[aarshx05]

Description

The chatbot assumes an LLM reply's content is a str. Newer models (e.g. recent Anthropic Claude models) return content as a list of structured content blocks (text / thinking / tool-use / signed text). In process_user_message this value is stored in the Mongo chat history and passed to add_to_chroma_collection → Document(page_content=...), which requires a str. A list therefore raises pydantic.ValidationError; the exception is caught in /genai/ask and the raw error is returned to the user and the turn is silently never persisted (chat appears broken and RAG/history storage fails).

This bugfix adds a small message_content_to_text() helper in agent_utils.py that normalizes any provider's content to plain text (joins text blocks, drops thinking/tool-use blocks, coerces other types safely) and uses it in process_user_message before the value is stored, embedded, or returned. No behavior change for providers that already return a string (e.g. OpenAI, older models) — those pass through untouched.

Reproduction: run the chatbot with CHATBOT_LLM_PROVIDER=anthropic and a current Claude model and send a message that elicits a multi-block/signed reply. Before this change the bot returns 1 validation error for Document page_content Input should be a valid string [...]; after, it replies normally and the turn persists.

Testing

• Added TestMessageContentToText in services/chatbot/src/chatbot/tests/test_agent_utils.py covering: plain-string passthrough, a list of signed text blocks (the crash case), skipping thinking/tool_use blocks, joining multiple and bare-string blocks, None, non-str/list coercion, and the "always returns str" contract.
• cd services/chatbot/src && python -m pytest chatbot/tests/test_agent_utils.py → all pass.
• Deployed locally with the Anthropic provider + a current Claude model: chat replies and history/RAG persistence now work where they previously raised the pydantic error.

Documentation

No documentation changes required — internal robustness fix, no public API/behavior change.

Checklist:

• My changes generate no new warnings
• I have added tests that prove my fix is effective
• Any dependent changes have been merged
• I have documented any changes if required in the docs (n/a)