llm-cli

A blazing fast, single-binary CLI for chatting with any OpenAI-compatible LLM from your terminal. No runtime dependencies. No Docker. No Node. Just a 10 MB binary that works on any Linux machine.

Heavily inspired by simonw/LLM but but built for people who want something simpler.

$ llm -s "What command shall I use to see all files and folders including hidden ones"
ls -la

macOS and Windows releases coming soon.

Install

wget -qO- https://raw.githubusercontent.com/Glaicer/llm-cli/main/install.sh | bash

That's it. The llm binary is dropped into ~/.local/bin. Make sure it's in your PATH:

export PATH="$HOME/.local/bin:$PATH"

Usage

Single shot — one question, one answer, done:

llm -s "Explain the difference between TCP and UDP"

Interactive mode — multi-turn conversation with full context memory:

$ llm
>> What command lists all files
ls
>> And hidden ones
ls -la
>> exit

You can also seed the conversation with an initial prompt:

llm "What is the capital of France"

REPL commands: exit, quit, Ctrl-D, or Ctrl-C to leave.

Config

On first run, llm walks you through setup interactively — base URL, model, and API key. The config is saved to:

~/.config/llm-cli/config.toml

Edit it directly to change settings:

base_url = "https://api.openai.com/v1"
api_key = "sk-..."
model = "gpt-5.4-mini"
system_instruction = "You are an AI CLI assistant. Your answers should be short and concise. If the user asks for command, output just command without any comments."

Option	Description
base_url	Any OpenAI-compatible API endpoint (OpenAI, Ollama, LM Studio, vLLM, etc.)
api_key	Bearer token for authentication
model	Model ID accepted by your endpoint (e.g. gpt-4o-mini, claude-opus-4)
system_instruction	System prompt prepended to every conversation

Environment overrides — set these to override the config file without editing it:

LLM_CLI_API_KEY=sk-...
LLM_CLI_BASE_URL=
LLM_CLI_MODEL=

Uninstall

Remove the binary:

rm ~/.local/bin/llm

Remove the config:

rm -r ~/.config/llm-cli

Development

git clone repo
cd llm-cli
cargo build
cargo test

The binary is built at target/debug/llm. For a release build:

cargo build --release

License

MIT