Shellmate

Shellmate is a CLI tool that uses AI to translate a description into a valid shell command. It supports local models using Ollama, or OpenAI models.

NOTE: AI models, especially less powerful ones running locally on your computer, may hallucinate and generate commands that are incorrect or invalid. You should verify the generated command is correct if there is a chance it could modify or delete something that you don't want it to.

Installation

npm install -g shellmate

Configuration

Create a file in your home directory called shellmate.json. It should contain a provider property. This is the AI provider to use. To use a local LLM with Ollama, set this to ollama. To use the OpenAI API, set this to openai.

See the sections below for provider configuration options.

Using Ollama

In shellmate.json, set provider to ollama. Then, add an ollama property to the configuration with the following properties:

• model: The name of the model to use. Make sure you have already downloaded this model by running ollama pull <model>.

Note that Ollama must already be running for Shellmate to be able to communicate with it.

Here is an example shellmate.json configured to use Ollama:

{
    "provider": "ollama",
    "ollama": {
        "model": "llama3.2"
    }
}

Using OpenAI

In shellmate.json, set provider to openai. Then, add an openai property to the configuration with the following properties:

• apiKey: Your OpenAI API key.
• model: The name of the OpenAI model to use.

Here is an example shellmate.json configured to use OpenAI:

{
    "provider": "openai",
    "openai": {
        "apiKey": "<your-api-key>",
        "model": "gpt-4o-mini"
    }
}

Running Shellmate

Run the shellmate command, specifying a description of the command you'd like to generate enclosed in quotes:

shellmate "Generate a 2048-bit RSA ssh key"

Shellmate will talk to the model, then return the command for you to run.