MiniMax CLI
https://github.com/hmbown/minimax-cli- Category
- AI Tools
- Rank
- No. 1619Tools index
- Pricing
- Open Source
- Type
- TOOL
- Builder
- Hmbown
- GitHub
- 27 stars
- Latest release
- v0.7.1
- Date
About
Rust-based CLI for working with MiniMax models.
What it does
MiniMax CLI combines an interactive terminal chat, one-shot prompts, coding agents, and media generation. It can read and change workspace files, run shell commands, search the web, delegate work, resume sessions, and expose a local runtime server.
Why it's ranked here
The breadth is unusually practical for a terminal client. Approval-aware agent modes, persistent sessions, large-context workflows, MCP integration, and an HTTP/SSE runtime make it useful beyond basic chat. The security model is uneven, however, because some modes and MCP tools bypass approval prompts.
What's good
Workspace boundaries constrain file tools by default, while normal agent workflows request approval for shell access, writes, and paid media. Background job and sub-agent metadata survives restarts with explicit recovery states. Large-file handling, session resumption, diagnostics, shell completions, and local API endpoints support serious repeatable workflows.
Tradeoffs
This is an unofficial client that requires MiniMax credentials. Image, video, music, and speech operations can spend credits and write files. YOLO and large-context modes automatically approve tools, while MCP tools currently receive no terminal approval prompt. Restarted processes cannot be reattached, so restored running jobs become orphaned and active sub-agents become failed.
How to use it well
It best suits engineers who want MiniMax chat and coding automation inside a terminal, especially for multi-step repository work, large-document analysis, or scripted access through a local server. Start with approval-gated modes and a narrow workspace. Reserve automatic approval for controlled environments. Do not choose it as a provider-neutral model client or a safeguard around untrusted MCP servers.
Technical notes+
Cargo.toml defines a Rust 2024 binary named minimax, with Axum, Tokio, Reqwest, Ratatui, Clap, Starlark, PDF extraction, and test dependencies including Wiremock. src/main.rs routes interactive, diagnostic, session, sandbox, MCP, runtime server, review, execution, RLM, Duo, and media smoke-test commands. src/mcp.rs implements JSON-RPC initialization, tool discovery, pooled stdio child processes, configurable timeouts, and process termination on drop. src/rlm.rs stores external context, supports regex search and several chunking strategies, and sizes recommended context from detected memory. src/duo.rs models the player-coach state machine and persists turn history and quality signals. src/client.rs centralizes authentication, streaming, base URLs, and retries for rate limits and server failures. pyproject.toml separately declares a Python launcher package that downloads and runs the Rust binary.
Observed
- License
- MIT
- Primary language
- Rust, using the 2024 edition
- Packaging
- Cargo crate, prebuilt GitHub Release binary, source build, and a declared Python launcher package
- Interfaces
- Interactive TUI, command-line interface, MCP stdio server, and local HTTP/SSE runtime API
- Model and media scope
- MiniMax text chat plus image, video, music, and text-to-speech generation
- Configuration
- TOML configuration, named profiles, environment-variable overrides, and configurable MiniMax base URLs
- Shell integration
- Completion generation for Zsh, Bash, and Fish
Read from README.md, Cargo.toml, pyproject.toml, src/main.rs, src/ui.rs, src/duo.rs, src/mcp.rs, src/rlm.rs, src/hooks.rs, src/smoke.rs, src/utils.rs, src/client.rs.
What it can do
Execute MiniMax model inference
Text prompts and model parameters → Model-generated responses
Load and configure MiniMax models
Model file paths and configuration settings → Initialized model instance
Process batch text inputs
Multiple text prompts or input file → Batch processing results
Manage model parameters
Parameter configuration commands → Updated model settings
Export inference results
Generated model outputs → Formatted output files or console display
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.