The gist

OpenSquilla is an open-source, microkernel AI agent built for token efficiency. It solves the problem of high LLM costs by using a local model router that sends each task to the cheapest capable model. The framework provides a single, unified turn loop with persistent memory, web search, and a layered sandbox, accessible via command-line, a web UI, or integrated chat channels. This offers a consistent agent experience across multiple interfaces while optimizing for performance and cost.

What it does

Routes tasks to the most cost-effective LLM provider using an on-device model router.
Connects to over 20 LLM providers, including OpenAI, Anthropic, and local models via Ollama.
Maintains persistent memory and performs web searches to inform agent actions.
Runs as a command-line tool, a local web UI, or an adapter for chat platforms like Slack and Discord.
Migrates existing configurations and data from OpenClaw and Hermes Agent.

How it works

OpenSquilla is a self-hosted Python application that runs a local gateway. Users interact by providing prompts through a command-line interface, a web UI, or integrated chat channels. A built-in router analyzes each prompt and dispatches it to the most token-efficient model from a list of over 20 configured providers. The tool is open-source and can be installed via desktop installers, terminal commands, or by building from the source code.

Best for

This tool is best for developers and technical users who want to run a cost-efficient, multi-provider AI agent locally. It's ideal for setting up a persistent agent with memory and tool access that can be controlled from the terminal, a browser, or a chat app.

Watch out for

The Windows desktop installer is currently unsigned, which may trigger security warnings from SmartScreen or other policies. Some features, like the advanced terminal UI, are only available when running from a development source checkout, not from the standard installation.