Choose the AI provider and model for each action. Auto is a deterministic, versioned routing policy — not another model asked to pick a model. Provider API keys stay on the server and are never shown here. No broker, no real orders: every trade stays a local paper record.
Each action routes via Auto (a deterministic, versioned policy) or a pinned model. Policy version: v1
| Action | Mode | Currently recommends | Explanation | Fallback | Max cost | Providers | Required caps |
|---|---|---|---|---|---|---|---|
| Market scan market.scan | AUTO | gpt-5.6-luna | Recommends GPT-5.6 Luna (gpt-5.6-luna) — quality=0.55, speed=0.9, cost=1, reliability=0.54, total=0.78. | — | — | any | structuredOutput |
| News filter market.news_filter | AUTO | gpt-5.6-luna | Recommends GPT-5.6 Luna (gpt-5.6-luna) — quality=0.55, speed=0.9, cost=1, reliability=0.54, total=0.78. | — | — | any | structuredOutput |
| News summary market.news_summary | AUTO | gpt-5.6-luna | Recommends GPT-5.6 Luna (gpt-5.6-luna) — quality=0.55, speed=0.9, cost=1, reliability=0.54, total=0.78. | — | — | any | — |
| Memory classification memory.classify | AUTO | gpt-5.6-luna | Recommends GPT-5.6 Luna (gpt-5.6-luna) — quality=0.55, speed=0.9, cost=1, reliability=0.54, total=0.78. | — | — | any | structuredOutput |
| Memory summarize memory.summarize | AUTO | gpt-5.6-luna | Recommends GPT-5.6 Luna (gpt-5.6-luna) — quality=0.55, speed=0.9, cost=1, reliability=0.54, total=0.78. | — | — | any | — |
| Company deep analysis company.deep_analysis | AUTO | gpt-5.6-terra | Recommends GPT-5.6 Terra (gpt-5.6-terra) — quality=0.85, speed=0.7, cost=1, reliability=0.54, total=0.794. | — | — | any | structuredOutput, reasoning |
| Portfolio analysis portfolio.analyze | AUTO | gpt-5.6-terra | Recommends GPT-5.6 Terra (gpt-5.6-terra) — quality=0.85, speed=0.7, cost=1, reliability=0.54, total=0.794. | — | — | any | structuredOutput, reasoning |
| Paper trade candidates paper_trade.generate_candidates execution | AUTO | Fail closed | No eligible model for this action after applying hard filters. | 2 | — | any | structuredOutput, reasoning |
| Paper trade final recommendation paper_trade.final_recommendation execution | AUTO | Fail closed | No eligible model for this action after applying hard filters. | 2 | — | any | structuredOutput, reasoning |
| Strategy Lab training recommendation strategy_lab.training_recommendation | AUTO | gpt-5.6-terra | Recommends GPT-5.6 Terra (gpt-5.6-terra) — quality=0.85, speed=0.7, cost=0.874, reliability=0.54, total=0.774. | — | — | any | structuredOutput |
| Strategy reflection strategy.reflect | AUTO | gpt-5.6-terra | Recommends GPT-5.6 Terra (gpt-5.6-terra) — quality=0.85, speed=0.7, cost=1, reliability=0.54, total=0.794. | — | — | any | structuredOutput, reasoning |
| Propose rule change strategy.propose_rule_change | AUTO | gpt-5.6-terra | Recommends GPT-5.6 Terra (gpt-5.6-terra) — quality=0.85, speed=0.7, cost=1, reliability=0.54, total=0.794. | — | — | any | structuredOutput, reasoning |
| Validate rule strategy.validate_rule | AUTO | gpt-5.6-terra | Recommends GPT-5.6 Terra (gpt-5.6-terra) — quality=0.85, speed=0.7, cost=1, reliability=0.54, total=0.794. | — | — | any | structuredOutput, reasoning |
| Compare challenger strategy.compare_challenger | AUTO | gpt-5.6-terra | Recommends GPT-5.6 Terra (gpt-5.6-terra) — quality=0.85, speed=0.7, cost=1, reliability=0.54, total=0.794. | — | — | any | structuredOutput, reasoning |
| Report explanation report.generate_explanation | AUTO | gpt-5.6-luna | Recommends GPT-5.6 Luna (gpt-5.6-luna) — quality=0.55, speed=0.9, cost=1, reliability=0.54, total=0.76. | — | — | any | — |
See which model Auto would pick — without calling a model or creating any trade.
When starting a manual analysis run you can pick one of these. Runs that use an experiment override are recorded with their own immutable config snapshot and classified EXPERIMENT/SHADOW, so their statistics never mix into a qualified Champion.