Vectra for VS Code

Your own AI pair programmer — running on your machine, not someone else's GPU.

Vectra is a repo-aware coding agent for VS Code. Point it at a GGUF model and llama.cpp, and it reads your code, proposes reviewed edits, and runs approved commands — no API key, no per-token bill, no code leaving your laptop.

Why Vectra

• Local-first — run instruction-tuned GGUF models through llama.cpp; your model and prompts never leave your machine.
• Repo-aware agent — reads, searches, and edits your workspace; every write is a reviewed diff, never a silent change.
• Approved execution — files, tests, and commands only run after you confirm them.
• Documents built in — parse and generate PDF, DOCX, PPTX, XLSX, RTF, code, and images.
• Any model, any hardware — switch to OpenAI, Anthropic, Gemini, or Ollama anytime; Auto/GPU/CPU device modes adapt to what you've got.

Quick start

1. Install llama.cpp so llama-server is on your PATH.
2. Download an instruction-tuned .gguf model (a quantized 3B–4B model is a good start).
3. Open the Vectra sidebar → Local Model → pick the file → Test.

Prefer a cloud model? Skip straight to API Key and pick a provider instead.

Requirements

VS Code 1.90+. llama.cpp only if you're running a model locally.

Privacy

Local prompts never leave your machine; cloud requests are sent only when you choose a cloud provider. Every edit is reviewed before it's applied, and command execution always asks first. Details: PRIVACY.md · SECURITY.md.

Created by Sudip Laudari. Source & issues · Changelog · Proprietary — see LICENSE.txt.