MiMo Code Commands Reference
Every slash command, keyboard shortcut, and work mode explained.
Slash Commands
| Command | Description | Usage |
|---|---|---|
| /init | Initialize project context and memory | Run first in any project. Creates .mimocode/ directory with project rules. |
| /connect | Connect or switch model provider | Change API provider mid-session without restarting. |
| /thinking | Toggle thinking/planning visibility | Show or hide the agent's reasoning process. |
| /memory | View, edit, or reset project memory | Manage persistent memory: view history, add rules, clear context. |
| /plan | Enter plan mode | Agent analyzes and proposes a plan before making changes. |
| /build | Enter build mode | Agent makes code changes immediately (skip planning). |
| /compose | Enter compose mode | Agent plans first, then builds (recommended for complex tasks). |
| /checkpoint | Save a checkpoint | Create a restore point before major changes. |
| /rollback | Roll back to a checkpoint | Undo changes since last checkpoint. Useful for failed experiments. |
| /dream | Consolidate project memory | Agent reviews and distills project context into compact form. |
| /distill | Distill task learnings | Save lessons learned from a completed task into memory. |
| /config | View or edit current config | Show current provider, model, mode, and environment settings. |
| /clear | Clear conversation context | Reset the current session's chat history (memory persists). |
| /status | Show session status | Display current model, token usage, memory state, and active tools. |
| /exit | Exit MiMo Code | Save session and return to terminal. |
Keyboard Shortcuts
| Shortcut | Action |
|---|---|
Ctrl+C | Interrupt agent's current action |
Ctrl+D | Exit MiMo Code (EOF) |
Ctrl+L | Clear screen |
Up Arrow | Previous command in history |
Tab | Autocomplete file paths |
Work Modes in Detail
Plan Mode
Use /plan or start with --mode plan. The agent analyzes your request, studies the codebase, and presents a detailed plan. No code is modified. Approve the plan, then switch to build mode.
Best for: Complex refactors, architecture changes, unfamiliar codebases.
Build Mode
Use /build or start with --mode build. The agent makes changes immediately. Good for simple, well-understood tasks.
Best for: Bug fixes, small features, known codebases.
Compose Mode
Use /compose or start with --mode compose. The agent plans first, then asks for approval before building. Best of both worlds.
Best for: New features, multi-file changes, most tasks (recommended default).
Memory System
MiMo Code's persistent memory system is what sets it apart for long-horizon tasks. Here's how it works:
Project Memory
Stored in .mimocode/memory.json. Contains project rules, architecture decisions, and learned patterns. Persists across sessions.
Checkpoints
Use /checkpoint before risky changes. Create named restore points. /rollback to undo. Stored in .mimocode/checkpoints/.
Dream & Distill
/dream consolidates project knowledge. /distill extracts lessons from completed tasks. These processes keep memory compact and relevant.
AGENTS.md / MEMORY.md
Create AGENTS.md at project root for rules the agent should always follow. MEMORY.md for project-specific context. Similar to Claude Code's CLAUDE.md.
Quick Tips
- Always run /init first in a new project — it sets up the foundation for everything else.
- Use checkpoints liberally — they cost nothing and save hours of rework.
- Write a good AGENTS.md — include coding conventions, tech stack, and key architectural decisions.
- Switch providers mid-task — use cheaper models for planning, then switch to a stronger model to execute.
- Disable telemetry if working with private code.
- Max Mode trades cost for quality — enable for complex tasks, disable for simple ones.