A project is a lasting home for work. A run is one task performed inside it.
Resource model
| Resource | Purpose | Example |
|---|---|---|
| Organization | Shares membership, permissions, credits, and limits | Your team |
| Project | Groups related files and workspaces | A research repository |
| Workspace | Owns an independent working folder and Git branch | A feature branch |
| Session | Keeps a compatible native agent conversation | An ongoing investigation |
| Run | Executes one prompt with a deadline and budget | Update the report |
| Checkpoint | Records a verified recoverable file revision | The files after a completed run |
| Connection | Grants access to a provider or tool | An MCP server or model key |
| Agent preset | Reuses harness, model, and run configuration | A code-review setup |
What persists
Persistent project files and compatible native session state survive ordinary run completion. Checkpoints provide verified recovery points. Detailed output and tool history have a separate plan retention policy. GitHub synchronization adds a remote version-control copy; it does not replace checkpoint persistence.
A lost sandbox can lose changes since the last published checkpoint. Saved checkpoints, current file revisions, and Git status are visible separately so you can identify what is recoverable.
Parallel work
Only one active writer can modify a workspace. Create another workspace for parallel tasks. Organization concurrency limits apply across workspaces and projects; available global capacity also affects when a run starts.
Three independent clocks
- Queue deadline: how long a run may wait before starting. New jobs default to 24 hours; requests can choose a shorter deadline.
- Execution timeout: how long the agent may run after starting, bounded by the organization's plan.
- History retention: how long detailed output remains available. Queue expiry does not delete the run's status or accounting record.
An accepted request is not a completed task
The API returns a run ID when it accepts work. Follow the status, stream events, or retrieve the result until it is final. Cancellation requests stop future work and preserve recoverable files; they cannot reverse actions already taken by external tools.
Read runs and agents, workspaces, and billing for the full behavior.