Files
crewAI/lib
Devin AI affe5709c1 fix: capture thought output from Gemini thinking models (issue #4647)
- Add thinking_config parameter to GeminiCompletion.__init__ (accepts dict or ThinkingConfig)
- Include thinking_config in _prepare_generation_config when set
- Rewrite _process_stream_chunk to iterate over parts directly instead of using chunk.text, avoiding warnings when non-text parts (thought, function_call) are present
- Convert _extract_text_from_response from staticmethod to instance method; separate thought parts from text parts and store thoughts in self.previous_thoughts
- Add 11 tests covering thinking config initialization, generation config integration, thought part extraction in streaming and non-streaming paths

Co-Authored-By: João <joao@crewai.com>
2026-02-28 12:14:10 +00:00
..
2026-02-27 09:44:47 -05:00