Gv

Gv is an autonomous Bring Your Own Model (BYOM) AI coding agent that runs right inside VS Code, built specifically for local models and self-hosted inference.

With tools to create and edit files, explore large projects, inspect linter diagnostics, use the browser for UI testing, and execute terminal commands (with your approval), Gv goes far beyond simple code completion. It provides an intuitive human-in-the-loop GUI to review every proposed file change and terminal command before execution, giving you full control and complete privacy over your codebase.

1. Bring Your Own Model (BYOM): Local & Self-Hosted AI

Gv eliminates lock-in to proprietary cloud APIs. Connect directly to your choice of local or private endpoints:

Ollama: Run open-weights models like qwen2.5-coder, llama3.3, deepseek-r1, mistral, and codellama locally on your machine at http://localhost:11434.
LM Studio: Connect directly to LM Studio's local server at http://localhost:1234/v1.
vLLM: Connect to high-throughput production inference servers using custom base URLs and token authentication.
OpenAI-Compatible APIs: Connect to any self-hosted, private gateway, or third-party endpoint conforming to /v1/chat/completions.

2. Autonomous File Creation & Editing

Gv creates and modifies files directly in your workspace:

Side-by-Side Diff View: Every edit is presented as an editable diff. You can review, refine, or revert changes before accepting them.
Proactive Linter Integration: Gv inspects compiler diagnostics and linter errors (syntax issues, missing imports, type mismatches) as it codes, fixing them automatically.
Timeline & Checkpoints: Workspace snapshots are taken at every step so you can compare and restore your files at any point.

3. Terminal Command Execution

Gv can execute commands in your terminal to build, test, and run your software:

Run tests (npm test, pytest, cargo test, bun test).
Manage dependencies and install packages.
Monitor dev servers and background processes, reacting dynamically when errors occur.
Human-in-the-loop: Every command prompts for your permission before executing (or can be auto-approved per your settings).

4. Interactive Browser Testing

For web development workflows, Gv can launch a browser instance, navigate pages, click buttons, enter inputs, and capture screenshots and console logs. It can reproduce visual bugs and verify front-end fixes autonomously.

5. Extensibility with Model Context Protocol (MCP)

Extend Gv's capabilities using the Model Context Protocol:

Connect to external databases (PostgreSQL, SQLite, MySQL).
Integrate with Git repositories, documentation search engines, and enterprise tools.
Add custom MCP tools directly from the extension settings.

6. Rich Context Mentions

Easily bring context into the chat:

@file: Mention a specific file to include its contents in the prompt context.
@folder: Include an entire folder hierarchy and files.
@problems: Pass active workspace compiler and linter errors to Gv for instant fixes.
@url: Fetch web pages and documentation directly as clean markdown.

Getting Started

Install the Gv extension (gv-*.vsix).
Click the Gv icon in the Activity Bar.
In the Settings tab (gear icon), select your preferred provider:
- Ollama (http://localhost:11434)
- LM Studio (http://localhost:1234/v1)
- vLLM (custom server URL & API key)
- OpenAI-compatible (any custom /v1/chat/completions endpoint)
Select your model and start coding!

Gv

gv

Gv