From workspace to your first traceable Run.

Sign in and confirm your workspace

Open Console and confirm the active workspace. Membership, provider credentials, Skills, Runs, quota, and audit records are all scoped to it.

• Do not reuse access tokens across workspaces.
• Automations should retain the workspace ID, not its display name.

Connect and verify a model credential

Create and verify a credential in Console under Model Provider. GoClaw only returns redacted metadata; provider keys never appear in read APIs, logs, or Artifacts.

A provider key calls the model; a GoClaw access token authorizes a user or client to access a workspace. They are different credentials.

Choose the right integration

Use Playground for interactive work, HTTP API or Go SDK for services, goclaw for terminal automation, and MCP for external agent clients. All share the same authorization, approval, and audit semantics.

Follow the Run state after submission

A successful Run submission returns 202. Poll its state or consume resumable SSE; when a high-risk action pauses, inspect the Approval and decide against its bounds and version.

• Send a stable Idempotency-Key on writes for safe retries.
• Cancellation is a request; it does not mean cleanup has already completed.
• Resume a disconnected stream with Last-Event-ID to avoid gaps.

Treat the Artifact as the deliverable

Download the Artifact when an Agent Run completes. It records model identity, token usage, pinned Skill versions, Tool call summaries, and digests for tickets, evaluations, or release evidence.

Do not copy Run input, model output, or Artifacts into public issues; they may contain user data.

Verify first, then expand automation

Start with a low-risk request to verify workspace, model, and output. Then add approvals, retries, event recovery, and Artifact downloads. Production clients must handle 401, 403, 409, 429, and 5xx responses.