LLM CLI

A fast command-line interface for interacting with large language models. The tool supports multiple providers through OpenAI-compatible APIs, configurable system prompts, inline file inclusion, and streaming responses.

Features

• Provider-scoped configuration with per-model aliases
• Named system prompts with {{info}} placeholder for runtime context (date, OS, shell, etc.)
• File and URL references via @path tokens in prompts
• Streaming or full-response modes
• Optional thinking/chain-of-thought display for reasoning models

Installation

1. Go to the releases page and download the binary for your platform.
2. Add the binary to a directory in your PATH.

• Example for Linux/macOS: sudo mv llm /usr/local/bin

3. Reload your shell or session to ensure the binary is available.

Build from source

1. Clone this repository
2. Install dependencies: go mod tidy
3. Build the binary: go build -o llm .

Configuration

Copy config.yaml.example to config.yaml and adjust:

• providers list with base URL, API key, and models per provider
• default_provider and default_model to choose the defaults
  • default_model does not support aliases, you must use provider>name instead
• Model aliases to map short names to provider>model
• system_prompts definitions; the default is used unless -p/--prompt is set
• Include {{info}} in a prompt to expand runtime context

Environment variables prefixed with LLM_ override config values (e.g., LLM_API_KEY).

Usage Examples

$> llm how can i list all the files in this directory sorted by size
>> gpt-oss-20b (oss):
Use `ls -lS`

$> llm summarize @README.md
>> gpt-oss-20b (oss):
**LLM CLI**
- Command‐line tool to talk to large language models via OpenAI‐compatible APIs.
- Supports multiple providers, streaming, full responses, chain‐of‐thought display.
**Features**
- Model aliases; one can be set [...continues...]

• Use @filename or @https://url to inject contents into the prompt
• --raw outputs responses without processing
• --think shows the model’s thinking phase when supported

Run llm --help for the full option list.