OpenPilot
Open-source desktop AI agent for any model you choose
72 followers
Open-source desktop AI agent for any model you choose
72 followers
OpenPilot is a free, MIT-licensed desktop AI agent. Connect any OpenAI-compatible model (OpenAI, OpenRouter, Ollama, LM Studio), pick a folder, and it reads and edits files and runs commands there. You approve sensitive actions. Windows now, macOS early beta.
Free
Launch Team
AWS Startup AdvisorPersonalized AWS guidance that meets you where you’re already building
Promoted
Hi Product Hunt, I'm Ibrahim, the solo developer behind OpenPilot (not related to comma.ai's openpilot).
I wanted the Cursor-style agent experience, where you describe a task and it reads your files, runs commands and fixes things, but open source and running on my own model. I couldn't find an open-source tool that worked the way I wanted, so I built one.
You paste a base URL, key and model name (OpenAI, OpenRouter, or a local Ollama / LM Studio server), pick a folder, and ask for something. Every tool call shows up as a card, and sensitive actions wait for your approval. It also has skills, MCP servers and per-model token usage. It's not an editor or an IDE, just an agent that works in a folder you choose.
Honest status: Windows works well, macOS builds are an early unsigned beta, and there's no Linux build yet. Small local models handle simple tasks but struggle with long multi-step work. Anonymous usage stats are currently on by default (no chats, files or keys) and can be turned off in Settings. It's early and it's just me.
I'd love to hear which models you want it tested with and what breaks on install. (Note: this comment was drafted with AI assistance and posted from the OpenPilot team account.)
Direct download for Windows (no account needed): https://github.com/ibmshaikh/OpenPilot/releases/latest/download/OpenPilot-Setup.exe
Windows may show a SmartScreen warning because the build isn't code-signed yet. Click "More info" then "Run anyway". Mac builds are an unsigned beta for now.
respect the honesty about small local models struggling with long multi-step work, most launches bury that in the fine print. is there a concrete signal that tells a user they've crossed that line, like the model repeating a tool call or losing track of the folder state, or does it just quietly produce worse edits and you only notice after reviewing the diff. a built-in "this model is probably too small for this task" warning would be a genuinely useful feature on top of what you've got
Thanks Gal, really appreciate that. Honest answer: right now there's no automatic warning. The clearest signs are the ones you named: the same tool call repeating, re-reading files it already read, or edits that drift from what you asked. Since every tool call shows up as a card, you usually spot it in the trail before the diff, but that still depends on you watching. A 'this model is probably struggling' warning based on repeated calls and loops is a great idea, and I'm adding it to the list. Thanks for the push!
@ibmshaikh that's a fair tradeoff honestly, a card-based trail you can scan is probably more trustworthy than an auto-warning you'd start ignoring after the third false positive anyway. curious if you've thought about rate limiting retries at the agent level too, or is that something you're leaving to the underlying model's own loop control
@galdayan Good question — right now it's on the model's own loop control. OpenPilot doesn't step in between the model and its retries, so a stuck small model can burn tokens repeating the same call until you stop it. The card trail is the only guardrail today.
An agent-level retry cap is going near the top of the list now: catch the same tool call repeating, stop the loop, and show a 'model looks stuck — here's what it tried' card instead of a noisy diff. Appreciate you pushing on this — exactly the kind of thing worth building early.
Good catch — we're treating repeated edits to the same file or function as one loop, not just identical calls.