Agent Pilot

Agent Pilot is a desktop AI workflow builder for people who want to combine model conversations with tools and code. It runs on Windows, Linux and Intel Macs and is open source under AGPL-3.0. You can chat with one model or build a graph of cooperating agents, with human input included where a task needs it.

Its model connections use LiteLLM, with support for local LLM backends such as Ollama and vLLM, custom API servers, and cloud providers including OpenAI, Anthropic and Gemini. The desktop app runs on your machine; choosing a cloud provider sends model requests to that service.

Workflows can combine prompts, text, code and other workflows, including steps that run in parallel. You can save a whole workflow as an agent and reuse it inside another. Reusable blocks hold shared prompts or logic, while agent tools can execute code or invoke a complete workflow. This makes it useful for tasks that need several linked steps rather than a single model response.

Chats support branching: you can edit a message, tool input or code and rerun it to explore a different result. Saved conversations and agents can be organized into folders. Open Interpreter handles code execution, and Instructor supports structured model output for tasks that need data in a defined format.

Developers can extend agent behavior through plugins, import custom addons, and use Python modules for persistent functionality. The interface also supports custom pages and editable configuration screens.