Connect a protected action. Verify the evidence.

1. Talk to us for a design-partner engagement, or sign in if you already have an account.
2. Create your organisation.
3. Create an API key.
4. Ask Guard for approval, make the provider call yourself, then settle actual usage.
5. Open Evidence after the first governed request.

Proxy integration is not currently enabled in this environment. Use the Guard API / sidecar path.

Proxy integration is not currently enabled in this environment.

Use Guard API / sidecar: ask Guard, execute if allowed, then settle actual usage.

Ask Guard for approval, make the provider call yourself, then settle actual usage.

1. Ask Guard — POST /v1/govern/llm
2. Execute if allowed
3. Record actual usage — POST /v1/settle

Risky actions need an extra gate before execution: tool calls, email/webhook sends, memory writes, deploys, connector writes and external fetches.

POST /v1/atg/gate — ATG (Agent Trap Guard)

Govern, settle and ATG gate responses can include a signed receipt. Verify with POST /v1/verify before treating the decision as authoritative downstream.

• POST /v1/auth/register · POST /v1/auth/login
• POST /v1/tenants · POST /v1/tenants/:id/keys — tenant create grants server-controlled opening credit
• POST /v1/govern/llm · POST /v1/settle
• POST /v1/atg/gate — for policy mutation use action_kind=policy_write; the server issues the write scope
• GET /v1/policies · GET /v1/policies/lifecycle — discover write_capability (no hidden URL to memorise)
• Policy lifecycle: create → preflight → canary → activate → rollback (each write needs a reservation-bound policy_write capability)
• GET /v1/statements/credits · GET /v1/statements/carbon · GET /v1/statements/enforcement · GET /v1/evidence/enforcement/:id
• POST /v1/verify

Enterprise V1 customer contract: PREFLIGHT, ENFORCE, RECORD, REPORT. Machine-readable manifest: docs/enterprise_v1_public_api.json.

Full operator tools remain in the console Developer section.

/v2/authority and /v2/approvals remain mounted for broker/agent authority v2. They are not the V1 ATG capability contract.

For high-risk actions in Enterprise V1, use POST /v1/atg/gate to obtain a capability, then POST /v1/capabilities/verify or POST /v1/capabilities/consume. Do not treat v2 authority as a second public V1 verb surface.

See the Trust Centre for identity, signing, enforcement, recovery, release evidence and the tamper-evident audit replica.

Customer-facing latency claims use the real-provider measurement only. The simulated harness is engineering evidence and is not used as a customer claim.

• Environment: Sparse Guard stage
• Provider / model: OpenAI gpt-4o-mini
• Concurrency: 1
• 200 paired requests (warmup discarded)
• Sidecar added latency: median 393.263 ms · p95 970.36 ms
• Proxy added latency: median 559.234 ms · p95 1030.189 ms
• Added TTFT sidecar: median 446.268 ms · p95 875.486 ms
• No provider errors, Guard errors, or rate-limit errors in the qualified sample
• No double settlement and no cross-tenant effects

Not a production benchmark. Not all models or regions. Relative percentage figures are not used as marketing claims.