local AI inline completions in Neovim.
Hits your local model with prefix+suffix around the cursor, streams back a completion, shows it as ghost text. Accept with <Tab>, dismiss with <Esc>, trigger manually with <M-CR>(alt+enter).
- Neovim >= 0.10 (uses
vim.text.diff) - Ollama running locally (default provider)
-- lazy.nvim
{
"luke/binford/geass.nvim",
config = function()
require("geass").setup({
provider = "ollama",
model = "qwen/qwen2.5-coder:1.5b",
url = "http://localhost:11434/api/generate",
})
end,
}| key | default | description |
|---|---|---|
provider |
"ollama" |
backend ("ollama" / "opencode") |
model |
"ibm/granite4:tiny-h" |
model name for the provider |
debounce |
300 |
debounce window (ms) |
throttle |
1000 |
throttle interval (ms) |
context_window |
16000 |
total chars sent to the model |
context_ratio |
0.75 |
fraction of window for prefix |
| action | default |
|---|---|
| trigger | <M-CR> |
| accept | <Tab> |
| dismiss | <Esc> |
- ollama provider
- automatic trigger on CursorMovedI
- multi-candidate cycling
- tests