A note from the person who built RepoMesh

I didn't build RepoMesh to sell software.
I built it because I was frustrated.

This wasn't a business plan. It was something I needed for myself—and eventually realized might help other people, too.

I love what we can build with AI. I don't love starting over.

I know plenty of people are vibe coding, building things they never thought they'd be able to make. I'm one of them. I use AI coding agents to work on real software, and I love what they're making possible.

But something kept driving me crazy: every new session, every switch between tools, it felt like I was asking another agent to rediscover the same repository.

Same files. Same architecture. More reading. More waiting. More context consumed before we could even get to the actual work.

I didn't want another coding assistant. I wanted the assistants I already used to stop starting from zero.

Then a random post sent me down a rabbit hole.

I saw someone mention a repository directory in a post. That was it. A passing idea—but it made me curious enough to start researching how coding agents find relevant files, understand symbols, and carry repository context between sessions.

The question wouldn't leave me alone: What if different AI coding agents could ask the same local system about the same codebase?

So I started building it. Not for customers. Not for a launch. For me.

That's how RepoMesh began.

The tool

One codebase brain. Your agents still do the coding.

RepoMesh is a free, local-first repository intelligence layer. You connect a local repository, index it, and let compatible AI coding agents request relevant information through the Model Context Protocol (MCP).

  • Index your repository to make files, symbols, dependencies and likely tests easier to find.
  • Use the agents you already like—including documented integrations for Claude Code, OpenAI Codex and Cursor, plus compatible MCP clients.
  • Watch what actually happens with recorded MCP activity and estimates of context delivered. These are not guarantees of model-token savings.
  • Reuse repository knowledge instead of having each session explore everything from scratch.

RepoMesh doesn't write your app for you or replace your model. It's the context layer underneath the tools you already use.

Why give it away?

Because I built it to solve a problem I had, and I don't think I should have to turn every useful thing into a subscription.

RepoMesh is free. I have no plans to charge for it. If it ends up saving someone else the same frustration, that already makes sharing it worthwhile.

And I'm not going to pretend it's perfect. I'm still learning, finding rough edges and working out how to make the experience better. I don't want to make up impressive numbers or promise that every task will use fewer tokens.

What I do want is honest feedback. Tell me what works. Tell me what breaks. Tell me what you'd change. If you're comfortable sharing what you learn, we can make something more useful together.