6 Commits

Author	SHA1	Message	Date
Norbert	c6232d262d	Fix: pass tool_call_id through to watsonx API ... watsonx requires tool_call_id on role=tool messages. Added field to ChatMessage model and passthrough in transformer.	2026-02-23 18:46:46 +00:00
Norbert	592b34cff0	Fix: accept array content format in ChatMessage + add extra=allow, update .gitignore	2026-02-23 18:40:28 +00:00
Michael	15c650f942	Allow extra fields in ChatCompletionRequest to fix 422 errors ... - Add model_config with extra='allow' to ChatCompletionRequest - Fixes 422 validation errors when OpenAI-compatible clients send unknown fields - Maintains backward compatibility with all existing fields - Tested with unknown_field and extra_param - request succeeds	2026-02-23 13:03:09 -05:00
Michael	27d9bfbad8	Add support for 'developer' role in chat messages ... - Add 'developer' to accepted role literals in ChatMessage model - Existing normalization in chat.py converts developer → system before watsonx API call - Fixes 422 validation errors when clients send developer role messages - Maintains backward compatibility with all existing roles	2026-02-23 12:50:23 -05:00
Michael	d924b7c45f	Add vLLM message normalization for OpenAI content format compatibility ... - Normalize 'developer' role to 'system' (vLLM doesn't support developer role) - Flatten array content to string for text-only messages - Preserve mixed content (text + images) as array - Add comprehensive unit tests for normalization logic Fixes HTTP 422 errors when clients send OpenAI multi-content format	2026-02-23 11:59:23 -05:00
Michael	2e2b817435	Add AGENTS.md documentation for AI agent guidance	2026-02-23 09:59:52 -05:00