSON snapshot of the current running configuration. Secret fields are masked as "***". The snapshot is stored in an in-memory ring (last 10 backups) and optionally persisted to disk if BACKUP_DIR is configured.

Response 200 OK

{
  "success": true,
  "timestamp": "2026-03-16T14:05:00Z",
  "fields": 42
}

fields is the count of configuration keys captured in the snapshot.

---

GET /admin/api/config/backup/list

List available config backups — timestamps and sizes (in-memory ring, up to 10 entries).

Response 200 OK

{
  "backups": [
    {
      "timestamp": "2026-03-16T14:05:00Z",
      "size": 2048
    },
    {
      "timestamp": "2026-03-16T12:00:00Z",
      "size": 2031
    }
  ],
  "count": 2
}

Backups are stored in memory only (ring buffer, last 10). They do not persist across restarts unless BACKUP_DIR is set.

---

POST /admin/api/config/restore

Restore configuration from a backup payload. Secret fields in the payload are ignored and cannot be restored via this endpoint.

Request body

{
  "config": {
    "log_level": "DEBUG",
    "policy_sync_interval": 60,
    "dlp_enabled": true
  }
}

Response 200 OK

{
  "success": true,
  "applied": 3,
  "skipped": 0,
  "skipped_keys": []
}

Response 400 Bad Request: Unknown config keys in payload, or missing config key.

Fields masked as "***" in the payload are automatically skipped. Secret fields (audit_hmac_key, policy_hmac_key, outpost_api_key, etc.) are always skipped. Audit-logged as admin_config_restore.

---

POST /admin/config/diff

Compare a proposed configuration against the running settings. Read-only — does not modify any state.

Request body

{
  "proposed_config": {
    "log_level": "DEBUG",
    "policy_sync_interval": 120
  }
}

Response 200 OK

{
  "changes": [
    { "key": "log_level", "from": "INFO", "to": "DEBUG" },
    { "key": "policy_sync_interval", "from": 60, "to": 120 }
  ],
  "change_count": 2,
  "has_breaking_changes": false,
  "summary": "2 changes: 2 modified"
}

has_breaking_changes is true if any changed key is in the breaking-change set (e.g., outpost_id, org_id, platform_management_url). All 26 configured settings fields are compared.

---

GET /admin/config/validate

Validate a proposed configuration JSON without applying it. Returns validation result, errors, warnings, and the diff against running settings.

Request body

{
  "log_level": "DEBUG",
  "rate_limit_requests_per_minute": 0,
  "tls_min_version": "TLSv1.0"
}

Response 200 OK

{
  "valid": false,
  "errors": [
    "rate_limit_requests_per_minute must be > 0, got 0",
    "tls_min_version must be 'TLSv1.2' or 'TLSv1.3', got 'TLSv1.0'"
  ],
  "warnings": [],
  "changes": [
    { "key": "log_level", "from": "INFO", "to": "DEBUG" }
  ]
}

Validation checks: rate-limit fields must be > 0; confidence/threshold fields must be 0–1; TLS cert paths must exist on disk; tls_min_version/tls_max_version must be "TLSv1.2" or "TLSv1.3". Unknown keys produce warnings, not errors.

---

GET /admin/api/config/reload-status

Return the timestamp and changed keys from the last config hot-reload.

curl -s http://outpost.internal:8301/admin/api/config/reload-status \
  -H "Authorization: Bearer $ADMIN_API_KEY"

Response 200 OK

{
  "last_reload_at": 1742137200.5,
  "last_reload_changes": [
    "LOG_LEVEL: debug -> info",
    "DLP_NER_ENABLED: true -> false"
  ]
}

last_reload_at is a Unix timestamp. last_reload_changes lists changed keys with old → new values.

---

POST /admin/api/config/reload

Trigger a config hot-reload programmatically — equivalent to sending SIGHUP to the outpost process. Re-reads all reloadable environment keys and applies changes in-place. Writes an admin_config_reload audit event.

Query parameters

Parameter	Type	Default	Description
dry_run	bool	false	Compute diff and validate but do not apply changes

# Preview what would change
curl -s -X POST "http://outpost.internal:8301/admin/api/config/reload?dry_run=true" \
  -H "Authorization: Bearer $ADMIN_API_KEY"

Response 200 OK

{
  "status": "ok",
  "changed_keys": ["LOG_LEVEL: debug -> info"],
  "restart_required_keys": [],
  "reloaded_at": 1742137200.5,
  "dry_run": false,
  "validation_errors": null
}

Response 422 Unprocessable Entity — pre-reload validation failed

{
  "status": "rejected",
  "validation_errors": [
    {
      "name": "AUDIT_HMAC_KEY",
      "status": "fail",
      "message": "AUDIT_HMAC_KEY must be at least 32 bytes"
    }
  ]
}

---

POST /admin/api/config/apply

Apply a single reloadable configuration key immediately without a restart. Only keys listed in RELOADABLE_KEYS can be applied at runtime. Non-reloadable keys (TLS cert paths, database URLs, crypto keys) require a full restart.

Request body

{"key": "log_level", "value": "DEBUG"}

Field	Type	Description
key	string	The OutpostSettings field name to update
value	string	New value — coerced to the field’s declared type

Response 200 OK

{
  "applied": true,
  "key": "log_level",
  "old_value": "INFO",
  "new_value": "DEBUG",
  "restart_required": false
}

Error responses

Code	Condition
400	Key not in RELOADABLE_KEYS, or missing key/value fields
422	Value cannot be coerced to the field’s declared type

Reloadable keys (common)

Key	Type	Description
log_level	string	Logging level: DEBUG, INFO, WARNING, ERROR
audit_flush_interval_seconds	int	Audit queue flush interval
credint_kanon_enabled	bool	CredInt k-anonymity (also available via /admin/api/credint/kanon)
heartbeat_interval_seconds	int	Platform heartbeat frequency
dlp_scan_timeout_ms	int	DLP scan timeout per request
budget_check_enabled	bool	Enable/disable budget enforcement

Sensitive values (api_key, secret, password, token) are redacted to *** in the response and audit trail. All applied changes are recorded in the config reload history ring buffer and written to the audit log.

---

GET /admin/api/config/reload-history

Return a paginated log of all config key changes applied via POST /admin/api/config/apply since startup.

Query parameters

Parameter	Default	Maximum	Description
limit	20	50	Entries per page
offset	0	—	Pagination offset

Response 200 OK

{
  "entries": [
    {
      "timestamp": 1742151720.4,
      "key": "log_level",
      "old_value": "INFO",
      "new_value": "DEBUG"
    },
    {
      "timestamp": 1742151600.1,
      "key": "audit_flush_interval_seconds",
      "old_value": "30",
      "new_value": "10"
    }
  ],
  "total": 2,
  "limit": 20,
  "offset": 0
}

The history ring buffer is in-memory and resets on outpost restart. Sensitive values are always redacted to ***.

---

10. Multi-Org

Multi-org mode allows a single outpost to serve multiple organizations simultaneously. Enabled via MULTI_ORG_MODE=true. This section also covers the policy cache, which is closely tied to per-org policy management.

GET /admin/api/orgs

Return org mode and active organisations.

Response 200 OK (single-org mode)

{
  "mode": "single",
  "orgs": [
    {
      "org_id": "org_abc",
      "last_seen": null,
      "request_count": 0
    }
  ]
}

Response 200 OK (multi-org mode)

{
  "mode": "multi",
  "max_orgs": 10,
  "active_orgs": 3,
  "orgs": [
    {
      "org_id": "org_abc",
      "last_seen": "2026-03-16T14:00:00Z",
      "request_count": 842
    },
    {
      "org_id": "org_xyz",
      "last_seen": "2026-03-16T13:59:00Z",
      "request_count": 315
    }
  ]
}

---

GET /admin/api/orgs/{org_id}/policy-status

Return policy status for a specific org in multi-org mode.

Path parameter: org_id — the org identifier.

Response 200 OK

{
  "org_id": "org_abc",
  "policy_loaded": true,
  "last_fetched": "2026-03-16T13:55:00Z",
  "cache_hits": 412,
  "cache_misses": 23
}

Response 400 Bad Request: Outpost is not in multi-org mode (MULTI_ORG_MODE=false).

---

GET /admin/api/cache/stats

Return policy cache statistics.

Response 200 OK

{
  "entries": 42,
  "hit_count": 9821,
  "miss_count": 412,
  "hit_rate_percent": 95.97,
  "eviction_count": 18,
  "ttl_seconds": 300,
  "enabled": true
}

If the policy cache is not configured, all counters are zero and enabled is false.

---

POST /admin/api/cache/clear

Clear all policy cache entries immediately.

Response 200 OK

{
  "success": true,
  "cleared": 42
}

cleared is the number of entries that were removed.

Caution

Clearing the cache forces all subsequent requests to re-evaluate policies from the policy bundle. In high-traffic deployments this can temporarily increase request latency until the cache warms up again.

---

11. Kill Switch & Drain

These endpoints provide immediate traffic control: the emergency kill switch halts all AI proxy traffic, drain mode enables graceful shutdown, routing overrides pin traffic to a specific provider, and prompt holds allow admin review of flagged requests.

POST /admin/api/emergency-kill

Activates or deactivates the emergency kill switch. When active, all LLM requests are rejected immediately with 503 — no DLP evaluation, no provider contact.

Immediate effect

Kill switch activation halts all AI proxy traffic immediately. Deactivation resumes traffic immediately. Both actions are logged to the audit trail with the requesting admin’s identity.

POST /admin/api/emergency-kill
Authorization: Bearer <token>
Content-Type: application/json

{
  "active": true,
  "reason": "Security incident — halting all AI traffic"
}

Field	Type	Required	Description
active	boolean	yes	true to activate, false to deactivate
reason	string	no	Reason logged to the audit trail

Response 200 OK

{
  "kill_switch_active": true,
  "activated_at": "2026-03-16T09:00:00Z",
  "reason": "Security incident — halting all AI traffic"
}

---

GET /admin/api/drain/status

Return graceful drain status. Draining mode is entered when the outpost is shutting down — no new requests are accepted and in-flight requests are allowed to complete.

Response 200 OK

{
  "draining": false,
  "inflight_requests": 7,
  "drain_start_time": null
}

When draining is active, drain_start_time is an ISO 8601 timestamp and draining is true.

---

POST /admin/api/routing-override

Pins all traffic to a specific provider and optional model, or clears an existing routing override. Routing overrides are in-memory and reset on outpost restart.

POST /admin/api/routing-override
Authorization: Bearer <token>
Content-Type: application/json

To set an override:

{
  "provider": "anthropic",
  "model": "claude-haiku-4-5-20251001",
  "reason": "OpenAI outage — routing all traffic to Anthropic fallback"
}

To clear the routing override and restore policy-default routing:

{"provider": null}

or:

{"clear": true}

Response 200 OK

{
  "override_active": true,
  "provider": "anthropic",
  "model": "claude-haiku-4-5-20251001",
  "set_at": "2026-03-16T09:05:00Z"
}

Returns 400 if the provider name is not in the policy bundle.

---

GET /admin/api/prompt-holds

Lists all prompt holds (pending and resolved). Prompt holds are raised when the DLP pipeline flags a request for human review. Holds are in-memory only and do not persist across outpost restarts.

GET /admin/api/prompt-holds
Authorization: Bearer <token>

Response 200 OK

{
  "holds": [
    {
      "hold_id": "hold_01J...",
      "user_id": "usr_01J...",
      "conversation_id": "conv_01J...",
      "status": "pending",
      "prompt_message": "Code generation requires confirmation.",
      "prompt_text_preview": "Write a Python script that...",
      "rule_ids": ["custom-secret-leak"],
      "created_at": "2026-03-16T09:10:00Z",
      "resolved_at": null,
      "resolved_by": null
    }
  ],
  "pending_count": 1
}

---

POST /admin/api/prompt-holds/{hold_id}/approve

Approves a pending prompt hold, forwarding the held request to the LLM.

POST /admin/api/prompt-holds/{hold_id}/approve
Authorization: Bearer <token>

Returns 404 if the hold is not found or already resolved.

Response 200 OK

{"hold_id": "hold_01J...", "status": "approved", "resolved_at": "2026-03-16T09:11:00Z"}

---

POST /admin/api/prompt-holds/{hold_id}/deny

Denies a pending prompt hold, returning an error to the requesting client.

POST /admin/api/prompt-holds/{hold_id}/deny
Authorization: Bearer <token>

Returns 404 if the hold is not found or already resolved.

Response 200 OK

{"hold_id": "hold_01J...", "status": "denied", "resolved_at": "2026-03-16T09:11:00Z"}

---

GET /admin/api/prompt-holds/events

Server-Sent Events (SSE) stream. Emits prompt_hold events in real time as new holds arrive. Backlogs all current pending holds on initial connect. Suitable for admin UI integrations.

curl -s -N "http://localhost:8301/admin/api/prompt-holds/events" \
  -H "Authorization: Bearer $ARBITEX_ADMIN_KEY" \
  -H "Accept: text/event-stream"

Event format

event: prompt_hold
data: {"hold_id": "hold_01J...", "status": "pending", "created_at": "2026-03-16T09:10:00Z"}

event: prompt_hold_resolved
data: {"hold_id": "hold_01J...", "status": "approved", "resolved_at": "2026-03-16T09:11:00Z"}

The stream remains open until the client disconnects.

---

POST /admin/api/prompt-holds/bulk-approve

Approve multiple pending prompt holds in one request.

Request body

Field	Type	Required	Description
hold_ids	string[]	yes	List of hold IDs to approve

curl -s -X POST \
  http://outpost.internal:8301/admin/api/prompt-holds/bulk-approve \
  -H "Authorization: Bearer $ADMIN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"hold_ids": ["hold_abc", "hold_def", "hold_ghi"]}'

Response 200 OK

{
  "results": {
    "hold_abc": true,
    "hold_def": true,
    "hold_ghi": false
  },
  "approved_count": 2
}

results maps each hold ID to true (approved) or false (not found or already resolved). Writes an admin_bulk_approve_prompt_holds audit event.

---

12. Plugins & Webhooks

The plugin system allows optional extensions to be loaded at startup (e.g., custom DLP detectors, routing hooks). Enabled via PLUGINS_ENABLED=true. Plugins are discovered from PLUGIN_DIR (default ./plugins). Each plugin .py file must expose a register_plugin() function returning an object implementing the PluginInterface protocol.

GET /admin/api/plugins

Lists all loaded plugins with their status.

Response 200 OK (plugins enabled)

{
  "plugins_enabled": true,
  "plugins": [
    {
      "name": "custom-redact-pii",
      "version": "1.2.0",
      "status": "enabled",
      "description": "Additional PII redaction patterns for EU deployments",
      "loaded_at": "2026-03-16T08:00:00Z",
      "hooks_registered": ["on_scan_complete", "on_startup", "on_shutdown"]
    },
    {
      "name": "siem-forwarder-v2",
      "version": "0.9.1",
      "status": "disabled",
      "description": "Alternate SIEM forwarding with batching",
      "loaded_at": "2026-03-16T08:00:00Z"
    }
  ]
}

Response 200 OK (plugins not enabled)

{"plugins": [], "plugins_enabled": false}

hooks_registered lists only hooks that are callable on the plugin instance.

---

POST /admin/api/plugins/{name}/enable

Enable a loaded plugin by name. Plugin name is case-sensitive and must match plugin.name.

Request body: {} (empty JSON; Content-Type: application/json required — returns 415 if missing)

Response 200 OK

{"status": "enabled", "name": "custom-redact-pii"}

Response 404: {"detail": "Plugin 'custom-redact-pii' not found"}

Response 503: {"detail": "Plugin system not enabled"}

---

POST /admin/api/plugins/{name}/disable

Disable a loaded plugin by name.

Request body: {} (empty JSON required)

Response 200 OK

{"status": "disabled", "name": "custom-redact-pii"}

Response 404: {"detail": "Plugin 'custom-redact-pii' not found"}

Response 503: {"detail": "Plugin system not enabled"}

Note

Disabling a plugin prevents it from receiving on_scan_complete and lifecycle hooks. The plugin object remains in memory and can be re-enabled without reloading the module.

---

GET /admin/api/webhook-emitter/stats

Returns webhook emitter statistics.

Response 200 OK (emitter configured)

{
  "enabled": true,
  "events_queued": 12,
  "events_sent": 4821,
  "events_delivered": 4821,
  "events_failed": 3,
  "events_in_flight": 2,
  "last_delivered_at": "2026-03-16T09:01:00Z",
  "last_event_at": "2026-03-15T10:23:44.123456+00:00",
  "last_failure_at": "2026-03-15T22:14:00Z",
  "last_failure_reason": "HTTP 503 from webhook endpoint"
}

Response 200 OK (emitter not configured)

{
  "enabled": false,
  "events_sent": 0,
  "events_failed": 0,
  "last_event_at": null
}

Configuration

Setting	Description
WEBHOOK_EMIT_ENABLED	Master switch (default false)
WEBHOOK_EMIT_URL	Destination URL for webhook POSTs
WEBHOOK_EMIT_SECRET	HMAC-SHA256 signing secret

Webhook payload format:

{
  "event_type": "scan.complete",
  "org_id": "org_abc",
  "action": "BLOCK",
  "entity_types": ["CREDIT_CARD"],
  "scan_duration_ms": 2.1,
  "timestamp": "2026-03-15T10:23:44.123456+00:00"
}

The request includes an X-Webhook-Signature header containing the HMAC-SHA256 hex digest of the payload body. Retries up to 3 times with exponential backoff (1s, 2s). events_failed increments after all retries are exhausted.

---

13. Software Updates

The outpost has two software update paths: the software update system (/admin/api/updates/*) manages downloading and staging update bundles for operator-applied container restarts, and the upgrade orchestrator (/admin/api/upgrade/*) manages the full automated lifecycle including apply and rollback.

GET /admin/api/updates/status

Return the current software update state: check result, download progress, staged version, and Ed25519 signature verification status. Requires SOFTWARE_UPDATE_RELEASE_URL to be configured.

Response 200 OK

{
  "status": "staged",
  "current_version": "0.19.1",
  "latest_version": "0.19.2",
  "release_notes": "Fix: routing override not persisting across restarts",
  "download_progress": 100,
  "staged_at": "2026-03-16T14:22:00Z",
  "staged_version": "0.19.2",
  "staged_path": "/var/outpost/updates/outpost-0.19.2.tar.gz",
  "signature_verified": true,
  "error": null
}

Field	Description
status	disabled | idle | checking | available | downloading | staged | error
download_progress	Integer 0–100. null until a download is in progress
staged_path	Filesystem path to the staged bundle. null if not yet staged
signature_verified	true when the Ed25519 bundle signature passed verification. null if not yet downloaded

When SOFTWARE_UPDATE_RELEASE_URL is not set, status is "disabled" and error is "SOFTWARE_UPDATE_RELEASE_URL not configured".

---

POST /admin/api/updates/check

Trigger an immediate version check against the configured release manifest URL. Returns 503 if SOFTWARE_UPDATE_RELEASE_URL is not configured.

Response 200 OK

{
  "update_available": true,
  "current_version": "0.19.1",
  "latest_version": "0.19.2",
  "release_notes": "Fix: routing override not persisting across restarts"
}

---

POST /admin/api/updates/download

Download and cryptographically verify the update bundle for the latest version. The bundle is verified against the embedded Ed25519 public key before being staged. Fail-closed: bundles with invalid signatures are rejected.

A staged update requires an operator-initiated apply (container restart with the new binary). There is no auto-apply path.

Returns 422 if signature verification fails.

Response 200 OK

{
  "status": "staged",
  "staged_version": "0.19.2",
  "staged_path": "/var/outpost/updates/outpost-0.19.2.tar.gz",
  "signature_verified": true,
  "bytes_downloaded": 18432000
}

Response 422 Unprocessable Entity (signature failure)

{
  "status": "error",
  "error": "Ed25519 signature verification failed — bundle rejected"
}

Tip

After staging, apply by replacing the outpost binary and restarting the container. The upgrade orchestrator (POST /admin/api/upgrade/run) handles this automatically for containerized deployments.

---

GET /admin/api/upgrade/status

Return the current upgrade orchestrator status.

curl -s "http://localhost:8301/admin/api/upgrade/status" \
  -H "Authorization: Bearer $ARBITEX_ADMIN_KEY"

Response 200 OK

{
  "phase": "idle",
  "last_result": "success",
  "last_upgraded_at": "2026-03-10T14:22:00Z",
  "current_version": "0.14.2",
  "history_count": 4
}

Phase	Description
idle	No upgrade in progress
checking	Checking for available updates
downloading	Downloading update package
verifying	Verifying download signature
backing_up	Creating backup of current installation
applying	Applying the update
post_verify	Verifying the updated installation
rolling_back	Automatic rollback in progress

---

POST /admin/api/upgrade/run

Execute the full upgrade lifecycle. The orchestrator runs all steps sequentially: check → download → verify → backup → apply → post-verify. On any step failure, the orchestrator automatically attempts to roll back.

Caution

This operation causes a brief restart of the outpost process during the apply step. Active requests are drained before the restart.

Response 200 OK

{
  "status": "success",
  "from_version": "0.14.1",
  "to_version": "0.14.2",
  "steps": [
    {"step": "check", "result": "ok", "duration_ms": 1200},
    {"step": "download", "result": "ok", "duration_ms": 8400},
    {"step": "verify", "result": "ok", "duration_ms": 230},
    {"step": "backup", "result": "ok", "duration_ms": 1100},
    {"step": "apply", "result": "ok", "duration_ms": 3200},
    {"step": "post_verify", "result": "ok", "duration_ms": 850}
  ]
}

Returns 409 Conflict if an upgrade is already in progress.

---

POST /admin/api/upgrade/rollback

Manually roll back the last staged upgrade. Use when upgrade/run failed without automatic rollback, or when a post-upgrade issue is discovered.

Response 200 OK

{"status": "rolled_back", "restored_version": "0.14.1"}

---

GET /admin/api/upgrade/history

Return the last 10 upgrade run summaries.

Response 200 OK

[
  {
    "run_id": "upg_01J...",
    "status": "success",
    "from_version": "0.14.1",
    "to_version": "0.14.2",
    "started_at": "2026-03-10T14:20:00Z",
    "completed_at": "2026-03-10T14:22:00Z"
  }
]

---

14. Operational Tools

These endpoints provide observability tooling: diagnostic bundles, metrics summaries, connection pool stats, circuit breakers, bandwidth adapter, DLP benchmarks, log export status, OpenTelemetry tracing status, Prometheus Pushgateway status, IP allowlist status, body hash config, and replay protection stats.

GET /admin/diagnostics

Export a sanitized diagnostic bundle for support and troubleshooting. All secret fields (API keys, HMAC keys, signing keys) are scrubbed from the output. Safe to share with Arbitex support.

Response 200 OK

{
  "generated_at": "2026-03-16T14:00:00Z",
  "generated_by": "[email protected]",
  "version": "0.12.0",
  "uptime_seconds": 3600,
  "config_summary": {
    "outpost_id": "outpost-prod-01",
    "org_id": "org_abc",
    "platform_management_url": "https://api.arbitex.ai",
    "log_level": "INFO",
    "dlp_enabled": true,
    "credint_enabled": true,
    "admin_api_key": "***",
    "audit_hmac_key": "***"
  },
  "audit_tail": [
    {
      "event_id": "a1b2c3d4-...",
      "timestamp": "2026-03-16T13:58:00Z",
      "action_taken": "ALLOW"
    }
  ],
  "metrics": [
    {
      "name": "outpost_requests_total",
      "count": 4821,
      "sum": 4821.0,
      "avg": 1.0
    }
  ],
  "health_checks": [
    {
      "check": "admin_key_set",
      "status": "pass",
      "message": "Admin API key configured"
    }
  ]
}

audit_tail contains the last 10 entries from the audit log file. config_summary omits all secret fields (shown as "***"). This endpoint writes an admin_diagnostics_export audit log entry.

---

GET /admin/api/metrics-summary

Return a JSON summary of the six custom OTel metrics with traffic-light status indicators, plus circuit breaker states. Returns 503 if prometheus-client is not installed.

Response 200 OK

{
  "metrics": [
    {
      "name": "outpost_requests_total",
      "display_name": "Total Requests",
      "unit": "requests",
      "status": "green",
      "display_value": 4821.0,
      "count": 4821,
      "sum": 4821.0,
      "avg": 1.0
    },
    {
      "name": "outpost_platform_reachable",
      "display_name": "Platform Reachable",
      "unit": "bool",
      "status": "green",
      "display_value": 1.0,
      "value": 1.0
    }
  ],
  "metric_count": 7,
  "circuit_breakers": [
    {
      "name": "platform_sync",
      "state": "closed",
      "status": "green"
    }
  ]
}

Traffic-light values: "green" (healthy), "yellow" (warning threshold), "red" (critical threshold). Histograms report count, sum, avg. Gauges report value. The synthetic outpost_platform_reachable metric reflects degradation manager state — 1.0 = reachable, 0.0 = degraded.

---

GET /admin/api/metrics/connections

HTTP connection pool statistics for the outpost proxy.

curl -s http://outpost.internal:8301/admin/api/metrics/connections \
  -H "Authorization: Bearer $ADMIN_API_KEY"

Response 200 OK

{
  "active_connections": 12,
  "idle_connections": 28,
  "max_connections": 200,
  "pool_started": true
}

---

GET /admin/api/circuit-breakers

List the state of every registered circuit breaker — both provider-level and per-model breakers.

curl -s http://outpost.internal:8301/admin/api/circuit-breakers \
  -H "Authorization: Bearer $ADMIN_API_KEY"

Response 200 OK

{
  "circuit_breakers": [
    {
      "name": "anthropic",
      "state": "closed",
      "failure_count": 0,
      "success_count": 8421,
      "last_failure": null
    },
    {
      "name": "openai:gpt-4o",
      "state": "half_open",
      "failure_count": 3,
      "success_count": 7,
      "last_failure": 1742137190.3
    }
  ]
}

States: closed (normal), open (blocking requests), half_open (testing recovery).

---

GET /admin/api/bandwidth

Bandwidth adapter statistics: throughput measurement, window sizing, and recommended audit batch size.

curl -s http://outpost.internal:8301/admin/api/bandwidth \
  -H "Authorization: Bearer $ADMIN_API_KEY"

Response 200 OK

{
  "enabled": true,
  "throughput_bps": 1048576,
  "window_size": 50,
  "window_capacity": 100,
  "recommended_batch_size": 50,
  "tier_name": "standard",
  "batch_min": 10,
  "batch_max": 100,
  "connection_health_status": "healthy"
}

Returns {"enabled": false, ...} when the bandwidth adapter is not configured.

---

GET /admin/api/benchmarks

Return DLP pipeline benchmark results from the most recent benchmark run.

Response 200 OK

{
  "run_at": "2026-03-15T08:00:00Z",
  "pipeline": "5-tier",
  "total_samples": 1000,
  "avg_latency_ms": 2.34,
  "p95_latency_ms": 6.12,
  "p99_latency_ms": 11.45,
  "tier_breakdown": {
    "tier0_tfidf":   { "avg_ms": 0.08, "p95_ms": 0.15 },
    "tier1_regex":   { "avg_ms": 0.21, "p95_ms": 0.45 },
    "tier2_ner":     { "avg_ms": 1.87, "p95_ms": 4.92 },
    "tier3_deberta": { "avg_ms": 8.43, "p95_ms": 18.20 },
    "tier4_credint": { "avg_ms": 0.34, "p95_ms": 0.72 }
  }
}

Response 404 Not Found: No benchmark results on disk — run benchmarks/dlp_pipeline_bench.py first.

Note

Benchmark results are read from benchmarks/benchmark_results.json in the outpost working directory. Results are static snapshots from the last manual benchmark run and do not update automatically.

---

GET /admin/api/log-export/status

Return log export status (file-based structured log export).

Response 200 OK

{
  "enabled": true,
  "path": "/var/log/arbitex/outpost",
  "format": "jsonl",
  "current_file_size_bytes": 204800,
  "total_files": 7,
  "oldest_file": "2026-03-10T00:00:00Z",
  "newest_file": "2026-03-16T14:00:00Z"
}

If log export is not configured, enabled is false and all other fields are empty/zero defaults.

---

GET /admin/api/otel/status

Return OpenTelemetry tracing status.

Response 200 OK

{
  "enabled": true,
  "endpoint": "http://otel-collector:4317",
  "service_name": "arbitex-outpost",
  "export_protocol": "grpc",
  "spans_exported": 12450,
  "export_errors": 0
}

If OTel is not configured, enabled is false.

---

GET /admin/api/pushgateway/status

Return Prometheus Pushgateway exporter status.

Response 200 OK

{
  "enabled": true,
  "url": "http://pushgateway:9091",
  "job": "outpost",
  "instance": "outpost-prod-01",
  "last_push_at": "2026-03-16T13:59:55Z",
  "last_push_status": "success",
  "consecutive_failures": 0,
  "is_circuit_open": false
}

If the Pushgateway exporter is not configured, enabled is false and all other fields are empty/zero defaults.

---

GET /admin/api/ip-allowlist/status

Return IP allowlist configuration and denied request count since startup.

Response 200 OK

{
  "enabled": true,
  "cidrs": ["10.0.0.0/8", "192.168.1.0/24"],
  "admin_exempt": true,
  "denied_count": 12
}

admin_exempt indicates whether the admin API (port 8301) is exempt from IP allowlist enforcement. When the allowlist is disabled (IP_ALLOWLIST_ENABLED=false), enabled is false and cidrs is an empty array.

---

GET /admin/api/body-hash/config

Return body hash logging configuration. Body hash logging writes HMAC-SHA256 hashes of request/response bodies into the DLP scan audit events. Enabled via BODY_HASH_LOGGING_ENABLED=true.

Response 200 OK

{
  "enabled": true,
  "algorithm": "sha256",
  "log_response": false,
  "key_source": "audit_hmac_key"
}

Field	Values
algorithm	"sha256" (default)
log_response	true = response body is also hashed; false = request body only
key_source	"audit_hmac_key" | "admin_jwt_secret" | "not_configured"

---

GET /admin/api/replay-protection/stats

Return replay protection cache statistics.

Response 200 OK

{
  "enabled": true,
  "cache_size": 248,
  "duplicates_blocked": 3
}

Replay protection rejects duplicate requests (same X-Request-ID) within the protection window. When disabled (REPLAY_PROTECTION_ENABLED=false), enabled is false and counters are zero.

---

Error Responses

Status	Cause
400	Bad request — missing or invalid field
401	Missing Authorization header, or empty token
403	Invalid admin API key, or operation not permitted
404	Resource not found (provider, rule, hold, etc.)
409	Conflict (duplicate request ID, upgrade already in progress, empty queue flush)
415	Missing Content-Type: application/json
422	Validation error (type coercion failure, bundle validation)
429	Brute-force rate limit exceeded — 5 failed auth attempts within 15 minutes
500	Unhandled server error or uninitialised component
503	Required component not initialised (e.g. no policy bundle loaded)

---

See also

• Outpost architecture overview — port layout and admin network isolation
• Credential Intelligence guide — bloom filter setup and CDN refresh
• Policy Simulator guide — detailed simulate response schema
• DLP Rules API reference — rule and ruleset IDs
• Outpost deployment guide — Kubernetes installation and initial configuration
• Outpost certificate management — Certificate rotation procedures
• PROMPT governance guide — How PROMPT holds are created and managed
• Audit Chain Integrity — HMAC chain design and verification