OpenSquilla is a token-efficient microkernel AI agent that routes tasks to the most cost-effective language model for each request.
The tool addresses the challenge of running capable AI agents within constrained token budgets. It uses a local model router to direct each turn to the cheapest model capable of handling it, paired with persistent memory, a layered sandbox, built-in web search, and on-device embeddings. All entry points—Web UI, CLI, and chat channels—flow through the same core loop, ensuring consistent tool dispatch, retries, and decision logging across interfaces. The agent supports multiple LLM providers including TokenRhythm, OpenRouter, OpenAI, Anthropic, Ollama, DeepSeek, Gemini, Qwen/DashScope, and over twenty others through a pluggable provider layer that requires no code or configuration schema changes.
Developers should choose this tool if they need an AI agent that maximizes capability within a fixed token budget and want consistent behavior across CLI, web, and chat interfaces. It suits projects where cost efficiency matters alongside agent performance, and where the ability to swap between multiple LLM providers without code changes is valuable. The tool's microkernel architecture and unified turn loop distinguish it from agents that handle different interfaces separately.
The project maintains active development with continuous integration checks passing. A technical report on agentic routing and multi-model ensemble routing is available, documenting how the router's approach surpasses certain baseline models. Documentation spans product guides, API references, and implementation details across multiple languages.
