███╗░░░███╗░█████╗░██████╗░███████╗██╗░░░░░░█████╗░██╗░░░░░██╗
████╗░████║██╔══██╗██╔══██╗██╔════╝██║░░░░░██╔══██╗██║░░░░░██║
██╔████╔██║██║░░██║██║░░██║█████╗░░██║░░░░░██║░░╚═╝██║░░░░░██║
██║╚██╔╝██║██║░░██║██║░░██║██╔══╝░░██║░░░░░██║░░██╗██║░░░░░██║
██║░╚═╝░██║╚█████╔╝██████╔╝███████╗███████╗╚█████╔╝███████╗██║
╚═╝░░░░░╚═╝░╚════╝░╚═════╝░╚══════╝╚══════╝░╚════╝░╚══════╝╚═╝
Call any LLM from the command line via models.dev.
- API-key based models — works with any provider that uses standard API keys (OpenAI, Anthropic, Google, xAI, Groq, Mistral, etc.)
- Some subscription models work — as long as they use an API Key (Z.AI's GLM, Minimax, etc.)
- No OAuth subscription models — does not support models behind OAuth login (e.g. ChatGPT Plus/Pro, Codex subscriptions)
npm install -g @blankeos/modelcli # npm bun install -g @blankeos/modelcli # or bun cargo binstall modelcli # or cargo-binstall (prebuilt binary, faster) cargo install modelcli # or cargo (build from source) curl -sSL https://raw.githubusercontent.com/Blankeos/modelcli/main/install.sh | sh # or linux/macos (via curl)
# 1. Connect to a provider (Any known provider thanks to models.dev) modelcli connect # 2. Browse models and set a default modelcli models # 3. Send a prompt modelcli "What is the meaning of life?"
modelcli [OPTIONS] [PROMPT]
| Command | Description |
|---|---|
connect |
Connect to a provider (add API key) |
models |
Browse and manage models |
| Flag | Description |
|---|---|
--model <provider/model-id> |
Model to use (overrides default) |
--stream |
Stream tokens as they arrive |
--thinking |
Show thinking/reasoning tokens |
--reasoning-effort <level> |
Reasoning effort: low, medium, or high |
--format json |
Output raw JSON instead of human-readable text |
# Use a specific model modelcli --model openai/gpt-4o "Explain quicksort" # Stream the response modelcli --stream "Write a haiku about Rust" # Enable reasoning modelcli --thinking --reasoning-effort high "Prove that √2 is irrational" # JSON output modelcli --format json "Hello"
You can add any OpenAI-compatible provider not listed on models.dev.
1. Add a credential:
modelcli connect
# Select "Other (custom provider)" → enter a provider ID and API key2. Configure the provider in ~/.config/modelcli.jsonc:
{
"provider": {
"myprovider": {
"name": "My AI Provider",
"baseURL": "https://api.myprovider.com/v1",
"models": {
"my-model": {
"name": "My Model", // optional display name
"reasoning": false, // optional, default false
"context": 200000, // optional context window
"output": 65536, // optional max output tokens
},
},
},
},
}Then use it like any other model:
modelcli --model myprovider/my-model "Hello!"The config file is auto-created the first time you add a custom provider. Both
.jsoncand.jsonare supported, but not both at the same time.
- Credentials and app data:
~/.local/share/modelcli/ - Custom provider config:
~/.config/modelcli.jsonc
modelcli enables piping LLM calls directly from your terminal—perfect for generating commit messages in lazygit (see PR #5389), or powering any other CLI app with AI capabilities. Quickly ask questions or pipe stdout from other tools to get instant AI-powered responses.
Inspired by OpenCode's seamless multi-provider experience and built on models.dev's unified LLM API.
🦀 Made w/ Rust. A fast, minimal but intuitive CLI made with Rust.