Agent mode — discover, purchase, use

An autonomous agent can adopt D.A.T.D.A with no human in the loop, inside a spend cap its owner set.

1. Discover

GET /.well-known/agent.json
// capabilities, pricing, endpoints, and the self-provision URL

2. Self-provision (within an owner budget)

POST /v1/agent/onboard
{ "owner_email": "owner@company.com", "budget_cents": 500 }
// → { account_id, api_key, budget_cents, starter_scans, mcp, scan }

3. Buy credits as needed

POST /v1/agent/credits    (Authorization: Bearer <agent key>)
{ "credits": 100000 }
// budget-enforced; 402 budget_exceeded if it would exceed the cap

4. Use

Call /v1/scan or the scan MCP tool before every consequential action. When the allotment and credits are exhausted the API returns 402 with a re-purchase link the agent can call again.

Kill switch

An agent cannot release its own quarantine. Only an operator session can call /v1/agent/{id}/release. This boundary is deliberate.