Local LLM Autocomplete

Ghost-text code completion for VS Code powered by a local LLM via any OpenAI-compatible endpoint (llama-server, ollama, etc).

Features

• Inline ghost-text suggestions triggered manually or automatically as you type
• Selection replacement — select code and trigger to have the LLM fix or improve it in place (undo with Ctrl+Z)
• Fully customizable prompts — tailor the completion and selection prompts to work best with your model
• Project context folder — drop text files, SQL schemas, instructions, or even images into a .llm-context folder in your workspace to give the model extra context
• Multimodal support — image files in the context folder are sent to vision-capable models
• Fully local — no data leaves your machine

Install from Release (easiest)

1. Download the .vsix from the Releases page
2. In VS Code: Ctrl+Shift+P → "Install from VSIX" and pick the file
3. Set your endpoint URL and model name in settings (search "Local LLM")

Build from Source

git clone https://github.com/lilblam/local-llm-autocomplete.git
cd local-llm-autocomplete
npm install -g @vscode/vsce
vsce package --allow-missing-repository
code --install-extension local-llm-autocomplete-0.0.5.vsix

Setup

1. Start your local LLM server (e.g. llama-server) with an OpenAI-compatible chat endpoint
2. Open settings and configure the endpoint URL and model name if they differ from the defaults

Settings

Setting	Default	Description
local-llm.endpointUrl	http://localhost:8013	Base URL of your server (don't include /v1 — it's appended automatically)
local-llm.modelName	Qwen3.5	Model name to request
local-llm.contextFolder	.llm-context	Folder in your workspace with extra context files
local-llm.enableAutoSuggest	false	Enable automatic suggestions as you type
local-llm.autoSuggestDelay	1000	Delay (ms) before auto-requesting a suggestion
local-llm.completionPrompt	(see below)	Prompt template for cursor-based completions
local-llm.selectionPrompt	(see below)	Prompt template for selection replacement
local-llm.imageContextPrompt	(see below)	Prompt appended when images are in the context folder

Usage

• Manual trigger: Press Ctrl+Shift+Space (Cmd+Shift+Space on Mac) to request a completion at your cursor position. The suggestion appears as ghost text — press Tab to accept or Escape to dismiss.
• Selection replacement: Select a piece of code, then press Ctrl+Shift+Space. The LLM will analyze the selected code and replace it with a corrected or improved version. Press Ctrl+Z to undo if needed.
• Auto-suggest: Enable in settings for automatic suggestions with a configurable delay.
• Context folder: Create a .llm-context folder in your workspace root and add any relevant files — schemas, documentation, instructions, or reference images.

Customizing Prompts

All prompts sent to the model are fully customizable in settings. Each prompt supports placeholder variables that get replaced at runtime.

Completion prompt (local-llm.completionPrompt) — used when triggering at a cursor position:

• {fileName} — name of the current file
• {textBefore} — all code before the cursor
• {textAfter} — all code after the cursor

Selection prompt (local-llm.selectionPrompt) — used when triggering with selected code:

• {fileName} — name of the current file
• {beforeSelection} — all code before the selection
• {selectedText} — the selected code
• {afterSelection} — all code after the selection

Image context prompt (local-llm.imageContextPrompt) — appended when image files are in the context folder:

• {imageCount} — number of images
• {imageNames} — comma-separated list of image file names

To edit these, open VS Code settings (Ctrl+,), search for "Local LLM", and modify the prompt fields. The settings UI provides a multiline text editor for comfortable editing.

To reset any setting to its default, click the gear icon next to it in the Settings UI and select "Reset Setting".

Output

Open the Output panel and select "Local LLM Autocomplete" from the dropdown to see request logs and diagnostics.