mirror of
https://github.com/Significant-Gravitas/AutoGPT.git
synced 2026-02-11 23:35:25 -05:00
Adds configuration to enable Anthropic's extended thinking feature via OpenRouter. This keeps the model's chain-of-thought reasoning internal rather than outputting it to users. Configuration: - thinking_enabled: bool (default: True) - thinking_budget_tokens: int (default: 10000) The thinking config is only applied to Anthropic models (detected via model name containing 'anthropic'). Fixes the issue where the CoPilot prompt expects thinking mode but it wasn't enabled on the API side, causing internal reasoning to leak into user-facing responses.