Installation
Offline demo: npx trigguard demo with no API key / no ~/.trigguard login → Decision: PERMIT · Receipt: VERIFIED. If an API key is present, the demo uses the live gateway and may return DENY (fail-closed). Live authorization: set TRIGGUARD_API_KEY, then authorize() against the Execution Gateway. Root SDK: authorize, verify, receipt, explain, createAgent, createClient - see SDK.
npm install trigguard
npx trigguard demo
pip install trigguard
pip install authgraph
pip install trigguard-litellm
# MCP: npx @trigguard/mcp
# Live: TRIGGUARD_API_KEY + authorize()
How it works
Architecture and decision semantics live on dedicated pages - this hub links into them.
Quickstart
Machine path: npm install trigguard then npx trigguard demo. Live path: TRIGGUARD_API_KEY + authorize(). Full walkthrough: quickstart.
Protocol reference
- TG-EXECUTION-AUTH-01 DraftPOST /execute, PERMIT / DENY / ESCALATE / SILENCE, fail-closed semantics
- TG-RECEIPT-SCHEMAReceipt fields, signature format, expiry semantics
- TG-KEY-DISCOVERY
/.well-known/trigguard-keys.json· key rotation - Decision vs enforcementDecisions vs EXECUTED/BLOCKED enforcement layer
- API referenceEndpoints, parameters, responses, error codes
- Authorization surfacesdeploy.*, infra.*, data.*, secrets.* taxonomy
- Receipt verificationCanonical verification guide
Runtime integration
Trust & verification
Deterministic authorization boundary - every request yields PERMIT, DENY, ESCALATE, or SILENCE. Report vulnerabilities to [email protected].
FAQ
- Protocol Verification vs Execution Gateway?
- Offline verification is live: Ed25519 over receipt JSON using
/.well-known/trigguard-keys.json. The Execution Gateway issues PERMIT / DENY / ESCALATE / SILENCE for irreversible actions viaauthorize()withTRIGGUARD_API_KEY- see TG-EXECUTION-AUTH-01. - API keys for verification?
- No. Use published keys with the CLI or /verify.
- Canonical quickstart?
- /docs/quickstart - offline demo first (
npx trigguard demo), then liveauthorize(). - SILENCE vs DENY?
- DENY is an explicit refusal. SILENCE withholds authorization without leaking policy signals.