OctoFriend: A Friendly, Open-Source Coding Assistant

OctoFriend is an open-source coding assistant designed to be small, friendly, and highly effective. Built with a cephalopod-inspired personality, it’s engineered to work seamlessly with any OpenAI-compatible or Anthropic-compatible large language model API, allowing users to switch between models mid-conversation when one gets stuck. This flexibility ensures continuous progress, especially during complex coding tasks. Octo enhances the performance of your primary coding LLM by optionally integrating custom, open-sourced machine learning models trained specifically for handling tool calls and code edit failures. These autofix models are compatible with any coding LLM, making Octo a powerful companion regardless of your preferred AI provider. It’s designed to work well with cutting-edge models like GPT-5, Claude 4, GLM-4.5, and Kimi K2—though it supports nearly any LLM you can connect to. One of Octo’s standout features is its intelligent handling of multi-turn conversations, particularly with advanced reasoning models such as GPT-5 and Claude 4, which may use encrypted or token-optimized responses. Octo carefully manages thinking tokens to maintain high performance and accuracy throughout interactions, giving users a consistently sharp and responsive experience. The developers believe Octo is the most effective multi-LLM tool available for managing complex, multi-step reasoning. Privacy is a core principle: Octo collects zero telemetry. When used with privacy-focused LLM providers—such as Synthetic—the code you write stays entirely under your control. However, Octo also supports integration with OpenAI, Anthropic, and local LLMs running on your own machine, offering full flexibility based on your needs. Octo isn’t just a tool—it’s a collaborator. While it has contributed to writing parts of its own source code, the system is built with a human-first philosophy. It’s meant to be a helpful, approachable assistant rather than a fully autonomous code generator. The default mode encourages user oversight, ensuring you stay in control. For those who prefer a more hands-off approach, Octo offers an “unchained” mode via the --unchained flag, which skips all tool and edit confirmations. Octo automatically looks for configuration files named OCTO.md in the current directory and every parent directory up to your home directory. The first matching file it finds is used, so you can define different rules for different projects or models by placing an OCTO.md in your project folder and another in your home directory. If you prefer not to clutter your home folder, you can place a global rules file at ~/.config/octofriend/OCTO.md. For advanced use cases, Octo can connect directly to MCP (Model-Controlled Protocol) servers to access rich, real-time data. After the first run, a config file is created at ~/.config/octofriend/octofriend.json5. To integrate with an MCP server—such as your Linear workspace—simply add the server’s URL and authentication details to this file, enabling Octo to pull in relevant data without relying on complex shell scripting.