OAuth3

Describe what you want to do in plain English. The TEE drafts API bindings via LLM, then a human approves.

1. Sign up

curl -X POST ${location.origin}/signup \
  -H 'Content-Type: application/json' \
  -d '{"name":"my-agent"}'
→ { "tenant_id": "...", "api_key": "...", "tee_url": "..." }

2. Request a permit with intent

curl -X POST TEE_URL/permit \
  -H 'Authorization: Bearer <token>' \
  -H 'Content-Type: application/json' \
  -d '{
    "description": "Manage GitHub issues",
    "intent": [{
      "name": "github",
      "goal": "Create and list issues on owner/repo",
      "doc_urls": ["https://docs.github.com/en/rest/issues"],
      "secret_hints": ["GITHUB_TOKEN"]
    }]
  }'
→ { approval_url, status_url, permit_id, capabilities }

The TEE reads the docs and drafts scoped API bindings. Present approval_url to your human.

3. Poll for approval

fetch(status_url)  // long-polls up to 120s
→ { status: "completed", permit_id }

4. Execute code

curl -X POST TEE_URL/execute \
  -H 'Authorization: Bearer <token>' \
  -H 'Content-Type: application/json' \
  -d '{
    "permit_id": "...",
    "code": "const r = await github(\"GET\", \"/repos/o/r/issues\"); console.log(JSON.stringify(r));"
  }'
→ { status_url }

Your code runs in a SES sandbox with only the capabilities the human approved.

5. Get result

fetch(status_url)
→ { status: "completed", result: { success: true, stdout: "..." } }

Endpoints