OpenPilot
Claim this toolOpen-source AI agent that works with your own LLM, no vendor lock-in.
Free, no paid tierVisit WebsiteFree Visit Website
Tracked since 2026
The Bottom Line
- Price
- Free, no paid tier. Compare plans
Key facts
- Open-source AI agent that runs locally on your machine with your own LLM.
- Provides filesystem, shell, web research, and MCP tools for real project work.
- No vendor lock-in, no account required, and full control over models and data.
Pros
- Fully open-source (MIT license) with no vendor lock-in
- Runs locally, keeping your API keys and data private
- Flexible model choice: use any OpenAI-compatible endpoint, including local LLMs
Cons
- Currently only available for Windows (macOS version coming soon)
- Requires you to provide your own LLM API key, which may incur costs
What is OpenPilot?
OpenPilot is a local desktop application that acts as an AI agent harness. It provides the agent with filesystem tools (create, read, edit, grep, glob), a live shell for running real terminal commands, web research capabilities via TinyFish, and support for the Model Context Protocol (MCP) to connect external systems. Users bring their own OpenAI-compatible API endpoint and key, allowing them to use any model from OpenAI, OpenRouter, local servers like Ollama or LM Studio, or custom endpoints. The agent can scaffold projects, run tests, install dependencies, and perform research, all while keeping data and keys local. It is MIT licensed and currently available for Windows, with macOS builds coming soon.
Louis Corneloup Source openpilot.in ↗
Key Features
- Create, read, edit, grep, and glob files across a workspace
- Run real terminal commands (install deps, start servers, run tests) with stdout streaming
- Live web research using TinyFish (search and fetch pages with cited answers)
- Skills and long-term memory via AGENTS.md files that the agent reads and updates
- Model Context Protocol (MCP) support for connecting databases, browsers, internal APIs
- Bring your own OpenAI-compatible model (cloud or local) with custom base URL and API key
- Session and lifetime token usage tracking in-app
- Workspace-based isolation: point at a folder and the agent gets filesystem and shell access there
Pricing
Free
OpenPilot is completely free to use with no hidden costs.
Reviews
$99 Free with your review
Write a reviewReview OpenPilot, get a free AI guide
Share your experience and we will send you Improve Your Thinking Patterns Using ChatGPT, free.
Best OpenPilot Alternatives
Top alternatives based on features, pricing, and user needs.
Cursor Freemium
AI code editor with advanced models and flexible pricing
Cognition AI Paid
Building AI agents that can reason and learn to solve complex problems.
Devin Paid
AI software engineer by Cognition Labs
Refact AI Freemium
Your open-source, autonomous AI agent that codes, thinks, and adapts to your workflow.
Magic AI Paid
Automating software engineering and AI research with frontier code models for safe AGI.
Kodezi Freemium
AI CTO for codebases that autonomously finds bugs, applies fixes, and updates documentation.
Indent Freemium
Software engineering agents that work smoothly with you and with each other, made for teams who move fast.
Still deciding?
Most buyers shortlist 2 or 3 tools before committing. Pull a side-by-side comparison or browse the full alternatives shortlist below.
Explore More
OpenPilot FAQ
Can I use OpenPilot with a local LLM like Ollama or LM Studio?
Yes. OpenPilot works with any OpenAI-compatible endpoint, including local servers. You simply set the base URL to your local address (e.g., localhost:11434) and provide the model ID. No internet connection is required for the model itself, though web research features will need internet access.
Does OpenPilot require an internet connection to function?
Not for core agent operations if you use a local LLM. The agent can create files, run shell commands, and use MCP tools offline. However, the live web research feature (TinyFish) requires internet access to search and fetch pages. If you use a cloud-based LLM endpoint, internet is needed for API calls.
How does OpenPilot handle file system access and security?
OpenPilot operates within a workspace folder you specify. The agent can read, write, edit, and delete files only inside that workspace. It does not have access to files outside the chosen directory. The shell commands also run within the workspace context. You control which folder the agent can see and modify.
What is TinyFish and how does it work?
TinyFish is OpenPilot's built-in web research tool. It allows the agent to search the web and fetch page content in real time. The agent uses the retrieved information to ground its answers and cite sources. You can invoke it with commands like web_search() and web_fetch() in the chat.
Can I extend OpenPilot with custom tools or integrations?
Yes. OpenPilot supports the Model Context Protocol (MCP). You can define MCP servers in an mcp.json file and enable them per chat session. This lets you connect databases, browsers, internal APIs, or any other system that implements the MCP standard.
Does OpenPilot store any data on its servers or require an account?
No. OpenPilot is a local application with no server-side component. It does not require you to create an account or log in to any vendor service. All data, including API keys and workspace files, stays on your machine. Token usage tracking is done locally in the app.
How do I set up skills and long-term memory in OpenPilot?
You create an AGENTS.md file in your workspace. In that file, you can write reusable playbooks, preferences, and rules (e.g., 'Prefer TypeScript strict', 'Use pnpm in this repo'). The agent reads this file on startup and can update it when you ask. This provides persistent memory across sessions.
Is there a macOS version of OpenPilot?
A macOS version is under development and will be released soon for both Apple silicon and Intel Macs. Currently, only a Windows 64-bit installer and portable executable are available. The macOS builds are listed as 'Coming soon' on the website.
Source: openpilot.in