Agent Lab
Agent Lab
Develop strategies with AI, preserve revisions, compare historical runs, and publish a selected version.
Agent Lab brings strategy editing, AI assistance, historical runs, and revision history into one experiment. Start with SJS for your main algorithm; use L3 when you need advanced orchestration.
An experiment belongs to your signed-in user and workspace. Importing an app takes a snapshot of its saved configuration. Changes stay in the experiment until you explicitly publish a revision.
The experiment workflow
New experiment or imported app
↓
Edit source and settings ← AI-assisted changes
↓
Saved revision
↓
Historical backtest
↓
Inspect → annotate → compare → revise
↓
Publish selected revision as an app
↓
Launch separately from the appKnow what you are working with
| Item | Purpose |
|---|---|
| Working draft | Your current editor buffers and strategy settings; edits can be unsaved |
| Revision | An immutable snapshot of strategy source and settings |
| Run | A historical execution attached to a specific revision |
| Baseline | A run you pin for comparison with another run |
| Publication | An explicit operation that saves a selected revision as an app |
Follow the guides
- Getting started: create an experiment and establish a baseline.
- Model connections: configure the assistant.
- Working with AI: give focused tasks and review the changes.
- Revisions: save, inspect, and restore strategy versions.
- Running backtests: select data and inspect a completed run.
- Annotations: attach observations to chart regions and trades.
- Comparing results: evaluate one change against a baseline.
- Publishing: create an app or update an imported app.
- Troubleshooting: resolve conflicts, failed runs, and interrupted work.
Continue after publication
Review Apps & Running Sessions before launch. Use the capability matrix to distinguish historical worker behavior from a running simulated or broker-backed session.