DocsTrain
Running a training run
Set a spending limit, choose what to work on and follow a training run in Spark Loops.
A training run tries changes to your agent’s instructions and tests them on examples. You review the results before approving an upgrade.
Before you start
- Choose one agent and a clear job.
- Add real examples of good and bad work.
- Choose AI reviewers from different model providers.
- Set a spending limit and a small round limit.
What to work on
You can add a short note about the weakness you want to improve. For example, focus on unclear replies or missed details.
Follow the run
The app shows the training progress, new versions and estimated costs. Read the results in Reading results.
You pay your model providers directly. Estimated costs use the provider prices configured in the app. Training stops at the spending limit or round limit you set. It can also stop if a problem is detected.
Start small. A run can take a while, and it may leave your current instructions unchanged.
Chat training
Grok Bot chat training and connections to outside helpers are coming soon. For now, start inside the desktop app.