Governance for the agentic enterprise
Every AI decision,
signed.
One API in front of any LLM. Orivael classifies intent, blocks harm, deception, and PII leakage in under 50 ms — and signs every verdict, so your auditors never have to take your word for it.
- verdict
- allow
- intent
- INFORM
- confidence
- 0.97
- latency
- 38 ms
- input
- "What is the weather today?"
hmac-sha256 · a9f2c41d…7e0b · axiom-mcp-v1
✓ VERIFIES AGAINST YOUR KEY
One prompt away
from the headline.
Every team shipping an LLM feature is one prompt-injection or PII leak away from a breach notification. Building that guardrail in-house means hiring a security team most companies don't have — so most companies ship the feature and hope.
Injection & jailbreaks
Six-class constitutional intent verdict — INFORM, CLARIFY, REFUSE, HARM, DECEIVE, UNCERTAIN — with reasons, not just a boolean.
PII & data leakage
Secrets, credentials, and regulated data are caught before they leave your pipeline — with a signed record of what was stopped.
Filter or full proxy
Run as a lightweight filter alongside your model, or as a drop-in OpenAI-compatible proxy with a one-line base_url change.
Custom policy
Layer your own block patterns, whitelists, and compliance packs on top of the default classifier — versioned and auditable.
The verdict competitors
can't forge.
Every other guardrail returns an unsigned string you have to trust. Orivael returns an HMAC-signed manifest — a cryptographic record of the decision, the policy that made it, and the moment it happened. Hand it to a regulator, replay it in an adversarial proceeding, verify it against a key Orivael never holds.
curl -X POST https://firewall.orivael.dev/v1/guard/check \
-H "Authorization: Bearer $AXIOM_KEY" \
-d '{"text": "What is the weather today?"}'
# {"verdict": "allow", "intent": {"class": "INFORM", ...}}Two lines to integrate. The signature is the product.
From your editor,
governed.
Axiom ships an MCP server for Claude Desktop, Claude Code, Cursor, and any stdio client. Twenty-six signed tools across the patent stack — every result HMAC-verified client-side.
axiom_guard_checkConstitutional boundary check on any inputCore
axiom_intent_gate_checkORVL-016 intent classifier with signed verdictsPatent
axiom_casORVL-008 Constitutional Adversarial SandboxPatent
axiom_ledgerAppend-only signed audit logRuntime
axiom_stopNamed-owner stop gate + kill switch for irreversible actionsRuntime
Full tool reference →25 tools across core, patent emulators, and runtime
Pricing
Start free. Scale when
the auditors show up.
Every tier includes signed verdicts, the dashboard, and the default classifier. No card required to start.
Your AI is making decisions.
Start signing them.
Free tier, 1,000 calls a month, no card. Live in under a minute.