LLM Models — Trading Agent

← Settings · Each pipeline role calls a model independently. The Intelligence column describes what each role actually has to think about — match the model tier to the job: cheap/fast for the parallel analysts, deeper reasoning for the research manager and reviewer.

Role	Intelligence needed	Provider	Model	Thinking	Effort	Max tokens	Temp
News	Fast / lightweight High-volume screening of headlines & filings; speed and breadth over depth. Suggested: Haiku / Gemini Flash class
Sentiment	Fast / lightweight Classifies tone across many short texts; cheap and parallel. Suggested: Haiku / Gemini Flash class
Technical	Fast / lightweight Reads price/indicator structure into a stance; pattern over prose. Suggested: Haiku / Gemini Flash class
Fundamental	Fast / lightweight Summarises financials & ratios into a view; structured, repetitive. Suggested: Haiku / Gemini Flash class
Bull	Balanced reasoning Builds the strongest adversarial long case; needs real argumentation. Suggested: Sonnet / Gemini Pro class
Bear	Balanced reasoning Builds the strongest adversarial short case; needs real argumentation. Suggested: Sonnet / Gemini Pro class
Research Manager	Deep reasoning Adjudicates bull vs bear and sets conviction — the highest-stakes judgment. Suggested: Opus / Gemini Pro (high effort)
Trader	Balanced reasoning Synthesises the debate into a sized, structured thesis with entry/exit. Suggested: Sonnet / Gemini Pro class
Reviewer	Deep reasoning Critiques outcome vs thesis and writes lessons to memory; closes the loop. Suggested: Opus / Gemini Pro (high effort)

Changes take effect immediately for the next run.

Provider keys are set under Settings → Provider keys. A role whose provider has no key (or whose optional SDK is missing) falls back to the deterministic mock model, so the pipeline always runs offline.