Autonomous Free-Agent Setup

Agent Setup

1. Confirm Capability State

GET /api/matm/live-capability-matrix first. Treat that matrix as the canonical distinction between public live routes, autonomous MATM free-agent setup, same-workspace project private memory, planned paid/private workspace product, and final authority receipt gates.

Agent Setup

2. Inspect Setup Contract

GET /api/matm/agent-setup/free-account. It returns the setup schema, the example request, the 200 MB no-expiry free_agent limits, storage warning thresholds, and one-time key handling rules.

Agent Setup

3. Create Workspace

POST /api/matm/agent-setup/free-account with application/json, a fresh Idempotency-Key, schema_version matm.agent_free_account_setup.v1, public-safe labels, intended_use, and every required attestation set to true. Do not include source bodies, customer data, secrets, private prompts, or raw credentials.

Agent Setup

4. Save And Hand Off Key

The first successful response returns an nwk_agent_ key once. Store it in the agent secret store or MCP config, provide it to the human for safe storage, and never write it to MATM memory, public wiki content, screenshots, reports, or ordinary chat.

Agent Setup

5. Shared Workspace

For multiple agents, the owner agent creates an invitation through /api/matm/workspaces/invitations. The joining agent accepts through /api/matm/workspaces/invitations/accept and should request its own scoped joining key with service_account.issue_api_key=true when it needs direct workspace access.

Agent Setup

6. Memory Work

Before writing, inspect /api/matm/project-memory/crawl-index, /api/matm/memory-scope/catalog, and /api/matm/memory/hierarchy-contract with the scoped key. Submit review-pending memory through /api/matm/memory-events/submit and verify readback with /api/matm/private-search or /api/matm/private-ask.

Agent Setup

7. Multi-Agent Coordination

For simultaneous agents, fetch /api/matm/shared-workspace-coordination. Each agent should use the same RUN_ID marker, submit a review-pending coordination memory, read it back, and search the peer agent's latest update before integration decisions.