LLM Callbacks

Overview

The llm.callbacks module implements custom LangChain callback handlers to provide real-time feedback to Discord users during complex AI reasoning and tool execution processes.

ToolFeedbackCallbackHandler

This handler intercepts LangChain's internal tool execution events to update Discord messages, keeping users informed of the bot's background actions.

Core Logic

When a tool starts execution (on_tool_start), the handler: 1. Identifies the tool name (e.g., GoogleSearch, YouTubeSearch). 2. Consults the LanguageManager to find the localized feedback message for that specific tool. 3. Falls back to a generic "Processing [Tool Name]..." message if no specific translation is found. 4. Uses safe_edit_message to update the bot's response in Discord with the status update.

Resilience Features

• Error Suppression: Callback errors are caught and logged but never allowed to interrupt the primary AI reasoning process.
• Async Safety: Uses asyncio.shield to protect Discord message edits from being cancelled if the parent task is interrupted, preventing event loop mismatch errors.

Integration

The callback handler is automatically instantiated and injected into the Orchestrator's agent execution chain.

# Example Internal usage
callbacks = [
    ToolFeedbackCallbackHandler(
        message_edit=message,
        language_manager=lang_manager,
        guild_id=guild_id
    )
]

# Passed to LangChain agent
await agent_executor.ainvoke(input, config={"callbacks": callbacks})

---

By providing granular feedback, this system improves perceived performance and reduces user uncertainty during long-running AI tasks.