Inspect the audit trail
Follow proposal, policy, approval, one-use authority, attempt, result, and denial evidence without treating the local chain as independent attestation.
Community quickstart
v0.1 developer previewChoose the zero-credential local proof or connect an eligible ChatGPT workspace through OpenAI Secure MCP Tunnel. Both paths use the same local gateway, deterministic policy, and three simulated tools.
Deterministic first run
The bundled Community proof uses deterministic mock tools. It sends no email, changes no customer, needs no SaaS account, and keeps the unauthenticated gateway on loopback.
You will see a read execute, an email-shaped proposal pause for a human decision, and a destructive proposal stop at policy. The labels are realistic; the effects are simulated.
Cold-build time varies with network and host performance. Warm reruns reuse the local Docker cache and retained SQLite volume.
Start from the root of a fresh source download. The concierge checks prerequisites, starts a checkout-specific Docker project, verifies the runtime, and opens the matching browser guide.
The fastest proof needs no account or runtime credential. It keeps audit evidence in a local SQLite volume across stop and restart.
./actionproxy localUse the Node/Corepack workspace when changing the gateway or web console. This remains supported alongside the Docker concierge.
nvm use
corepack enable
corepack pnpm install --frozen-lockfile
corepack pnpm dev
# Open http://127.0.0.1:5173/#/demoFollow proposal, policy, approval, one-use authority, attempt, result, and denial evidence without treating the local chain as independent attestation.
Change a demo rule and confirm allow, deny, and require-approval behavior remain deterministic.
Edit a supported single-reviewer proposal and inspect the preserved original input plus policy re-evaluation.
Keep provider credentials in an external runner or downstream MCP server and use ActionProxy for authorization and lifecycle evidence.
Continue exploring