Cliff

Cliff (Command Line InterFace Friend) is an AI assistant that helps you come up with Unix commands. Given an objective (for example, "kill the process running on port 8080"), Cliff will generate a command that does the objective and add it to your paste buffer for you to easily paste into your terminal.

MacBookPro:~ $ cliff list my files and their sizes in KB, descending by size
ls -lS | awk '$5 > 0 {printf "%-20s %8.2f KB\n", $9, $5/1024}'
MacBookPro:~ $ ls -lS | awk '$5 > 0 {printf "%-20s %8.2f KB\n", $9, $5/1024}'
foo.txt                 10.29 KB
bar.txt                  5.28 KB
baz.txt                  2.61 KB

Cliff is compatible with LLMs from OpenAI, Anthropic, etc. as well as local models running with Ollama.

Why?

It's annoying having to open the browser when I forget how to do something in the terminal.

Requirements

• At least one of the following:
  • A valid API key from OpenAI, Anthropic, Google, Cohere, Groq, Replicate, Mistral, or Cerebras.
  • An LLM running locally with Ollama.
• A Unix-like operating system
• Python >= 3.9

Installation

You can install Cliff with homebrew:

brew tap pkelaita/cliff
brew install cliff

Or with pip:

pip install cliff-cli

Configuration

If you'd like to use models from an API-based provider, add its credentials as follows:

cliff --config add [provider] [api key]

The provider can be any of openai, anthropic, google, cohere, groq, replicate, mistral, or cerebras.

Otherwise if you want to use a local model, add it like this:

cliff --config add ollama [model]

Configuration tips

• In order to use local models, make sure you have Ollama installed and running and have the model loaded (their docs).
• If you add multiple models, you can set your default model with: cliff --config default-model [model].
• If you want to edit your config file directly, it's located at ~/.cliff/config.json.

For a full overview of the configuration system, run cliff --config help, and for a full list of supported models for each provider, see L2M2's docs.

Usage

Get started by running cliff with an objective.

cliff kill the process running on port 8080

Cliff will automatically add the command to your paste buffer, so no need to copy-paste it.

If needed (i.e., to avoid escaping special characters), you can use quotes.

cliff "kill the process that's running on port 8080"

If you want to specify which model to use, you can do so with the --model flag.

cliff --model gpt-4o kill the process running on port 8080

To view the man page, run cliff with no arguments.

---

Chat Memory

By default, Cliff has chat memory enabled with a sliding window size of 10 turns. You can view your memory with cliff --memory show and clear it with cliff --memory clear.

If you'd like to change the window size, run cliff --config memory-window [new size]. If you want to disable memory, just set the window size to 0.

---

Storing Command Outputs

Cliff's chat memory does not have access to command outputs, but you can optionally share them with Cliff to help it debug and improve its responses.

• To run a command and store its output for Cliff, run cliff -r [command] or cliff --recall [command].
• To view all recalled commands and their outputs, run cliff --show-recall or cliff -sr.
• To clear Cliff's recall storage, run cliff --clear-recall or cliff -cr.

Tip: You'll usually have to put quotes around your command if it contains special characters – e.g., cliff -r "ps ax | head -n 10" – for Cliff to properly capture and execute it.

---

That's it! It's pretty simple which is the point.

Planned Features

• Regular updates with new models, etc.
• UX improvements: cosmetic updates, model aliases, etc.
• Not sure what else this thing really needs, but open to suggestions!