Model connections

You can edit source and run backtests without an AI connection. Configure a connection when you want the assistant to investigate results or change strategy source.

Connect a model

Open the Agent panel and its connection controls. For the OpenAI connection, enter your Model ID and OpenAI API key, then choose Save connection. Use a model available to your API account.

The key is stored encrypted on the server and is not returned to the browser or included in model context. The key input is cleared after saving. Use Remove to remove the saved OpenAI credential.

Administrators may also see Backend Codex. That connection uses the backend host's existing Codex configuration and login. Select a model from its catalog and save the connection. If loading the catalog fails, retry after the backend configuration is available.

What the assistant receives

The authoring workflow supplies strategy source, selected configuration, saved annotations, and bounded historical results as context. Treat these as content you are sharing with the selected model provider. Do not put credentials in strategy source or prompts.

The Lab's authoring connection is separate from any model provider used inside an L3 strategy. Connecting the assistant does not enable real L3 provider calls during historical backtests; those calls are blocked in that workflow.

Set task limits

Each Send uses the token budget and timeout beside the prompt. Defaults are 100,000 tokens and 30 minutes. Both values must be positive integers; the timeout is in whole minutes.

The token budget counts reported input and output tokens cumulatively across that task, rather than setting a model context-window size. The last model call can exceed the remaining budget. The timeout starts when the queued task begins execution.

Check usage and status in the conversation before starting another task. See working with AI for an effective first prompt.