API reference

Base URL: https://firewall.orivael.dev

All API endpoints require an Authorization: Bearer axfw_... header.

`POST /v1/guard/check`

Classify a prompt and return a verdict.

Request

POST /v1/guard/check
Authorization: Bearer axfw_3kPa7QxR9mNvL2eFhJtBcDeFgHiJkLmNoPqRsTuVwXyZ
Content-Type: application/json

{"text": "What is the weather today?"}

Field	Type	Required	Description
`text`	string	yes	The prompt to classify. UTF-8, any length.

Response (`200 OK`)

{
  "verdict": "allow",
  "intent": {
    "class": "INFORM",
    "confidence": 0.55,
    "signals": [],
    "signature": "c487f1..."
  }
}

Field	Type	Description
`verdict`	`"allow"` \| `"block"`	What to do with this prompt.
`intent.class`	string	One of `INFORM`, `CLARIFY`, `REFUSE`, `HARM`, `DECEIVE`, `UNCERTAIN`.
`intent.confidence`	float	0.0 – 1.0. Higher = the classifier is surer.
`intent.signals`	string[]	Pattern hits that contributed (e.g. `["harm:1"]`).
`intent.signature`	string	HMAC-SHA256 of the verdict for audit replay.

Errors

Status	Body	Meaning
400	`{"detail": "Field 'text' must be a string"}`	Malformed request.
401	`{"detail": "Invalid or missing API key"}`	Bearer token wrong or revoked.
429	`{"detail": "...", "used": N, "limit": N, "retry_after_seconds": N}`	Free-tier quota exhausted. `Retry-After` header set to seconds until next month.
500	`{"detail": "..."}`	Firewall API error.

Latency target

p50: 5 ms
p99: 50 ms

(Measured server-side, excluding network round-trip. Add 30 – 100 ms for global users connecting to the US region.)

Data Gate

All Data Gate endpoints require Authorization: Bearer axfw_... plus X-Axiom-Tenant: <tenant_id>.

`PUT /data_policy/rule`

Create or replace the access rule for an agent.

PUT /data_policy/rule
Authorization: Bearer axfw_...
X-Axiom-Tenant: acme
Content-Type: application/json

{
  "agent_id": "callguard",
  "blocked_data_classes": ["PAN", "CVV", "SSN", "GDPR-9"],
  "allowed_actions": ["read", "summarise"],
  "blocked_actions": ["store", "forward", "log"]
}

Field	Type	Description
`agent_id`	string	Identifies the agent this rule applies to.
`blocked_data_classes`	string[]	Data class codes to deny (prefix-matched: `"GDPR-9"` blocks all `GDPR-9-*` sub-classes).
`allowed_data_classes`	string[]	If non-empty, only these classes are allowed. Empty = allow all (subject to blocked list).
`blocked_actions`	string[]	Actions to deny (case-insensitive: `"store"`, `"forward"`, `"log"`, etc.).
`allowed_actions`	string[]	If non-empty, only these actions are allowed.

Response 200 OK: the saved rule with rule_id and created_at.

`POST /data_policy/check`

Ask whether an agent is allowed to perform an action on a data class.

POST /data_policy/check?agent_id=callguard&action=store&data_class=PAN
Authorization: Bearer axfw_...
X-Axiom-Tenant: acme

Response:

{
  "allowed": false,
  "agent_id": "callguard",
  "action": "store",
  "data_class": "PAN",
  "reason": "blocked_data_class"
}

reason values: rule_allow · blocked_data_class · blocked_action · data_class_not_in_allowlist · action_not_in_allowlist · no_rule_sensitive_default_deny · no_rule_default_allow

`DELETE /data_gate/erasure`

Right-to-erasure for a data subject. Deletes all decision records that contain the subject identifier and returns a signed deletion certificate.

DELETE /data_gate/erasure?subject_id=user%40example.com
Authorization: Bearer axfw_...
X-Axiom-Tenant: acme

Response:

{
  "cert_id": "uuid",
  "tenant_id": "acme",
  "subject_id_hash": "sha256(subject_id)",
  "records_erased": 3,
  "erased_at": "2026-06-04T12:00:00.000000",
  "scope": "decision_log_only",
  "limitation": "Latent encodings in model weights are outside scope...",
  "signature": "hmac-sha256:..."
}

The signature is HMAC-SHA256 over the certificate payload using AXIOM_MASTER_KEY. Present the certificate to a regulator as proof of erasure. The scope and limitation fields are required disclosures per GDPR Article 17(3)(e).

Flight Recorder

`POST /flight_recorder/search`

Search the decision log.

POST /flight_recorder/search
Authorization: Bearer axfw_...
X-Axiom-Tenant: acme
Content-Type: application/json

{
  "verdict": "block",
  "intent_class": "HARM",
  "since": "2026-06-01T00:00:00",
  "until": "2026-06-04T23:59:59",
  "limit": 100,
  "offset": 0
}

All fields are optional. Returns {"decisions": [...], "limit": N, "offset": N}. Input/output text is not included in list results — fetch the individual record for full content.

`GET /flight_recorder/decision/{decision_id}`

Retrieve a full decision record including input_text and output_text.

GET /flight_recorder/decision/d3f4a1b2-...
Authorization: Bearer axfw_...
X-Axiom-Tenant: acme

`POST /flight_recorder/replay/{decision_id}`

Re-evaluate a historical decision against the current loaded policy and return a delta report.

POST /flight_recorder/replay/d3f4a1b2-...
Authorization: Bearer axfw_...
X-Axiom-Tenant: acme

Response:

{
  "decision_id": "d3f4a1b2-...",
  "original_verdict": "allow",
  "original_intent_class": "INFORM",
  "original_confidence": 0.72,
  "original_timestamp": "2026-05-01T...",
  "replay_verdict": "block",
  "replay_intent_class": "HARM",
  "replay_confidence": 0.94,
  "policy_delta": true
}

policy_delta: true means the current policy would have decided differently. Use this to detect regressions when tightening rules.

`GET /flight_recorder/export`

Export decisions in bulk for compliance or SIEM ingestion.

GET /flight_recorder/export?fmt=csv&verdict=block&since=2026-06-01
Authorization: Bearer axfw_...
X-Axiom-Tenant: acme

`fmt`	Content-Type	Description
`json`	`application/x-ndjson`	One JSON object per line
`csv`	`text/csv`	CSV with header row
`splunk`	`application/x-ndjson`	Splunk HEC format: `{"time": "...", "event": {...}}`
`datadog`	`application/json`	Datadog Logs JSON array

Additional query params: verdict, intent_class, since, until, limit (max 10 000 per request).

`PUT /flight_recorder/alerts`

Configure outbound alert destinations.

PUT /flight_recorder/alerts
Authorization: Bearer axfw_...
X-Axiom-Tenant: acme
Content-Type: application/json

{
  "webhook_url": "https://your-server.example/axiom-hook",
  "slack_webhook_url": "https://hooks.slack.com/services/T.../B.../...",
  "email_to": "secops@example.com",
  "email_from": "axiom@orivael.dev",
  "alert_on_verdicts": ["block"],
  "alert_on_intents": ["HARM", "DECEIVE"]
}

Alerts fire synchronously on record_decision() when the verdict/intent filters match. smtp_password is accepted on write but never returned on read.

`POST /signup`

Browser-only — used by the dashboard. Not part of the public API.

Per-IP rate-limited: 5 attempts per hour. Exceeding returns 429 with a Retry-After header.

`POST /login`, `POST /logout`

Browser-only — session-cookie auth for the dashboard. Not part of the public API.

`POST /research/run`

Run the 9-agent constitutional research pipeline against a question. Safety and Ethics agents can HALT the pipeline early if critical risks are detected. All per-step results are HMAC-signed.

Uses the same NIM > Anthropic backend as the guard API — configure NIM_API_KEY / NVIDIA_API_KEY for NIM or ANTHROPIC_API_KEY for Anthropic. NIM_MODEL controls the model when NIM is active.

Request

{
  "question": "Does intermittent fasting reduce inflammation?",
  "steps": 9,
  "model": null
}

Field	Type	Description
`question`	string	Research question to investigate
`steps`	integer 1–9	Number of pipeline steps (default 9)
`model`	string \| null	Override model for all agents (optional)

Response (`200 OK`)

{
  "question": "Does intermittent fasting reduce inflammation?",
  "steps_completed": 9,
  "halted": false,
  "halt_reason": null,
  "manifests": [...],
  "context": {
    "hypothesis": {...},
    "literature": {...},
    "simulation": {...},
    "critique": {...},
    "safety": {...},
    "ethics": {...},
    "data": {...},
    "experiment": {...},
    "report": {...}
  }
}

If halted is true, the pipeline stopped early (step 5 Safety or step 6 Ethics). halt_reason explains why. Steps after the halt are absent from context.

Errors

Code	Meaning
`400`	Empty question
`500`	Pipeline error (backend unavailable, model error)
`503`	ResearchPipeline not available — check server logs

Billing endpoints

POST /billing/upgrade/{tier}, POST /billing/portal, POST /billing/webhook — used internally by the dashboard and by Stripe. Contact sales@orivael.dev for pricing details.

Webhooks (Phase 2+)

Webhooks for verdict events (e.g. "notify my Slack on every block") are planned for Phase 2.

API reference

POST /v1/guard/check

Request

Response (200 OK)

Errors

Latency target

Data Gate

PUT /data_policy/rule

POST /data_policy/check

DELETE /data_gate/erasure

Flight Recorder

POST /flight_recorder/search

GET /flight_recorder/decision/{decision_id}

POST /flight_recorder/replay/{decision_id}

GET /flight_recorder/export

PUT /flight_recorder/alerts

POST /signup

POST /login, POST /logout

POST /research/run

Request

Response (200 OK)

Errors

Billing endpoints

Webhooks (Phase 2+)

`POST /v1/guard/check`

Response (`200 OK`)

`PUT /data_policy/rule`

`POST /data_policy/check`

`DELETE /data_gate/erasure`

`POST /flight_recorder/search`

`GET /flight_recorder/decision/{decision_id}`

`POST /flight_recorder/replay/{decision_id}`

`GET /flight_recorder/export`

`PUT /flight_recorder/alerts`

`POST /signup`

`POST /login`, `POST /logout`

`POST /research/run`

Response (`200 OK`)