Mirafold
Your Agent with Generative UI
Last checked:
What is Mirafold?
Mirafold is a browser-based UI layer for terminal coding agents (Claude Code, Codex, Gemini CLI) that converts terminal output into interactive, generative components like cards, tables, and charts. It runs locally on your machine, keeps API keys private, and works with both cloud APIs and local models like Ollama.
Mirafold pricing
Pricing model: Freemium
Free (open source, MIT) with full local features; Pro $12/mo or $99/yr for multi-device access and encrypted relay
Mirafold pros
- Runs entirely locally with API keys never sent to browser; works with local models (Ollama, LM Studio, vLLM) for complete privacy
- Generates intelligent UI components (cards, tables, charts, task lists) automatically based on agent output instead of dense scrollback
- Maintains full feature parity with terminal agent—same engine, settings, memory, and permissions; install with one npm command
- Multi-session management and multi-device access on Pro with end-to-end encrypted relay; real PTY passthrough for sudo/ssh prompts
Mirafold cons
- Requires Node 22+ to install
- Cannot use Claude/ChatGPT/Gemini subscription accounts—must use API keys (except with local models)
- Pro tier ($12/mo) adds cost for multi-device access; some features like real-time sync across devices gated behind paid plan
Frequently asked questions about Mirafold
Is this a web UI for Claude Code, Codex, and Gemini CLI?
Yes—a browser interface that runs your terminal agent with the same engine, settings, and permissions, plus generative UI on top. Install with `npm i -g mirafold` and run `mirafold` in your project directory.
Does it work with local models?
Yes—fully local when pointing at Ollama, LM Studio, or vLLM, keeping code, keys, and tokens entirely on your machine.
Where does my API key live?
In your local daemon's environment like the terminal. On Pro, the encrypted relay carries only ciphertext and connection metadata, never your key or conversation.
Can I use my Claude / ChatGPT / Gemini subscription?
Claude and Gemini require API keys (subscriptions prohibited by their terms). ChatGPT/Codex subscriptions work locally, though OpenAI's third-party terms are not explicitly written.
How is it safe when the agent generates UI?
Two-zone architecture: agent output renders only into an output zone with validated components and sandboxed iframes; the prompt box and credentials stay in a trusted shell the agent cannot access.