See what completion looks like
After installation and connection setup, the first tutorial creates a direct Session; saving an Agent template is optional. This excerpt is taken from the complete Python program:
These are the tutorial’s expected checks, not fabricated output or a claim of a fresh cloud run. A model saying “done” does not replace them. Cloud/model usage is billed through your configured accounts.
The 2026-11-20 first-Session record provides an actual scoped result for the linked programs:
Both used Vercel Sandbox, Pi 1.5.2 and the configured Messages endpoint with
deepseek-ai/DeepSeek-V4-Flash. This dated result covers the first task and retained artifact; it does not establish Tools, approvals, preview or other combinations. Original resource IDs stay in private journals. The complete example is the runnable entry point; copying only the creation excerpt leaves an open Session to manage.
Choose your next task
The complete project tutorial covers upload, fixed tests, private preview, a second modification and retained exports; dated independent dual-SDK cloud and browser checks passed for the documented Vercel/Pi/Messages combination. Use cases explains product fit and boundaries.
Understand where work runs
The control plane owns Session/Run state, durable events, credentials and cloud resources. Runtime supervises Harness processes inside the cloud Sandbox. Original provider/model/Tool credentials stay in the control plane; the guest receives scoped access. “Local” and “hosted” describe the control plane’s location, not local versus cloud task execution. See execution architecture and security boundaries.Enter by your responsibility
Coding Agents can use the integration reading guide. Browse the complete documentation directory for all existing pages, including provider guides, concepts, extension guides and reference entries.