We're here to help.
After installing Agent Constitution, the app displays a pairing code. Tell your AI agent the code — it will claim it and establish a secure channel. From there, your agent can make constitution checks before performing restricted actions. See the GitHub docs for integration details.
Open the app → Directives tab → tap "+" to add a directive. Write your directive in plain English — for example, "Ask before sending work emails" or "Require approval for purchases." Directives can cover any action: communication, purchases, social media, smart home, development commands, file operations, and more.
Approval/denial responses expire after 60 seconds. Make sure your agent is polling for responses with the wait parameter. If you're self-hosting the relay, check that it's running and the agent's pull request is reaching it.
Open the app → Directives tab → scroll to the Audit Log section. You'll see a chronological list of every constitution check, including what action was requested, which directive was triggered, and whether you approved or denied.
Yes! The relay is open source. Clone the repo, run npm install && npx tsx src/relay-node.ts, or deploy to any cloud provider. See the GitHub repo for deployment instructions.
AI agents can "drift" over time — forgetting directives or finding loopholes. Drift testing automatically sends realistic test prompts to your agent, evaluates responses, and shows you exactly where compliance dropped. Run tests from the app's Compliance tab.
Agent Constitution monitors 13 critical files on your agent (directives, scripts, encryption utilities). On each heartbeat, file hashes are computed and signed with HMAC. If your agent tampers with any protected file, the app shows a tamper warning immediately.
Yes! Vault data, health metrics, and constitution checks are protected with AES-256-GCM end-to-end encryption. The relay server cannot read your data — only your iPhone and your agent can decrypt it.
Yes — it's a pure HTTP API. Any agent framework that can make HTTP requests can perform constitution checks. We provide a ready-made skill package for OpenClaw, but you can integrate with any agent by calling the API directly.
Found a bug? Please open an issue on GitHub:
github.com/arunrlverma/agent-constitution/issues
If you've found a security vulnerability, please report it responsibly. Do not open a public issue. Instead, email us directly:
For general questions, feedback, or anything else:
Arun Verma
Email: contact@agentconstitution.com
GitHub: github.com/arunrlverma/agent-constitution