First call in five minutes

Sign up at app.edge.scrutari.ai and pick Growth: the AI gateway is a Growth capability, and the first 30 days are free with no card (free Starter covers the PQ-TLS edge, not AI routing). In the dashboard, add a provider credential (your own key for OpenAI, Anthropic, Google, AWS Bedrock, or Azure; we never resell tokens) and mint an API key with the inference:invoke scope.

curl https://api.edge.scrutari.ai/v1/chat/completions \
  -H "Authorization: Bearer $SCRUTARI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-4-6",
    "messages": [{"role": "user", "content": "Say hello."}]
  }'

The model name selects an entry from your workspace catalog; the entry decides which upstream provider serves it. Anthropic-native clients keep their SDK and use the messages endpoint unchanged.

Every workspace has a monthly token quota, and it fails closed. No silent degradation.

Every call writes an Ed25519-signed audit record; anchors make history tamper-evident. Evidence-pack export ships with Enterprise (Settings → Compliance export in the dashboard); once a pack is exported, anyone can verify it on any machine, air-gapped included, with the open-source verifier:

cargo install scrutari-verify
curl -LO https://scrutari.ai/demo-evidence-pack.jsonl
scrutari-verify --pack demo-evidence-pack.jsonl

The sample pack is fully synthetic (fictional tenant, throwaway keys, no real data) and passes every check. Flip one byte in it and run again to watch verification break.

It proves the pack is complete, untampered, included under the signed anchors, and signed by the declared keys, without talking to us.

Two surfaces are live in early access, and our own outreach fleet runs on both: governed agent memory (redaction before storage, per-entry provenance, access control at read time, an identical-write loop guard) and the governed MCP proxy with per-server allow-lists and a signed record per tool call. Ask for access from the dashboard, or read more on the MCP gateway page.