Vline — AI Coding Agent for VS Code

Vline is an AI coding agent that lives in the VS Code sidebar. Chat with cloud or local LLMs, let the agent read and edit files in your workspace (with one-click undo), attach images for vision, connect MCP tools, and even generate images — all without leaving your editor.

Status: Beta. Features are evolving quickly; feedback and bug reports are welcome.

✨ Features

Multi-provider chat — works with free-tier cloud models and fully local models:
- Ollama (local, auto port detection)
- Groq, Google Gemini, OpenRouter, NVIDIA NIM, Mistral AI, and any custom OpenAI-compatible endpoint
- Automatic FREE-model filtering (sofa.freeOnly)
Agentic file tools — the agent can create, read, edit, append and patch files in your workspace, with streaming previews of every change
One-click Undo — every file change the agent makes can be reverted instantly; files it created are removed cleanly
Plan / Act mode — restrict the agent to read-only planning, or let it execute changes
@-file mentions — type @ to add workspace files to the context
Image attachments & vision — attach screenshots and images for multimodal models
Image generation — connect custom image-generation models (OpenAI-compatible)
MCP support — connect Model Context Protocol servers (stdio & HTTP), including Claude/Cursor-style mcpServers configs
Reasoning modes — Low / Medium / High / Max thinking effort for reasoning models
Context window awareness — per-model context badges (128K / 256K / 1M) and automatic context trimming
Chat export — export your chat history as Markdown

📦 Installation

Install from the VS Code Marketplace (search for "Vline"), or install the .vsix manually:
- Command Palette → Extensions: Install from VSIX…
Open the Vline panel from the activity bar.
Sign in and add at least one provider (see below).

🔑 Configuration

Add your API keys in Settings → Extensions → Vline (keys are stored locally in VS Code settings):

Setting	Description
`sofa.providers.groq.key`	Groq API key
`sofa.providers.google.key`	Google Gemini API key
`sofa.providers.openrouter.key`	OpenRouter API key
`sofa.providers.nvidia.keys`	NVIDIA NIM API keys (supports rotation)
`sofa.providers.mistral.key`	Mistral AI API key
`sofa.ollama.host` / `sofa.ollama.port`	Local Ollama endpoint (auto-detected by default)
`sofa.freeOnly`	Show only FREE models
`sofa.maxToolLoops`	Max tool-call loops per question
`sofa.reasoning`	Enable reasoning (thinking) mode

Optional OAuth login (Google / GitHub) can be configured via sofa.auth.* — see LOGIN-SETUP.txt in the extension folder.

💬 Usage

Click the Vline icon in the activity bar.
Pick a model from the model picker (header chip or ✦ Model button).
Ask a question or describe a task. Use Plan mode to review the agent's plan before letting it change files.
Hover any file-change card and click Undo to revert that change.

🔒 Privacy

API keys are stored locally in your VS Code settings and are only sent to the provider you choose.
Chat content is sent only to the selected model provider (or stays fully local with Ollama).
The extension collects no analytics.

Vline Agent

Vline Corporation