The dev harness for coding agents

Coding agents are fast. They're also new to your project every single time.

Each session starts from nothing. The agent guesses at your stack, reaches for a version of the library from two years ago, wires up payments a way the docs stopped recommending, and forgets what it finished yesterday. So you write a rules file, paste in the same instructions again, and review the same mistakes again.

Today we're launching gyve to fix that. One MCP connection hands your agent your stack, the patterns to follow, current docs and the next step. Then it remembers what got done, for the next session and the next teammate.

Why we built it

We build products with agents every day, and the code they write is only as good as the context they start with. A rules file helps, but it's one repository's notes, written once and slowly going stale. It doesn't know that a library changed its advice last month, what order a project should be built in, or what your teammate's agent already did this morning.

We wanted the agent to start every session already knowing those things. So we wrote them down, kept them current, and put them where any agent can reach them.

How it works

1. Start a project on a path. A path is the route most projects of a kind follow, written as ordered steps, each with a check for when it's done.
2. Connect your agent. Make a key on the Connect page and add gyve to Claude Code, Cursor, VS Code, the Claude API or your own agent. It's one MCP server and one key.
3. Let it work. The agent asks gyve for the project's stack and the next step's instructions, does the work in your code, and records what it finished with a note for whoever comes next.

How we're different

Kits that are checked, not scraped. The catalog covers frameworks like FastAPI, Django, Rails, Laravel, Phoenix, Go, Spring Boot, ASP.NET Core and Next.js, services like Stripe and Claude, MCP servers, and architecture patterns. Each kit has setup steps, the patterns to follow, the mistakes to avoid and open-source projects that show it done well, and each is dated with when it was last checked against its own docs.

A plan, not just a prompt. Paths give your agent an order to work in and a way to know each step is really done. If your project doesn't fit a path, your agent still gets each kit's guide, and can add kits as the project grows.

Shared memory for the whole team. Progress belongs to the project, not to one chat. Every agent and every teammate sees which steps are done, who did them and their notes, and any step can be reopened.

It never sees your code. Your agent runs where it always does and works on your code there. gyve only holds your projects, their progress and the notes agents leave. Every tool call is recorded in your audit log.

You pay for people, not agents. Agents connect with keys and don't take seats, so running more of them never costs you more seats.

Try it today

Every plan starts with a 14-day free trial, with no card needed, for up to three people on Team. Make a key, add gyve to your agent, and your next session starts already knowing the plan.

See the plans, or see how it works. Questions? Email us at hello@gyve.dev. We read every message.