Repolode
Submit
TokenRhythm

TokenRhythm/opensquilla

A token-efficient, microkernel AI agent for CLI, Web UI and chat channels that routes each turn to the cheapest capable model.

Open-source alternative to:ChatGPT

View on GitHubWebsite
Stars
7,081
Forks
583
Open issues
271
Last commit
2 hours ago

What is it?

OpenSquilla is an AI agent that runs locally and talks to many LLM providers. A local model router called SquillaRouter classifies each turn on length, language, code, keywords and embeddings, then sends it to the cheapest model that can handle it. Classification runs on-device, so the prompt does not leave the machine for that decision.

Web UI, CLI and chat channels such as Telegram, Slack, Discord, Feishu, DingTalk, WeCom, Matrix and QQ all run through the same turn loop, so tool dispatch, retries and logging behave the same everywhere. Persistent memory uses a curated MEMORY.md plus SQLite full-text and sqlite-vec semantic search, and tools are sandboxed with Bubblewrap on Linux, Seatbelt on macOS, or a native Windows backend.

The project also includes on-demand skills, MCP client and server support, SQLite-backed sessions, a cron-style scheduler, and migration commands for importing state from OpenClaw or Hermes Agent. It is distributed as a Python package with desktop installers for macOS and Windows.

Key features

  • 01Routes each turn to the cheapest capable model using a local LightGBM and ONNX classifier.
  • 02Supports 20+ LLM providers, including OpenAI, Anthropic, Ollama, OpenRouter and DeepSeek.
  • 03Offers Web UI, CLI and chat channels that share one turn loop.
  • 04Keeps persistent local memory searchable by keyword and semantic embeddings.
  • 05Runs tools in a layered sandbox with Standard, Strict and Locked policy tiers.
  • 06Includes built-in web search, file and shell tools, and MCP client and server support.

Use cases

  • Cutting LLM API spend by sending simple requests to cheaper models.
  • Running a self-hosted personal assistant that answers over chat apps like Telegram or Slack.
  • Exposing agent sessions to another MCP client for automated workflows.
  • Migrating existing OpenClaw or Hermes Agent setups to a new local agent.

Quick start

  1. 1

    Install uv (Linux/macOS: run the official install script from the README)

  2. 2

    Install OpenSquilla with uv from the release wheel

    uv tool install --python 3.12 "opensquilla[recommended] @ https://github.com/TokenRhythm/opensquilla/releases/download/v0.5.6/opensquilla-0.5.6-py3-none-any.whl"
  3. 3

    Configure OpenSquilla

    opensquilla onboard
  4. 4

    Start the gateway

    opensquilla gateway run

Things to know

  • The default install pulls in heavy router dependencies such as ONNX Runtime and LightGBM, so the first install needs a sizable download.
  • On Windows and macOS, the router may need native system libraries (Visual C++ Redistributable or libomp) that the quick terminal install does not provide.
  • Building from source requires Git LFS and Node.js 22.12+ with npm.

Topics

#agent#llm#ai-agents#mcp#llm-routing#memory#sandbox#chatbot#ai#deep-learning

FAQ

What is opensquilla?
A token-efficient, microkernel AI agent for CLI, Web UI and chat channels that routes each turn to the cheapest capable model. OpenSquilla is an AI agent that runs locally and talks to many LLM providers. A local model router called SquillaRouter classifies each turn on length, language, code, keywords and embeddings, then sends it to the cheapest model that can handle it. Classification runs on-device, so the prompt does not leave the machine for that decision.
Is opensquilla free to use? What license does it have?
opensquilla is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is opensquilla still maintained?
Actively maintained. The last commit to opensquilla was on Oct 22, 2026.
What can I use opensquilla for?
Cutting LLM API spend by sending simple requests to cheaper models. Running a self-hosted personal assistant that answers over chat apps like Telegram or Slack. Exposing agent sessions to another MCP client for automated workflows. Migrating existing OpenClaw or Hermes Agent setups to a new local agent.
How do I install opensquilla?
Install uv (Linux/macOS: run the official install script from the README) Install OpenSquilla with uv from the release wheel: uv tool install --python 3.12 "opensquilla[recommended] @ https://github.com/TokenRhythm/opensquilla/releases/download/v0.5.6/opensquilla-0.5.6-py3-none-any.whl" Configure OpenSquilla: opensquilla onboard Start the gateway: opensquilla gateway run
Where can I get opensquilla?
The source code and installation instructions are on GitHub: https://github.com/TokenRhythm/opensquilla

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects