fateweaver.nvim

A Neovim plugin that provides intelligent code completion and predicts future code changes using a local LLM served over Ollama.

Requirements

• Neovim 0.8.0+
• Local chat-capable LLM: The plugin expects a model served by Ollama's chat endpoint by default (http://localhost:11434/api/chat)

Installation

Using lazy.nvim

{
  "Axot017/fateweaver.nvim",
  dependencies = { "nvim-lua/plenary.nvim" },
  config = function()
    require("fateweaver").setup({
      -- Configuration options (see below)
    })
  end,
}

Using packer.nvim

use {
  "Axot017/fateweaver.nvim",
  requires = { "nvim-lua/plenary.nvim" },
  config = function()
    require("fateweaver").setup()
  end
}

Configuration

require("fateweaver").setup({
  -- Logging configuration
  log_level = "ERROR", -- "ERROR", "WARN", "INFO", "DEBUG"
  logger_fn = vim.notify,
  
  -- Context and tracking options
  context_opts = {
    max_tracked_buffers = 5,          -- Maximum number of buffers for which plugin will store changes history
    max_history_per_buffer = 3,       -- Maximum number of recent changes to keep in history per buffer
    context_before_cursor = 80,       -- Number of lines before cursor to include as context for the LLM
    context_after_cursor = 80,        -- Number of lines after cursor to include as context for the LLM
  },
  
  -- Model configuration
  completion_endpoint = "http://localhost:11434/api/chat",
  model_name = "hf.co/Axottee/fateweaver-30B-A3B-GGUF:Q4_K_M",
  
  -- Performance options
  debounce_ms = 1000, -- Debounce time for completion requests
})

Usage

Basic Commands

-- Request completion for current buffer
require("fateweaver").request_completion()

-- Accept the current completion suggestion
require("fateweaver").accept_completion()

-- Dismiss the current completion
require("fateweaver").dismiss_completion()

Key Mappings (Example)

-- Add these to your Neovim configuration
vim.keymap.set("n", "<leader>fc", require("fateweaver").request_completion, { desc = "Request completion" })
vim.keymap.set("i", "<C-y>", require("fateweaver").accept_completion, { desc = "Accept completion" })
vim.keymap.set("i", "<C-x>", require("fateweaver").dismiss_completion, { desc = "Dismiss completion" })

Setting Up a Model

The plugin now sends requests in chat format so they match Fateweaver-style training data:

• system: You are a code completion assistant...
• user: ### Recent Edits: + diff + ### Code Excerpt: + fenced code excerpt

By default it uses Ollama's native chat endpoint, but OpenAI-compatible chat endpoints should also work.

Example: Using Ollama

1. Install Ollama: Follow the Ollama installation guide
2. Download your model:

   ollama pull hf.co/Axottee/fateweaver-30B-A3B-GGUF:Q4_K_M

   Replace this with your own trained model tag if needed.
3. Start Ollama server:

   ollama serve

The plugin will automatically connect to the Ollama chat endpoint at http://localhost:11434/api/chat.

Alternative Methods

You can also use other inference engines that provide chat-compatible APIs:

• llama.cpp
• vLLM
• llamafile
• Any other OpenAI-compatible server

Simply update the completion_endpoint in your configuration to point to your chosen inference server. For OpenAI-compatible servers, use a chat endpoint such as /v1/chat/completions.

Local testing with your normal Neovim config

You can launch Neovim with your usual config and this local plugin repo injected on top with:

mise run test

This runs the task file:

.mise/tasks/test

This script:

• prepends the current repo to runtimepath after your normal config loads
• keeps your normal Neovim config enabled
• calls require("fateweaver").setup() automatically if the plugin was not already loaded by your config
• sets log_level = "DEBUG"
• logs via require("fateweaver.logger").file_logger(...)
• uses FATEWEAVER_COMPLETION_URL for the completion endpoint from .mise/config.toml
• uses FATEWEAVER_LOG_FILE from .mise/config.toml for file logging
• adds an insert-mode <C-y> mapping that calls require("fateweaver").accept_completion()

Adjust .mise/config.toml to match your local environment before running the task.

Note: your regular config still needs nvim-lua/plenary.nvim available, since Fateweaver depends on plenary.curl.

Useful variants:

# Start via mise
mise run test

# Open a specific file
mise run test lua/fateweaver/init.lua

# Override the completion endpoint for one run
FATEWEAVER_COMPLETION_URL=http://localhost:11434/api/chat mise run test

# Override the log file path for one run
FATEWEAVER_LOG_FILE=/tmp/fateweaver.log mise run test

# Keep your normal config, but skip the automatic setup() call
FATEWEAVER_SKIP_SETUP=1 mise run test

# Run the task file directly if you want
./.mise/tasks/test lua/fateweaver/init.lua

When launched this way, <C-y> in insert mode accepts the current Fateweaver completion.

Contributing

This project is in early development. Contributions, issues, and feature requests are welcome!

License

MIT License - see the LICENSE file for details.

Acknowledgments

• Local open-weight code models and the Ollama ecosystem
• Cursor for inspiration on AI-powered code editing