OpenSquilla
OpenSquilla is an open-source AI agent framework that helps developers slash token costs by 60–80% using a local smart router and memory consolidation, with almost no quality loss.
Tool overview
OpenSquilla inserts a locally running model router (built with LightGBM/ONNX) between the agent and LLM providers. It assesses the complexity of each turn, directing simple queries to cheap models and reserving powerful ones for complex reasoning. Together with a 4-tier memory that caches ~80% of tokens, it dramatically reduces token expenditure. In PinchBench testing across 25 tasks, OpenSquilla scored 0.9251 compared to 0.9255 for an agent always using Claude Opus 4.7, while the total cost dropped from $6.233 to $0.688.
Key strengths: massive cost savings without sacrificing quality; local routing preserves prompt privacy; a micro‑kernel architecture (~100 lines of core code) ensures plugin crashes don’t affect the core and makes it easy to swap models or add tools (5 lines of code); one‑click migration from OpenClaw; built‑in channels like Slack, Discord, WhatsApp, as well as Chinese workplace apps; and self‑verification in v0.4.0 with automatic test execution and failure rollback.