Ollama Copilot allows users to integrate their Ollama code completion models into Neovim, giving GitHub Copilot-like tab completions.

Offers Suggestion Streaming which will stream the completions into your editor as they are generated from the model.

• Debouncing for subsequent completion requests to avoid overflows of Ollama requests which lead to CPU over-utilization.
• Full control over triggers, using textChange events instead of Neovim client requests.

• Language server which can provide code completions from an Ollama model
• Ghost text completions which can be inserted into the editor
• Streamed ghost text completions which populate in real-time

To use Ollama-Copilot, you need to have Ollama installed github.com/ollama/ollama:

curl -fsSL https://ollama.com/install.sh | sh

Also, the language server runs on Python, and requires two libraries (Can also be found in python/requirements.txt)

pip install pygls ollama

Make sure you have the model you want to use installed, a catalog can be found here: ollama.com/library

# To view your available models:
ollama ls

# To pull a new model
ollama pull <Model name>

Lazy:

-- Default configuration
{"Jacob411/Ollama-Copilot", opts={}}

-- Custom configuration (defaults shown)
{
  'jacob411/Ollama-Copilot',
  opts = {
    model_name = "deepseek-coder:base",
    stream_suggestion = false,
    python_command = "python3",
    filetypes = {'python', 'lua','vim', "markdown"},
    ollama_model_opts = {
        num_predict = 40,
        temperature = 0.1,
    },
    keymaps = {
        suggestion = '<leader>os',
        reject = '<leader>or',
        insert_accept = '<Tab>',
    },
}
},

For more Ollama customization, see github.com/ollama/ollama/blob/main/docs/modelfile.md

Ollama copilot language server will attach when you enter a buffer and can be viewed using:

:LspInfo

Smaller models (<3 billion parameters) work best for tab completion tasks, providing low latency and minimal CPU usage.

• deepseek-coder - 1.3B
• starcoder - 1B
• codegemma - 2B
• starcoder2 - 3B

Contributions are welcome! If you have any ideas for new features, improvements, or bug fixes, please open an issue or submit a pull request.

I am hopeful to add more on the model side as well, as I am interested in finetuning the models and implementing RAG techniques, moving outside of using just Ollama.

This project is licensed under the MIT License.