API Reference

Core decision endpoint. 87-module scoring pipeline.

New: grok_context param, sgraal_override / grok_decision / formal_override / notification_sent response fields

Batch scoring — up to 100 entries per call.

Zero-knowledge mode — SHA-256 proof hash, no content exposure.

Repair memory entries: REFETCH, VERIFY_WITH_SOURCE, REBUILD_WORKING_SET.

Natural language explanation — developer/compliance/executive audiences.

Close outcome — success/failure/partial with component attribution.

Side-by-side Sgraal vs Grok comparison. Formal contradiction detection.

Trace memory propagation — cascade risk, containment, affected agents.

Certify memory fidelity with cryptographic proof.

Multi-agent conflict detection with anti-consensus safeguard.

Decision audit trail with timestamp + omega mapping.

Fleet-wide decision analytics.

MemCube v2 JSON Schema.

SMRS v1.0 formal score definition.

Verify cryptographic integrity of the SHA256-chained audit log. Returns the verified-up-to entry id and the first divergence point if any.

Export the full hash-chained log. ?format=json | cef | leef. Each record carries the prior-hash for independent re-verification.

Verify a single audit entry's signature and chain position by id.

Generate an EU AI Act conformity declaration scoped to your account. Article-level evidence mapping based on system risk classification. JSON output suitable for regulator submission.

SLA report for the current billing period. Latency, availability, decision counts.

Forensic decomposition of a past preflight decision. Body: {"request_id": "..."}. Returns: detection layers fired, memory entries contributing to the decision, replay path, and a stable forensics_id.

Retrieve a previously-generated forensic report by id.

Trigger a calibration run against the built-in adversarial corpus using your account's current configuration.

Returns the most recent calibration report.

Cases the calibration run flagged for human review.

Mark a human-review case as resolved with a decision label.

List per-namespace write firewall rules. POST to upsert, DELETE /v1/firewall/rules/{namespace} to remove.

Recent write firewall violations. Pagination via ?limit= and ?cursor=.

Current circuit breaker state per agent and per cross-domain rule. Includes which agents are currently blocked and remaining cooldown.