Download & double-click
Unzip, run the start file, your browser opens. No installer, no admin rights.
ONE BRIEF. THREE AGENTS. NO COPY-PASTE.
Give one task. A Planner sets the criteria, a Worker produces it, a Reviewer finds the errors and sends it back — automatically — until it passes.
Works with Claude, GPT and Gemini. Mix vendors so the reviewer is never grading its own work. Runs on your computer with your own API keys; nothing is uploaded to us.
Windows · macOS · Linux · needs Node.js 24 · no account · no telemetry
Issues go back to the Worker.
Repeat until it passes.
HOW IT WORKS
Demo mode shows the whole loop with no key and no cost. Add keys when you are ready.
Unzip, run the start file, your browser opens. No installer, no admin rights.
Synthetic agents plan, draft, get sent back and pass — so you see exactly what to expect.
Anthropic, OpenAI or Google — one or all three. Encrypted on your machine, verified with a tiny test call.
Paste any facts as material. Pick which vendor plays which role. Press run.
Every issue found and fixed is on screen. Accept, copy or export with the full record.
DOWNLOAD
Free · open source · v1.0.0
One zip for every platform. About 300 KB.
Download agentsfair-agent-studio.zip ↓
Requires
Node.js 24+ (free).
Then: Windows → start-windows.cmd; macOS → start-macos.command; Linux
→ start-linux.sh.
Your chat assistant can hand a brief to Studio and fetch the result — no copy-paste. Add the
bundled studio-connector.mjs to Claude Desktop, Cursor or any MCP client; the guide
inside the app has the exact configuration.
No browsing, files, email, payments or publishing — agents only write and critique text. It does not control other apps on your computer. It does not guarantee correctness; verify facts before use. Provider charges go to your own accounts.
FAQ
No. Studio runs on your computer. The only outbound requests are the model calls you authorise, straight to the provider you chose. If you give us your email above, that is the only thing we receive.
Nothing to us. Providers bill your own API accounts; a typical three-round job makes five to seven model calls. Demo mode is free.
No. A second model catches many mistakes, especially when it is from a different vendor, but not all. Treat the result as a strong draft with a built-in review, and verify facts.
Any model name your provider account can access. Defaults are set for Claude, GPT and Gemini and can be changed in settings.
程序内含完整中文说明(Guide / 中文说明),安装、加 Key、运行、接入 Claude/Cursor 都有逐步指引。