Why export
For compliance you need evidence outside the system. For fine-tuning you need your own runs as training data. Either way, the history is yours to take.
Export it
- Open the History view.
- Choose Export (or use the API).
- Pick the format — JSONL is the default.
- Download the file.
What is in the file
Every entry: the agent, the action, the decision, the reason, the tokens, the latency, and the cost. The hash chain is preserved, so you can verify integrity later.
For fine-tuning
Rated runs (the good/bad feedback you leave on answers) make a clean preference dataset. Export, filter to rated runs, and you have signal to train on.