Llama VS Code Agent

Local VS Code extension that uses Ollama as a coding agent.

Features

• Ask the model questions about the current project
• Read and search files
• Write files with confirmation
• Run terminal commands with confirmation

Requirements

• VS Code
• Ollama running locally
• A local model pulled, for example:

ollama pull llama3.1:8b

Setup

npm install
npm run build

Press F5 in VS Code to open the Extension Development Host.

Suggested prompts

• Find possible re-render issues in the current file
• Search where this hook is used in the workspace
• Run tests and explain the main failure
• Refactor this component with minimal changes

Safety

• File writes require confirmation
• Terminal commands require confirmation

18) First improvements after MVP

1. Add a sidebar chat with webview.
2. Show diffs before applying writes.
3. Limit writable paths to the workspace.
4. Add read_many_files and list_files tools.
5. Add streaming responses.
6. Add patch-based edits instead of full file overwrite.
7. Add ripgrep integration for faster search.
8. Add ignore rules like .gitignore and custom allowlists.

19) Good first commit sequence

git init
npm init -y
npm install -D typescript esbuild @types/node @types/vscode vsce

Then create the files above and run:

npm run build

If Ollama is not running, start it and test:

ollama run llama3.1:8b

20) MVP scope recommendation

For your first version, keep only this:

• command palette integration
• ask prompt
• read file
• search workspace
• write file with confirm
• run command with confirm

Do not start with:

• autonomous edits in many files
• git automation
• background loops
• webview UI
• embeddings/RAG

Make it work small first.

code-loom