XiaomiMiMo/MiMo-Code

MiMo Code: Where Models and Agents Co-Evolve

What it solves

MiMoCode is a terminal-native AI coding assistant designed to handle long-horizon development tasks. It solves the problem of AI agents losing context over long sessions and the lack of structured, reproducible workflows for complex software engineering tasks like spec-to-ship feature delivery.

How it works

It operates as a CLI tool that can read/write code, execute commands, and manage Git. It employs several key mechanisms to maintain continuity:

  • Persistent Memory: Uses SQLite FTS5 to maintain project-wide knowledge (MEMORY.md), session checkpoints, and task progress across restarts.
  • Context Management: Automatically reconstructs context from checkpoints and project memory when the model's token limit is approached.
  • Agent System: Features specialized agents (Build, Plan, Compose) and the ability to spawn parallel subagents for complex tasks.
  • Workflows: Uses deterministic JavaScript scripts to orchestrate multi-phase pipelines (e.g., Brainstorm $\rightarrow$ Design $\rightarrow$ Implement $\rightarrow$ Verify) with bounded retries and parallelization.
  • Skills: A library of reusable instruction sets (e.g., for arXiv research or PDF generation) that can be loaded automatically or invoked via / commands.

Who it’s for

Developers who prefer a terminal-centric workflow and need an AI assistant capable of managing large-scale project architecture, autonomous research, and structured development pipelines.

Highlights

  • Terminal-native TUI: Full integration with the shell, Git, and system commands.
  • Long-horizon memory: Cross-session persistence via structured checkpoints and project memory.
  • Deterministic Workflows: Pre-defined JS-based pipelines for deep research, fact-checking, and feature implementation.
  • Self-Evolution: The /dream and /distill commands allow the agent to extract knowledge and package repeated patterns into new skills.
  • ** uma Multi-provider support**: Compatible with MiMo Auto, OpenAI, Claude, and any OpenAI-compatible API.
  • Voice Input: Integrated real-time streaming voice transcription for hands-free interaction.