Buddy AI Coding Agent

An AI coding agent for VS Code.

Ask questions in chat, let it read your repo, edit files, run commands, and search the web — with the provider and model you choose.

Buddy runs as @buddy in VS Code Chat or in a dedicated sidebar panel. It keeps conversation context per workspace, shows diffs before applying edits, and asks before running terminal commands (unless you opt into trusted mode).

Install

From the Marketplace

code --install-extension ishankrs.buddy-ai-coding-agent

Or search Buddy AI Coding Agent in the Extensions view.

From source (development)

git clone https://github.com/ishankrs/Buddy.git
cd Buddy
npm install
npm run compile

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

From a VSIX

npm run package

Then in VS Code: Extensions → … → Install from VSIX… → select buddy-ai-coding-agent-0.2.1.vsix.

Setup

1. Open the Command Palette and run Buddy: Select Provider and Model (or use the dropdowns in the sidebar panel).
2. Pick a provider and enter an API key when prompted. Keys are stored in VS Code SecretStorage — not in settings files.

Supported providers:

For a self-hosted or proxy API, use Buddy: Configure API Endpoint (URL + Key) instead.

Using Buddy

In VS Code Chat:

@buddy explain this function
@buddy /plan refactor the auth module
@buddy /debug why is this test failing?
@buddy /think how should I split this service?
@buddy /swarm add login end-to-end
@buddy /subagent write tests for auth.ts

Chat commands

You can also say things like “open a subagent to refactor the logger” — Buddy will pick that up in normal chat.

UI modes (buddy.uiMode)

• both — Chat and sidebar panel (default)
• chat — @buddy only
• panel — Sidebar only (activity bar icon)

Switch anytime via Buddy: Switch UI (Chat / Panel / Both).

What Buddy can do

The agent loop can call tools to:

• Read and search files in your workspace (requires rg in PATH)
• Edit files with a diff preview before apply
• Run terminal commands (with confirmation)
• Search the web and fetch public URLs
• Spawn subagents for isolated subtasks

It uses your active editor, selection, @ file references, and diagnostics as context.

Configuration

Common settings (Settings → search buddy):

buddy.provider          openai | anthropic | openrouter | ollama | custom
buddy.model             Model ID (provider-specific)
buddy.maxIterations     Tool loop limit per request (default 25)
buddy.autoApproveReadOnly   Auto-run read-only tools (default true)
buddy.trustedMode       Skip terminal approval (default false)
buddy.maxMemoryTurns    History kept per workspace (default 20)

Web search (buddy.webSearch.*) is on by default. Provider auto uses Serper, Brave, Tavily, or Google when keys exist; otherwise DuckDuckGo. Optional keys via Buddy: Set Web Search API Key.

Provider-specific base URLs: buddy.openaiBaseUrl, buddy.anthropicBaseUrl, buddy.openrouterBaseUrl, buddy.baseUrl (custom), buddy.ollamaBaseUrl.

Commands

The status bar shows the active provider and model; click it to change them.

Development

Requires Node 20+, VS Code 1.95+.

npm run compile      # build extension
npm run watch        # rebuild on change
npm run check        # typecheck
npm run package      # create .vsix

Requirements

• VS Code 1.95 or newer
• An API key for your chosen provider (except Ollama)
• ripgrep (rg) on PATH for workspace search

License

MIT — see LICENSE.