Chatbot

Chatbot

api/views/chatbot_view.py exposes the public chat endpoint used by the frontend widget.

Endpoint

• POST /api/v1/chatbot/

Responsibilities

The chatbot view:

• accepts unauthenticated messages
• resolves the client IP and optional session ID
• builds a system prompt that constrains the assistant to DeepAI OCR support
• pulls recent conversation history for the session
• calls LiteLLMAPIClient and parses the response with ResponseParser
• stores the turn in ChatbotHistory
• returns a reply under multiple keys for frontend compatibility

Prompt and response contract

The system prompt forces a narrow product-support scope and requests a one-sentence JSON response. That prompt is not decorative; the downstream parser and frontend both rely on the compact response shape.

History semantics

If a session_id is present, the last ten turns are loaded in chronological order and appended to the LLM message list. That means conversation continuity is session-based rather than user-account-based.

Dependencies

• api.models.chatbot_history.ChatbotHistory
• api.services.litellm.litellm_api_client.LiteLLMAPIClient
• api.services.litellm.litellm_response_parser.ResponseParser

Validation

• tests/integration/test_chatbot.py
• tests/unit/services/test_litellm_response_parser.py

Scope boundary

This page covers the public chatbot widget only. For the LLM routing service that powers it, read Services.