Local Models

Local models

browser_run_goal hands routine multi-step navigation (fill a form, pick options) to a small local model that picks one step at a time from a compact table of the page. Your agent supervises.

• Qwen (via Ollama) decides by default. With --decider cascade, Laya (via Ollaya) decides first and Qwen takes over when Laya is unsure.
• Every step says which model decided, with its confidence. The run hands back to your agent when the models are unsure.
• It only types values you pass in, never navigates to other sites, and goes through the same approval gate as every other action.
• After a run, the agent checks the result and calls browser_goal_feedback (good or bad). That feeds the scorecard in knowledge_list (traces.byDecider) and the Dashboard's Local models tab.
• Record me and good runs can be exported as training data: knowledge_export_dataset, then docs/FINETUNING.md. No model has been fine-tuned on them yet (issue #9).

Safety notes: Security-and-Approvals.