600 Stars. One Email. OpenAI Just Hired the Guy Who Built a Better Prompt Tool.

If you’re building AI tools, here’s a story you need to know.

An indie developer named Eric Provencher built a macOS utility called RepoPrompt. A few months ago, Romain Huet, Head of Developer Experience at OpenAI, reached out directly and invited him to join OpenAI. Provencher said yes, but on one condition: he wanted to settle his existing paying users first, make the product free, and open-source the entire codebase.

Now the community edition (RepoPrompt CE) is live on GitHub: Apache 2.0 license, nearly 500 stars, ready to use out of the box.

This wasn’t an acquisition. It was a talent grab, but one that’s more powerful than any funding round. By bringing him in, OpenAI was saying with its actions: What you built is exactly what we need.

What does this tool actually do?

In one sentence: It solves the shared pain of everyone coding with AI – when it comes to the code you feed an AI, less is more.

If you use Cursor, Claude, or ChatGPT for coding, you’ve probably done this:

• Dragged your entire project folder into the chat and said “fix this bug”

• Got back a wall of text that changed things it shouldn’t have

• Watched the model start hallucinating code that doesn’t exist

The reason is simple: you’re drowning the AI in too much information. AI models don’t get smarter with more data; past a certain threshold (say, 32K tokens), irrelevant context overwhelms the signal, and the model actually performs worse.

RepoPrompt does one thing: it helps you precisely select which files from your repo the AI actually needs to see, and assembles them into a high‑quality prompt.

Silicon Valley has a name for this now: Context Engineering.

What’s different about the open‑source version?

The original RepoPrompt was a straightforward “pick files, build a prompt” tool. The community edition flipped the architecture on its head: instead of the app managing agents itself, a built‑in MCP Server sits at the core, with different Agent executors you can plug in underneath.

In plain English: you can use a strong reasoning model like GPT‑4o or o3 to plan and break down tasks, then farm out the sub‑tasks in parallel to different agents: Claude Code handles the frontend, Codex deals with the backend, Gemini CLI runs the tests… Each agent only sees its own set of files, with zero interference.

A smart “brain” plus a team of obedient “hands and feet” is the next shape of AI‑assisted coding.

What can developers and founders learn?

Three takeaways from this story:

1️⃣ Pain Points Hide in the Details
RepoPrompt started by doing one thing: selecting files and stitching together a prompt. It sounds trivial to the point of being unremarkable, but it was a real, everyday pain for anyone coding with AI.
Small and sharp beats big and vague – that’s what gets a giant’s attention.

2️⃣ Architecture Needs Foresight
Flipping from a single app to “MCP Server + pluggable agents” landed the project right at the intersection of the Model Context Protocol (MCP) wave. Anthropic is pushing MCP, OpenAI is adopting it, and RepoPrompt sits in the exact middle layer that both sides need.
Pick an infrastructure‑level lane, and giants can’t work around you.

3️⃣ Founder Decency as a Strategy
Before Provencher was hired away, the first thing he did was take care of his users – free, open‑source, no loose ends. That move earned him serious goodwill in the community. In Silicon Valley, people sometimes say “nice guys finish last,” but this case shows that decency itself is a brand asset. The contributors on that GitHub repo today? Mostly former paying users. He walked into his new role with an open‑source team already in place.

How to try it yourself

Currently macOS only, with cross‑platform support in development. The easiest way (Homebrew):

bash

brew trust repoprompt/repoprompt-ce
brew tap repoprompt/repoprompt-ce
brew install --cask repoprompt-ce

Once installed, it just works. You can also build from source – no Xcode required.

GitHub repo:
👉 github.com/repoprompt/repoprompt-ce

Final thoughts

RepoPrompt isn’t a world‑changing product. But it lands squarely on a need that’s only going to grow as AI‑powered coding becomes the norm.

In the age of AI, a tool’s value isn’t measured by what it can do on its own – it’s measured by what it enables AI to do for you. And OpenAI’s “talent grab” just validated that thesis in the clearest possible way.

If you’re building AI dev tools, or just writing code with AI – RepoPrompt CE is worth ten minutes of your time. Install it, try it, and see.