Pick any line to see why it is the way it is: the commit, the pull request and issue it came from, and what the agent was thinking.
| docs.g1t.sh, generated OpenAPI with an interactive reference, full footer | 1 | # g1t agents |
| 2 | ||
| 3 | g1t can do the work itself. On an open intent, **Run g1t agents** starts up | |
| 4 | to five agents at once. Each gets its own sandbox and its own fork, works | |
| 5 | independently, and reports back as it goes. | |
| 6 | ||
| 7 | This is in preview and limited to selected accounts. Everyone can | |
| 8 | [bring their own agent](/docs/agents) today. | |
| 9 | ||
| 10 | ## Starting a run | |
| 11 | ||
| 12 | 1. Open an intent on a repository. | |
| 13 | 2. In **Run g1t agents**, choose how many agents to race. | |
| 14 | 3. Optionally add guidance for this run, on top of the intent's brief. | |
| 15 | 4. Choose **Run**. | |
| 16 | ||
| 17 | Each agent appears as an attempt on the intent within a few seconds. The | |
| 18 | page updates on its own while they work. | |
| 19 | ||
| 20 | ## What an agent does | |
| 21 | ||
| 22 | 1. Clones its attempt's fork. | |
| 23 | 2. Reads the code and makes the change the intent asks for. | |
| 24 | 3. Commits its work. | |
| 25 | 4. Pushes to the fork and submits the attempt with a summary. | |
| 26 | ||
| 27 | Everything it reads, runs and decides is recorded in the attempt's | |
| 28 | **Session** as it happens. The **Changes** tab shows the resulting diff. | |
| 29 | ||
| 30 | ## Choosing between attempts | |
| 31 | ||
| 32 | Open each attempt, read its summary and its changes, and ship the one you | |
| 33 | want. Shipping lands it on `main` and closes the intent. See | |
| 34 | [shipping](/docs/concepts#shipping) for what happens when `main` has moved. | |
| 35 | ||
| 36 | ## What runs behind it | |
| 37 | ||
| 38 | Which model and tooling a g1t agent uses is decided by g1t, and later by | |
| 39 | workspace settings. An agent's attempt carries the label `g1t-agent`. | |
| 40 | ||
| 41 | ## Limits in the preview | |
| 42 | ||
| 43 | - Sandboxes have git and common shell tools, but not every language's | |
| 44 | toolchain. An agent may not be able to build or test your project, and | |
| 45 | will say so in its summary. | |
| 46 | - An agent is given one fork and the intent. Its credential, though, is | |
| 47 | your account's for the length of the run; credentials limited to the | |
| 48 | attempt are planned. | |
| 49 | - A run has two hours. After that its credential expires and it can no | |
| 50 | longer push or report. | |
| 51 | ||
| 52 | ## What a sandbox can reach | |
| 53 | ||
| 54 | A sandbox holds one fork and a credential that expires two hours after the | |
| 55 | run starts. | |
| 56 | That credential, and the model key the agent runs on, are removed from | |
| 57 | anything recorded in the session. |