Skip to content

What is mimo-tui

mimo-tui is an AI coding assistant that runs in your terminal, powered by Xiaomi’s MiMo model. You chat with it, and it reads your code, edits files, runs commands, fetches docs, manages git — the same workflow as Claude Code, Codex CLI, and Gemini CLI, just with a different underlying model.

Why this project exists

Xiaomi’s MiMo (especially MiMo-V2.5-Pro — a trillion-parameter MoE with 1M context) is one of the most cost-effective Chinese LLMs in 2026. It supports the Anthropic Messages protocol, tool use, MCP, prompt caching, and thinking — everything Claude Code uses.

But Xiaomi never shipped a corresponding terminal tool:

  • The XiaomiMiMo GitHub org has 11 repositories — zero CLI / TUI / SDK
  • The “official” path is “configure Claude Code, then override env vars” — broken UX, config conflicts, and a model that thinks it’s Claude
  • The only existing “MiMo TUI” project is built on reverse-engineered Claude Code — legal grey area

So we built one. Original, compliant, open source, free forever.

Relationship to Claude Code

AspectClaude Codemimo-tui
Built byAnthropicCommunity OSS
ModelClaude Opus / SonnetMiMo-V2.5-Pro
ProtocolAnthropic Messages APIAnthropic Messages API ✓
Tool Use
MCP
Skills✓ Anthropic formatAnthropic format compatible
Sub-agents
Sandbox
Config dir~/.claude/~/.mimo/
PricingClaude planMiMo Token Plan
Paid?YesAlways free

Skills are fully compatible — you can drop your existing Claude Code skill folder into mimo-tui and use it as-is.

Design principles

  1. Native UX, no skin-on-top. Model identifies itself correctly (“I’m MiMo”). All UI, commands, and docs are MiMo-native.
  2. Deep MiMo optimization. Live thinking rendering. Cache hit visualization. Plan usage bar. 1M context management.
  3. Stick to standards. Anthropic protocol, MCP protocol, Skills format — no NIH.
  4. Free and open forever. MIT licensed. No Pro / Enterprise tier.

Who’s this for

  • Existing MiMo Token Plan subscribers who want a real CLI
  • Devs who want a Chinese-model alternative but won’t tolerate the env-var hack
  • Compliance-sensitive users who don’t trust reverse-engineered tools
  • Chinese devs / international devs wanting to try Chinese LLMs

Next