vsuite

I built vsuite because I kept doing the same thing by hand in every repository: writing agent prompts for one assistant, copying them for the next, tweaking them for a third, and watching all three quietly drift apart. A month later nobody, including me, could say which version was the real one.

The problem I wanted to solve

If you use more than one AI coding assistant, or work with a team that does, you probably know these pains:

• Prompt files drift. Every assistant has its own format, so the same rules live in several places and slowly diverge.
• No view of the team. You have a pile of markdown files, not a picture of who does what and who may call whom.
• Standards don't travel. Your conventions stay in one repository instead of following you to the next project.
• No feedback loop. You cannot tell whether your agents actually do a good job.

What vsuite does

You describe your agent team once, in vsuite.json. vsuite validates it and generates the project-local agents, skills and MCP configuration for OpenCode, Claude Code and GitHub Copilot. Change the config, regenerate, and all three stay in sync.

The desktop app gives you a visual workspace with one-click generate, and the agent graph makes the structure explicit: a coordinator delegates to specialists, call permissions are enforced, and you can export the graph. vsuite detects your stack (frontend, backend, ORM, database, design library) and adds matching convention skills plus official MCP servers such as Context7 and Playwright.

Built for real teams

• Enterprise packs: versioned, integrity-locked standards you can pull from a folder, npm or git.
• Job metrics: agents log outcome, tool calls and review verdicts, so you finally get a feedback loop.
• Per-target models: choose the right model for each assistant.
• Safe by design: secrets live in a git-ignored .vsuite/.env and are referenced as ${VAR}; writes are atomic and migrations versioned.

Not sure where to start? The AI-assisted setup wizard walks you through it, optionally using a local model via Ollama.

The agent team behind this website

Here is my favourite proof: the agent team that builds and maintains this very website was generated with vsuite. The backend, frontend and review agents in this repository, their skills and their MCP wiring all come from one vsuite.json, and every job they finish is logged to the vsuite job log. The text you are reading was seeded into the CMS by one of them.

Describe your agent team once. Let every assistant speak the same language.

Who it is for

Consultants who set up AI tooling for many clients, engineering leaders who want shared standards across teams, designers who want their design system respected, and developers who simply want their assistants to behave consistently.

Downloads and source code

Builds are available for macOS, Windows and Linux. The macOS build is not notarized and the Windows and Linux builds are unsigned, so your operating system may ask you to confirm the first launch. Every versioned release ships with SHA256SUMS so you can verify your download. The source code is currently under review and will be published later. Everything you need to get started is in the documentation.