This example demonstrates SochDB's Context Query Builder - a powerful feature for assembling LLM context under strict token budgets.

The Context Query Builder assembles the final prompt for an LLM by combining:

• System Message: Instructions and persona for the LLM
• User Query: The current user input
• Conversation History: Recent exchanges
• Retrieved Context: Semantically relevant information from vectorsearch

All of this is done under a token budget with priority-based truncation, ensuring the most important information fits within the LLM's context window.

• Token Budget Management: Automatically fits content within your specified token limit
• Priority-Based Trucation: Keeps the most important content (system + current query), then adds history and retrieval as space allows
• TOON Format Integration: Uses SochDB's to_toon() for compact retrieval formatting
• Flexible Truncation: Intelligently truncates conversation history from the middle, keeping recent and oldest context

• Chatbots: Keep conversation context relevant and within limits
• RAG Systems: Balance retrieved documents with conversation flow
• Long Conversations: Maintain coherence even after hundreds of turns
• Multi-Modal Agents: Coordinate different types of context (text, metadata, tool outputs)

pip install -r requirements.txt
python3 runner.py

The script demonstrates:

1. Building context with ample token budget → All content fits
2. Building context with tight budget → Intelligent truncation
3. Priority ordering → System and query always included first