Frequently Asked Questions

OmegaEngine is the governance layer between AI agents and the real world. It's a decision API that evaluates every autonomous action for risk before it executes, enforces your policies, runs multi-model arbitration on high-stakes calls (Business+ plans), and signs every decision into a cryptographic audit trail.

Teams whose AI agents touch things that matter: platform teams that can't let an agent near production unchecked, enterprises that need to prove why an agent did what it did, regulated industries that need replayable audit trails, and security teams that want their guardrails red-teamed continuously.

One API call where your agent asks for judgment on an action. Each decision includes risk scoring, policy evaluation, a verdict (approve / deny / escalate), a cryptographic signature, and an audit-trail entry. You're not paying for tokens — you're paying for judgment.

The policy, safety, and proof steps add only light overhead — in a full decision, most of the time is the LLM call itself, which depends on the model you route to. Deterministic safety checks run in under a millisecond. We return per-stage timing on every response so you can measure latency for your own workload instead of trusting a marketing number.

Create a free account at omegaengine.ai/signup, then go to Dashboard → API Keys and generate one. No credit card required for the free tier. Pass it as the x-api-key header on every request.

Fail-closed by design — it's an OmegaConstitution invariant. If the system fails, actions are blocked, not silently approved. Circuit breakers route around unhealthy model providers automatically. Live availability is published on our status page.

Install the SDK (npm install @omegaengine/sdk for TypeScript, pip install omegaengine for Python), initialize with your API key, and wrap agent actions with omega.authorize() — if the decision is 'approved', execute; otherwise escalate. Framework drop-ins like anthropic-omega, langchain-omega, and crewai-omega are also live on PyPI.

Yes. The MCP server (npx -y @omegaengine/mcp) gives Claude, Cursor, Windsurf, and any MCP client three tools: check_action (govern before execution), scan_agent (red-team an agent), and verify_proof (signature verification). Agents can also read omegaengine.ai/llms.txt for a machine-readable API summary.

OpenAI, Anthropic, Google, Mistral, Groq, DeepSeek, and Meta (Llama) — plus any OpenAI-compatible endpoint via BYOK. The full, current model catalog lives in our Model Support docs and is driven by the same file the router uses.

Yes — drop-in integrations exist for LangChain, LlamaIndex, CrewAI, AutoGen, Semantic Kernel, and the Vercel AI SDK, plus a GitHub Action for CI gates and webhooks for everything else. See the Integrations docs for setup guides.

Yes. The core is open source under Apache-2.0: clone the repo, configure .env.local, and run docker compose up — same API as the hosted service, on your infrastructure. Enterprise deployments in your own VPC are available by arrangement.

Not yet certified — we hold no formal certifications and we'd rather say so plainly than imply otherwise. The controls are implemented (signed audit trails, encryption, access controls) and a third-party audit is the remaining step. If you need a specific certification, talk to us about timelines.

Never. We commit in writing — in our DPA, for every tier — that your inputs, outputs, and decision payloads are never used to train, fine-tune, or improve AI models. Only aggregated, anonymized reliability telemetry (latency, error rates) is used to run the service.

In a managed PostgreSQL database, encrypted at rest and in transit; credentials you give us (API keys, webhook secrets) are additionally encrypted at the application layer with AES-256-GCM. Retention windows are published in our Privacy Policy, and request logs keep hashed IPs rather than raw addresses.

Both are live API endpoints, not an email queue: full data export and account deletion are available in-product, in line with GDPR portability and erasure rights. You can also reach privacy@omegaengine.ai for any data-subject request.

No — only agents and systems you own or are explicitly authorized to test, per our Acceptable Use Policy (omegaengine.ai/aup). During scans, tool calls are recorded, never executed, so no real side effect can fire. Unauthorized testing of third-party systems may violate computer-crime laws.

It's signed, offline-verifiable red-team evidence: every finding carries a cryptographic proof, mapped to OWASP, MITRE ATLAS, NIST AI RMF, and EU AI Act Article 15. It is not a certification or an audit — it's point-in-time evidence you can hand to your auditor, who can re-verify it without trusting us.

Yes — 100 decisions per month, no credit card required. Enough to integrate, run real scenarios, and evaluate the platform. Paid plans start with Developer at $299/month.

Yes — every plan is BYOK. You pay model providers directly and we never store one-off scan keys. OmegaEngine is a control plane: we never resell or manage inference, so there is one price per plan (no managed upcharge).

Plans are billed monthly or annually in advance; each includes a monthly decision allowance with transparent per-decision overage rates. You get alerts at 80% and 95% of your limit.

Cancel anytime — your plan stays active until the end of the current billing period, with no lock-in. Beyond that, fees are non-refundable except as required by law, as set out in our Terms of Service. The one-time $499 attestation is self-assessed; refund terms are on our Refund Policy page.

Starts at $15,000/month (BYOK). Unlimited decisions, 5-model arbitration, contractual SLA with service credits, SSO, custom integrations, custom retention, and deployment in your own VPC by arrangement. Contact sales@omegaengine.ai.

Email support@omegaengine.ai (we target a response within 2 business days, prioritized for paid plans), or use GitHub Discussions and Issues for community and bug reports. Security reports go to security@omegaengine.ai and are triaged ahead of everything else.