MimoCode: Xiaomi's Open-Source Terminal AI Coding Assistant 113

MimoCode is an open-source AI coding assistant released by Xiaomi's MiMo team on June 11, 2026. It is a fork of OpenCode, licensed under MIT, and free to use for both individuals and enterprises.

MimoCode primarily addresses two issues: AI gradually forgetting earlier conversations as long tasks progress, and the AI wrapping up prematurely before the task is actually finished. The team positions it as an exploratory project aimed at comprehending the full codebase, planning modification strategies, editing code safely, and working in tandem with the team's existing development tools.

MimoCode Core Features

Persistent Memory

This is where MimoCode differs most from other AI coding tools. Most AI coding assistants rely on the model itself to take "notes" during the dialogue, and once the session ends or the context fills up, they forget everything. MimoCode offloads memory to a dedicated system, so the main AI just focuses on the work.

Specifically, it maintains three types of memory files:

• MEMORY.md: Project-level long-term information, such as architecture decisions and coding conventions, preserved across sessions

• checkpoint.md: Records of key checkpoints for the current session

• tasks/*/progress.md: Completion progress for each task

There is also a /dream command that runs automatically every 7 days, where an independent AI reads all historical sessions and existing memory files to merge, deduplicate, validate, and compress them, consolidating scattered information into a single current-state snapshot. Combined with SQLite FTS5 full-text search, you can retrieve historical memories across sessions.

Three Operating Modes

Press Tab to switch between primary agents:

• Build: The default mode, with full tool permissions for writing and running code

• Plan: Read-only mode, ideal for exploring the code and designing solutions

• Compose: Orchestration mode that can automatically run through the entire process of requirements analysis → coding → testing → review

Compose mode is one of MimoCode's highlights. Combined with the Harness system, it enables deep collaboration between the model and the development workflow.

Context Management

When the context window is about to fill up, MimoCode does not simply reconstruct history from scratch. Instead, it performs a structured rebuild from checkpoints, project memory, and recent messages, injecting content into the window based on importance to avoid bloating the window with irrelevant information.

Model Support

It comes with the built-in MiMo-V2.5 multimodal model, featuring million-token-level context, with performance on par with Claude Sonnet 4.6, currently available for free during the limited period. Simply select the "MiMo Auto" channel on first launch to use it directly—no registration required.

It is also compatible with 75+ mainstream model providers, including DeepSeek, Kimi, and GLM, and supports one-click import of existing credentials from Claude Code.

Voice Control

With MiMo-V2.5-ASR speech recognition, you can issue voice commands, modify code, or simply say "send" or "execute" to perform actions without using the keyboard.

Stop Conditions

Use the /goal command to set completion criteria. When the AI wants to end the task, it must be evaluated by an independent judge model to determine whether the goal has truly been achieved, reducing instances of "false completion."

Platform Support

Supports Mac, Linux, and Windows, with a fully localized Chinese interface and a status dashboard for real-time progress monitoring.

MimoCode Use Cases

• Large-scale project refactoring: The memory system maintains an understanding of the project across sessions, ideal for complex refactoring tasks

• Long-horizon development: For tasks with 200+ steps, MimoCode achieves a win rate of over 65%

• Specs-driven development: Compose mode can automatically complete the process from requirements breakdown to code review

• Beginner-friendly: Zero configuration, full Chinese localization, ready to use out of the box—ideal for developers new to AI coding assistants

• Terminal workflow: Suited for developers who prefer working in the terminal, integrating smoothly with Git and testing tools

MimoCode Pros and Cons

Pros

• The long-horizon memory capability is the biggest highlight—the four-layer memory system solves the forgetfulness problem in long sessions

• Completely free and open-source under the MIT license, no registration required, with a built-in free model ready to use

• Scores 62% on SWE-Bench Pro V2 and 73% on Terminal Bench 2, outperforming Claude Code's 57% and 68% respectively

• Supports 75+ model providers, avoiding lock-in to any single model

• Compose mode strings design, coding, testing, and review into a complete workflow

• Voice interaction boosts operational efficiency

• Supports one-click credential migration from Claude Code

Cons

• In complex tasks, it can sometimes be "overly aggressive"—starting to act before fully reading the code, potentially missing critical methods

• Pure command-line interface, though localized, still poses a barrier for users unfamiliar with the terminal

• MiMo-V2.5 is free for a limited time; long-term cost remains uncertain

• Product positioning is somewhat unclear—it's hard to tell whether it's a formal product or an experimental demo

• Some features lack stability; users have encountered installation failures, WSL compatibility issues, and upgrade failures

MimoCode vs. Competitors

FAQ

Q: How do I use it for free?

Select MiMo Auto on first launch—this is the limited-time free channel and requires no registration.

Q: Which models are supported?

It comes with built-in MiMo-V2.5 (free for a limited time) and supports 75+ mainstream model providers, including DeepSeek, Kimi, and GLM.

Q: Where is the configuration file?

.mimocode/mimocode.json in the project directory, or the global configuration at ~/.config/mimocode/mimocode.json.

Q: What if the MiMo Auto channel fails to load?

Delete the ~/.config/mimocode folder and re-run /connect to switch back to MiMo Auto.

Q: Copying in WSL produces garbled text?

Install xsel: sudo apt install xsel.

Q: Chinese garbled text on Windows?

Enable system-wide UTF-8 support: Settings → Time & Language → Language & Region → Administrative language settings → Change system locale → Check "Beta: Use Unicode UTF-8 for worldwide language support" → Restart.

Q: How do I upgrade?

mimo upgrade. If the upgrade fails, refer to GitHub Issue #1317.