Auth for Agents

Auth for Agents is a proxy layer between AI agents and OAuth-based APIs that enforces human approval and policy rules before actions are executed. It records every requested action, decision, and token usage for auditability.

Why This is Cool

Once you give an AI agent your OAuth tokens, it can do anything with no oversight. Auth for Agents sits between AI and your services, requiring your approval before actions execute. You maintain full control and get a complete audit trail of everything your AI agents do.

Getting Started

1. Go to the dashboard at https://auth-for-agents.vercel.app/
2. Create an API key - Go to "Agents" → "New Agent" and copy your API key
3. Create approval flows - Go to "Rules" → "New Rule" to define when actions need approval
4. (Optional) Test run with the sample app: Include your API key in the Authorization header:

./start-ai-agent.sh YOUR_API_KEY YOUR_EMAIL@gmail.com

Send a test email to yourself or test against another custom rule, to see how the agent reasons about each email

1. (Optional) Integrate your agent to your code: Build custom agentic Gmail flows with our API and your API key

curl -H "Authorization: Bearer ak_live_0212e25f9aad..." \
https://your-app.com/api/endpoint

Example: Request Gmail Access

curl -X POST https://your-app.com/api/oauth/initiate \
-H "Authorization: Bearer ak_live_0212e25f9aad..." \
-H "Content-Type: application/json" \
-d '{
"service_provider_id": "gmail",
"scopes": ["gmail.send", "gmail.readonly"]
}'

That’s it. The AI agent will monitor your Gmail, compose replies, and ask for your approval before sending.

License

MIT

auth-for-agents