llm

llm provides a simple way to talk to LLMs available via OpenRouter right from many programmers home (the terminal).

Table of Contents

• Install
• Examples
  • Basic Usage: Ask Anything
  • Model Selection (-m or --model)
  • Model Listing
  • Streaming Output
  • Clipboard Copy (-C or --copy)
  • Templates (-t or --template)
  • Configurable
  • API Keys
  • Shell Completion
  • Verbose Mode (-v or --verbose)
• Coming Soon
• Note

Install

Clone the repository and install the llm executable to your Go binary path ((go env GOPATH)/bin):

git clone https://github.com/flacial/llm
cd llm
go install .

Examples

Basic Usage: Ask Anything

You can send a prompt in a few ways:

1. As a direct argument:

   llm "What's the capital of Sudan?"

2. Via standard input (stdin):

   echo "Tell me a short story about a brave dragon and a sleeping cow." | llm

3. From a file (-f or --file):

   echo "Summarize the key points of the paper Attention Is All You Need." > summary.txt

   Then, run llm with the file:

   llm -f summary.txt

Model Selection (-m or --model)

Override your default model (if set) or specify a particular model for a single query.

llm -m google/gemini-flash-1.5 "Explain the concept of recursion in programming as if I'm a grug programmer."

Model Listing

View all available LLM models from OpenRouter:

llm models

Streaming Output

By default, llm streams responses live.

llm "Write a haiku about a bustling city at sunset."

Clipboard Copy (-C or --copy)

Automatically copy the LLM's response to your system clipboard.

llm -C "What is the chemical symbol for gold?"

(After running, you can paste the answer (Au) into any text field.)

Templates (-t or --template)

Use predefined prompts for common tasks.

Setup: Add a new template to your ~/.llm/templates/ folder.

# ~/.llm/templates/brainstorm.tmpl.yaml
name: "brainstorm"
description: "Generates creative ideas, concepts, or solutions for a given topic."
system_message: |
  You are a creative brainstorming assistant. Your role is to generate a diverse range of ideas, concepts, or solutions based on the user's input. Think broadly, explore different angles, and provide innovative suggestions. Encourage out-of-the-box thinking.
user_prompt_template: |
  I need some brainstorming ideas for:

  {{.UserPrompt}}

  Please provide at least 5 distinct ideas or approaches.

Usage: Provide a prompt and apply the template.

llm "Vacation plans for going to paris" -t brainstorm

Configurable

Set a default model or other options in your configuration file so you don't have to specify them every time.

Configuration File Location: The configuration file is located at $XDG_CONFIG_HOME/llm/config.yaml (typically ~/.config/llm/config.yaml on most systems). If XDG_CONFIG_HOME is not set, it defaults to ~/.config. You can also specify a custom location using the --config flag.

Setup: Add a default model to your configuration file:

# ~/.config/llm/config.yaml
model: openai/gpt-3.5-turbo # This will be used if -m is not specified
api_key: "your-api-key-here"  # Optional if using environment variable

Model Aliases: You can use shorter aliases for commonly used models:

# ~/.config/llm/config.yaml
model: fast # Uses the built-in alias for openai/gpt-4.1-nano
# Or define custom aliases:
models:
  aliases:
    my-model: "anthropic/claude-3-5-sonnet"
    quick: "google/gemini-flash-1.5"

Built-in Aliases:

• fast: openai/gpt-4.1-nano
• 10x: anthropic/claude-sonnet-4
• smart: google/gemini-2.5-pro
• gpt4: openai/gpt-4o

Usage: Now you can run llm without the -m flag:

llm "What is the capital of Sudan?"
# Or use an alias
llm -m fast "Quick question here"

API Keys

Ensure your LLM_API_KEY environment variable is set, or include api_key: "YOUR_KEY_HERE" in your ~/.llmrc.yaml.

# Example of setting an API key via environment variable (for current session)
export LLM_API_KEY="sk-or-..."
llm "Hello!"

Shell Completion

Bash:

To load completions for the current session:

$ source <(llm completion bash)

To load completions for each new session, execute this once:

• Linux:

  $ llm completion bash > /etc/bash_completion.d/llm

• macOS:

  $ llm completion bash > /usr/local/etc/bash_completion.d/llm

Zsh:

To load completions for each session, execute this once:

$ llm completion zsh > ~/.zsh/_llm

Restart your terminal for it to work.

Fish:

To load completions for the current session:

$ llm completion fish | source

To load completions for each new session, execute this once:

$ llm completion fish > ~/.config/fish/completions/llm.fish

Verbose Mode (-v or --verbose)

See detailed output, including API requests and responses, useful for debugging.

llm -v "hello world" # For debugging use --debug

Note

This is a personal tool. It works well, but isn't built for production workloads. Use at your own risk.