A month ago I launched OpenPilot on Product Hunt. It's an open-source desktop AI agent: you connect any OpenAI-compatible model (including local ones in Ollama or LM Studio), pick a folder, and it works there. I built it because I wanted that Cursor-style agent loop on my own model and couldn't find an open-source tool that worked the way I wanted.
I wrote a one-week update on October 15. This is the one-month version, same rule: only numbers from GitHub and Product Hunt, nothing rounded up.
The numbers
| Oct 9 (launch) | Oct 15 (week 1) | Nov 7 (today) | |
|---|---|---|---|
| GitHub stars | 1 | 18 | 21 |
| Forks | 0 | 4 | 5 |
| Windows .exe downloads (all releases, all time) | ~170 | 423 | 605 |
| macOS downloads (dmg + zip) | 6 | 14 | 26 |
| Open issues | 0 | 1 | 1 (still unanswered — on me) |
Product Hunt launch day finished at #9 with about 88 upvotes and 54 followers. Almost no comments from other people on day one. That part hasn't changed much since.
What actually moved
Most of the star growth happened in week one. Stars went 1 → 18 in the first week, then only +3 over the next three weeks. Downloads kept climbing (Windows .exe about +182 since Oct 15), but more slowly — a few installs a day rather than the early post-launch bump.
LICENSE finally landed. The repo now has a real MIT LICENSE file, so GitHub detects it. That unblocked awesome-list pull requests; three are still open and waiting on maintainers. I haven't pinged them.
Writing tutorials beat writing "we're live" posts. The posts that got any conversation were the setup guides and the honest comparison. Announcement posts went quiet fast.
One open issue has sat unanswered since October 12. That's the worst look for a solo project asking for early adopters. It's a detailed Linux install report, and I still owe a reply.
What I got wrong (or delayed)
- I still haven't posted a Show HN. New accounts hit rate limits, and I wanted the first comment in my own words. A month later, that post still isn't up.
- Telemetry is still on by default. I say that plainly every time because the FAQ used to sound like nothing left the machine, and that isn't true.
- No Linux build yet, even though someone already built from source on Linux and filed notes.
- macOS builds are still unsigned. Same for Windows SmartScreen.
What's next
- Answer that open issue.
- Keep shipping small, useful posts (skills, MCP, AGENTS.md) instead of more launch noise.
- A Show HN when the first comment is ready in my own words.
- Signing, Linux packaging, and making telemetry opt-in properly.
If you tried OpenPilot with a local model and it fell over, tell me which model and what broke. Repo: github.com/ibmshaikh/OpenPilot. Site: openpilot.in.
I used an AI assistant to help draft and edit this post. The numbers are from GitHub's API and Product Hunt.
Top comments (2)
Thanks for sharing the unrounded numbers — the contrast between stars stalling at 21 after week one while downloads kept climbing to 605 is a useful reminder that the two measure different kinds of interest. Your point that setup guides and honest comparisons drew conversation while launch posts did not matches what I have seen too. Answering that unanswered Linux issue first seems like the right next step, since it is the most visible signal to early adopters weighing whether to try it.
telemetry still on by default changes the trust story for a desktop agent. what would have to change for opt-in to be the default without making setup harder?