Grom — Free, Open-Source AI Coding Assistant for VS Code (Ollama, LM Studio, Anthropic, and More)

I've been building Grom, a free and open-source VS Code extension that brings agentic AI coding to your machine. No telemetry, no mandatory account, no subscription. If you use Ollama or LM Studio, nothing ever leaves your machine.

What is it?

Grom is a chat + agentic coding extension that lives in the VS Code sidebar. You can talk to it like a regular AI assistant, or switch it into BUILD mode where it reads files, writes code, searches your codebase, and runs terminal commands autonomously — pausing for your approval before anything destructive happens.

Key features

Chat

Streaming chat with PLAN and BUILD modes
Multiple sessions, compact history, export to Markdown
Persistent memory — custom instructions injected into every chat
Per-session system prompt override

Agentic loop

Built-in file tools: read, write, delete, search, list directory, run terminal
MCP (Model Context Protocol) server support
Per-action approval for destructive operations — nothing gets written or deleted without you seeing it first
Diff-aware undo — after the agent writes files, an undo button appears so you can revert any or all of the changes
Task log showing every tool call with args and results

RAG

Your codebase is automatically indexed with BM25 + optional semantic embeddings via Ollama
Relevant files are attached to every message without you having to think about it

Inline autocomplete

Ghost-text completions as you type
Adaptive debounce — slows down automatically when you rarely accept suggestions
Word-by-word partial accept
Per-language model routing

@ Context mentions

@filename — attach any workspace file
@selection — currently selected code in the editor
@git — your uncommitted diff
@terminal — recent terminal output
@problems — all VS Code errors and warnings
@url:https://... — fetch and attach a web page
@docs — search indexed documentation sources

Providers
Works with pretty much everything:

Local: Ollama, LM Studio, Open Code
Cloud: Anthropic (Claude), OpenAI (GPT-4o), Groq, Mistral, Gemini
Custom: any OpenAI-compatible or Anthropic-compatible endpoint

API keys are stored in the OS keychain — never in settings.json.

Why I built it

I wanted something like Cursor or GitHub Copilot but with full control over where my code goes. Most existing extensions either require a cloud subscription, send your code to a third party by default, or don't support local models well. Grom is designed so that local-first is the default, and cloud providers are opt-in extras.

Quick start

Install from the VS Code Marketplace
Install Ollama and run ollama pull qwen2.5-coder
Open the Grom panel from the activity bar and start chatting

For cloud providers, select one from the dropdown and paste your API key when prompted — it gets stored in the OS keychain immediately.

推荐订阅源

DEV Community