Open source project OpenSquilla: intelligent routing and local retrieval, significantly reducing LLM usage costs

AIMPACT News, May 14 (UTC+8), the open-source project OpenSquilla proposes a solution to the high token consumption issue in large language model applications by combining intelligent model routing with local vector retrieval. The system can automatically assess task complexity, routing simple questions to inexpensive models, while assigning more complex tasks to more powerful models, with routing decisions made locally, avoiding token consumption. Through incremental sending and cache hit mechanisms, actual token transmission has been reduced by over 90%. Its memory system can automatically filter and compress key information when the context is full, supporting hybrid retrieval. The project also features cost statistics, a security sandbox, support for one-click migration with OpenClaw, and scheduled tasks, significantly improving efficiency and cost-effectiveness. (Source: AiHot)