Unofficial terminal UI (TUI) + CLI for the MiniMax platform: chat with MiniMax-M2.1, run an approval-gated tool-using agent, and generate media (images, video, music, TTS).
Highlights:
- Streaming chat via MiniMax’s Anthropic-compatible API format
- Tool-using agent with approvals + workspace sandbox
- Built-in MiniMax media tools (image/video/music/TTS/voice) that save outputs to your workspace
- Skills (
SKILL.md) and external tools via MCP - Project-aware prompts via
AGENTS.md(and.claude/instructions.md/CLAUDE.md)
Not affiliated with MiniMax Inc.
- Get an API key from https://platform.minimax.io
- Run
minimaxand paste your key when prompted (saved to~/.minimax/config.toml), or setMINIMAX_API_KEY - Press
F1or run/helpfor the in-app command list
The NPM package is a thin wrapper that downloads the platform-appropriate Rust binary from GitHub Releases.
# npm / bun (installs `minimax`)
npm install -g @hmbown/minimax-cli
bun install -g @hmbown/minimax-cli
### From source (Rust)
```bash
cargo install minimax-cliDownload a prebuilt binary from https://github.com/Hmbown/MiniMax-CLI/releases and put it on your PATH as minimax.
minimax # Interactive TUI
minimax -p "Write a haiku" # One-shot prompt (prints and exits)
minimax doctor # Diagnose config + API key
minimax sessions # List auto-saved sessions (~/.minimax/sessions)
minimax --resume <id> # Resume by ID/prefix (or "latest")
minimax --continue # Resume most recent session
minimax --workspace /path/to/project
minimax --yolo # Start in Agent mode + auto-approve tools (dangerous)
minimax init # Generate a starter AGENTS.md for the current directoryShell completions:
minimax completions zsh > _minimax
minimax completions bash > minimax.bash
minimax completions fish > minimax.fishThe TUI can save your API key during onboarding. You can also create ~/.minimax/config.toml manually.
Minimal config:
api_key = "YOUR_MINIMAX_API_KEY"
default_text_model = "MiniMax-M2.1" # optional
allow_shell = false # optional
max_subagents = 3 # optional (1-5)Note: api_key must be a non-empty string.
Useful environment variables:
MINIMAX_API_KEY(overrides config)MINIMAX_BASE_URL(default:https://api.minimax.io; China users may usehttps://api.minimaxi.com)MINIMAX_PROFILE(selects[profiles.<name>]from config; errors if missing)MINIMAX_CONFIG_PATH(override config file path)MINIMAX_ALLOW_SHELL,MINIMAX_SKILLS_DIR,MINIMAX_MCP_CONFIG,MINIMAX_NOTES_PATH,MINIMAX_MEMORY_PATH,MINIMAX_OUTPUT_DIR,MINIMAX_MAX_SUBAGENTS
See config.example.toml for a fuller config reference.
If your workspace has an AGENTS.md, the TUI loads it into the system prompt automatically (and will also look in parent directories up to the git root). Use it to tell the agent how to work in your repo: commands, conventions, and guardrails.
Create a starter file:
- CLI:
minimax init - In-app:
/init
Switch modes with Tab or /mode:
- Normal: chat
- Edit: file-focused assistance
- Agent: multi-step tool use (with approvals)
- Plan: design-first prompting
- RLM: load/search/chunk large files in an in-app sandbox
The built-in help (F1 or /help) is always up to date. Common commands:
| Command | What it does |
|---|---|
| `/mode [normal | edit |
/model [name] |
View/set model name |
/skills |
List skills |
/skill <name> |
Activate a skill for your next message |
/save [path] |
Save current chat to JSON |
/load <path> |
Load chat JSON (or load a file into RLM context in RLM mode) |
/export [path] |
Export transcript to Markdown |
/yolo |
Switch to Agent mode + enable shell tool (still prompts for approval) |
/trust |
Allow file access outside workspace |
/tokens |
Token totals + metadata |
/context |
Context usage estimate |
/cost |
Pricing reference for paid tools |
/subagents |
Show sub-agent status |
MiniMax CLI exposes a tool set to the model (file read/write, patching, web search, sub-agents, and MiniMax media APIs). By default, the TUI asks before running tools with side effects:
- File writes:
write_file,edit_file,apply_patch - Shell:
exec_shell(disabled unlessallow_shell=trueor/yoloor--yolo) - Paid/Media:
generate_image,generate_video,generate_music,tts, voice tools, file upload/download
The built-in web_search tool is backed by DuckDuckGo HTML results and is auto-approved.
File tools are restricted to the --workspace directory unless you enable /trust.
MiniMax CLI includes first-class tools for MiniMax’s media endpoints. In practice: ask for an image/video/music/TTS and the assistant can generate it and save outputs into your workspace.
Built-in MiniMax tool names (for power users): generate_image, generate_video, query_video, generate_music, tts, analyze_image, voice_clone, voice_list, voice_delete, voice_design.
Skills are reusable workflows stored as SKILL.md files inside a directory.
- If your workspace contains
./skills/, the TUI uses that. - Otherwise, it falls back to
~/.minimax/skills/.
Use /skills to list and /skill <name> to activate.
This repo includes example skills like video-studio, voiceover-studio, music-video-generator, and audiobook-studio under skills/.
MiniMax CLI can load additional tools via MCP (Model Context Protocol). Configure ~/.minimax/mcp.json (supports servers and mcpServers keys), then restart the TUI.
For Coding Plan MCP setup, see docs/coding-plan-integration.md.
docs/ARCHITECTURE.mddocs/coding-plan-integration.mdCONTRIBUTING.md
cargo build
cargo test
cargo fmt
cargo clippyMIT
MiniMax is a trademark of MiniMax Inc. This is an unofficial project.