OpenPilot is an open-source, desktop-oriented AI agent whose core positioning is a "bring your own model" local AI harness rather than a hosted model marketplace. It emphasizes no model lock-in and no vendor login required; users can connect to any OpenAI-compatible endpoint and have the agent actually execute tasks within a local workspace.
Purpose OpenPilot is aimed at users who want to use an AI agent to accomplish real work while avoiding vendor lock-in. It connects capabilities such as filesystem tools, a live shell, web research, MCP, skills, and memory with a model of the user's choice, and is used to create files, scaffold projects, execute commands, search for information, and deliver complete projects.
Main capabilities
- File operations: The agent can read, write, edit, grep, and glob files within the workspace; it can be asked to generate a complete project and populate a file tree, then iterate through follow-up instructions.
- Terminal execution: Run real terminal commands, such as installing dependencies, starting servers, and running tests, and stream stdout back into the chat. The official description calls the shell a first-class tool.
- Real-time web research: Through TinyFish, search and fetch pages in real time when documentation, APIs, or error information lie outside the codebase, and answer based on what was actually read.
- Skills and long-term memory: Reusable playbooks can be placed as skills; preferences can be persisted in AGENTS.md. The agent loads them into context and can update memory when the user requests it.
- MCP extensions: Connect to Model Context Protocol servers and enable them per chat to integrate existing systems such as databases, browsers, and internal APIs.
- Bring your own model: Supports any OpenAI-compatible endpoint, including OpenAI, OpenRouter, local servers, etc.; each model can be configured with a custom base URL and API Key, providing a view of session and lifecycle token usage.
How to use Officially, the basic flow is three steps: 1) Add a model, paste the base URL, API Key, and model ID; no account registration with OpenPilot is required; 2) Open a workspace and point it to a folder; the agent gains filesystem and shell access in that directory; 3) Describe the desired result and ask it to build, fix, research, or automate, and you can view the tool call trace while it works.
Use cases Suitable for developers and technical users who want to use an AI agent to scaffold projects, modify codebases, run commands, conduct web research, or automate workflows, and is especially suitable for users who do not want to be locked into a single model vendor and are willing to bring their own API Key.
Limitations and platforms OpenPilot itself is a harness, not a hosted model marketplace; users need to provide an OpenAI-compatible endpoint and API Key themselves. It currently offers a native Windows application, requiring Windows 10 or higher (x64), including a 64-bit installer and a portable executable; Apple silicon and Intel versions for macOS are coming soon. Information about other platforms is not currently provided.
