MiMo Code
MiMo Code is Xiaomi's terminal-native coding agent. Rechecked 2026-08-25: github.com/XiaomiMiMo/MiMo-Code had 12,863 stars, license MIT, plus a separate Use Restrictions file. The README says it is a fork of OpenCode with persistent memory, compose workflows, subagents, and /dream / /distill self-improvement. The CLI itself is $0. Auth is Xiaomi MiMo OAuth, Codex OAuth, imported Claude Code auth, or any OpenAI-compatible key. The marketing site says you can try a multimodal model without login. That is not a published dollar seat table. Do not invent one.
Compare OpenCode if you wanted the upstream harness, Qwen Code if you wanted Alibaba's open CLI, and Claude Code if you wanted Anthropic's first-party agent.
Key Features
- OpenCode fork plus extras: TUI, LSP, MCP, and plugins stay. Added pieces include SQLite FTS5 memory, compose mode, workflows, and goal/stop judging.
- Multiple agents:
build,plan, andcompose. Tab switches primary agents. Compose locks after the first message. - Auth menu: Xiaomi platform OAuth, ChatGPT/Codex OAuth, Claude Code import, catalog providers, or a custom OpenAI-compatible endpoint.
- Yolo flag:
mimo --dangerously-skip-permissionsorMIMOCODE_DANGEROUSLY_SKIP_PERMISSIONS=1. The README calls this dangerous. - macOS Terminal.app: first-party docs say the built-in Terminal is unsupported. Use iTerm2 or VS Code's terminal.
Limitation: we did not open a logged-in Xiaomi checkout, so there is no public dollar table on this page. Quality still depends on the model you attach. "Unlimited context" on the marketing page is a product claim, not an audited window.
Use Cases
- People who will stay on Xiaomi OAuth or BYOK and treat the CLI as MIT tooling.
- People who already like OpenCode and want Xiaomi's memory / compose layer on top.
- People comparing Chinese open CLIs against Qwen Code.
Pricing
CLI is free. Model access on 2026-08-25.
| Path | Price | Notes from first-party pages |
|---|---|---|
| MiMo Code CLI | $0 | MIT. 12,863 stars on the check date. Use Restrictions still apply. |
| Xiaomi MiMo platform | OAuth / platform bill | Homepage says try without login. Recheck the live Xiaomi page. Do not invent a dollar seat. |
| Codex / Claude import / custom | Provider or $0 local | ChatGPT Pro/Plus OAuth, imported Claude Code auth, or any OpenAI-compatible key. |
If a leftover blog prints a Xiaomi $19 or $200 seat, treat this table as the source.
Getting Started
- Install with
curl -fsSL https://mimo.xiaomi.com/install | bash, ornpm install -g @mimo-ai/cli. - Run
mimo. Complete the first-launch auth wizard. - Use
/initstyle project files under.mimocode/if you want project config. - Do not pass
--dangerously-skip-permissionsoutside a disposable sandbox.
First-party start: mimo.xiaomi.com/coder and the GitHub README.
Frequently Asked Questions
Is the CLI still free?
Yes. The MIT CLI is $0. Models are billed by the provider you attach.
Same as OpenCode?
No. It is a fork. Upstream is OpenCode. This page is Xiaomi's agent plus memory and compose.
Does --dangerously-skip-permissions skip deny rules?
No. The README says an allow-all base is injected under your config. Explicit deny and leftover ask rules still win.
Alternatives
- OpenCode: upstream harness without Xiaomi's extras.
- Qwen Code: Alibaba Apache-2.0 CLI plus ModelStudio.
- Claude Code: Anthropic first-party agent.
Tips
- Quote 12,863 stars and MIT from the GitHub check.
- Do not invent a Xiaomi dollar seat.
- Skip macOS Terminal.app. Use iTerm2.
Conclusion
MiMo Code is a 12,863-star MIT OpenCode fork with Xiaomi memory and compose, not a leftover free-forever seat. Start at github.com/XiaomiMiMo/MiMo-Code, attach a live provider, then decide whether OpenCode already covers the harness you need.
Comments
No comments yet. Be the first to comment!
Related Tools
Autohand Code
www.autohand.ai/code
Self-evolving terminal agent. Rechecked: Apache-2.0 with ARR cap, 178 stars. CLI $0 plus BYOK. Enterprise is sales, not a public $20 seat.
Mistral Vibe
github.com/mistralai/mistral-vibe
Mistral terminal agent. Rechecked: Apache-2.0, 4,872 stars. CLI $0 plus Mistral API key. --agent auto-approve / --yolo. Not a $20 seat.
OpenClaude
openclaude.gitlawb.com
Open terminal agent. Rechecked: 30,818 stars. MIT for contributor changes; derived from Claude Code. CLI $0 plus BYOK.
Related Insights
What Locks Codex Is the Picker, Not the Models
You already pay for OpenCode Go, Grok, and Z.ai, but the Codex picker still shows mostly GPT. The community project codex-router does not teach another install ritual. It puts subscriptions you already bought back into the selector, keeps keys on the machine, and leaves native GPT alone.
Hook OpenCode Go into Codex on Windows. Do Not Open a Second Toolkit.
A ChatGPT-signed Codex desktop app still shows mostly GPT in the picker. On Windows, enable only OpenCode Go and the Grok, GLM, Kimi, DeepSeek, and MiniMax models you already pay for appear in the same selector. Keys stay local. Native GPT stays put.
Seven AI Coding CLIs, Six Months: No Matter How Strong the Model, Work Needs Supervision
Claude Code, Codex, opencode, pi, omp and DeepSeek Harness all have personalities. After six months of deep use I run a division of labor: pi for the fastest cheapest reviews, omp for complex PRs, DeepSeek Harness on V4 Flash for high-frequency low-cost review, and Claude Code, Qoder and Cursor for writing. No matter how strong the model, work needs supervision — ideally from an independent third party.