Good Buddy

Good Buddy is a VS Code extension that uses a local Ollama server for inline code completions, a chat view, and selected-code explanations.

Features

• 💻 Inline code completions using a local Ollama server.
• 🗨️ Chat view for interactive discussions and explanations.
• 🧩 Selected-code explanations to understand complex code snippets.

Usage

Once installed and configured, you can:

• Trigger inline completions by typing in the editor.
• Open the chat view via the Good Buddy: Open Chat command.
• Request explanations for selected code using the Good Buddy: Explain Selection command.

Contributing

Contributions are welcome! Please open issues or submit pull requests on the GitHub repository.

Installation

Good Buddy does not call any cloud API — it talks to a local Ollama server, so you need Ollama installed and running before the extension is useful.

1. Install Ollama and make sure it's running (ollama serve, or the desktop app's background service).

2. Pull at least one completion model and one chat model, for example:

   ollama pull qwen2.5-coder:7b
   ollama pull qwen3:8b
   

3. Install/enable the Good Buddy extension in VS Code, then open Settings and configure it under Good Buddy (search goodBuddy):

   • goodBuddy.endpoint — Ollama server URL. Defaults to http://localhost:11434, which is correct for a local install.
   • goodBuddy.completionModel — required, must match a model tag you've pulled (see suggestions below). The default (moophlo/Qwen3-Coder-30B-A3B-Instruct-GGUF:latest) is a large 30B model that may time out on modest hardware.
   • goodBuddy.chatModel — required, must match a model tag you've pulled (see suggestions below).
   • goodBuddy.inlineCompletionsEnabled — toggle ghost-text completions on/off (also available via the Good Buddy: Toggle Inline Completions command).
   • goodBuddy.maxContextLines — how many lines of surrounding code to send with each completion request.
   • goodBuddy.completionTimeoutMs — how long to wait for a completion before giving up; raise this if you use a larger model or have slower hardware.

   Good Buddy does not validate these model names — if a configured model isn't pulled locally, requests will fail and the error surfaces in the status bar tooltip and the Good Buddy output channel.

Suggested models

Pick models sized for your hardware; smaller models respond faster and are a better fit for inline completions where latency matters.

Completion (goodBuddy.completionModel) — should support fill-in-the-middle (FIM):

• qwen2.5-coder:7b — good balance of speed and quality, recommended default.
• qwen2.5-coder:1.5b — fastest option, best for low-resource machines.
• codellama:7b-code — alternative code-focused model.

Chat / explanations (goodBuddy.chatModel):

• qwen3:8b — matches the extension's default, good general-purpose reasoning.
• llama3.1:8b — widely used general-purpose alternative.
• deepseek-r1:7b — stronger reasoning if you don't mind slower responses.

Browse the full catalog at ollama.com/library for other sizes/quantizations that fit your hardware.

Project layout

• src/extension.ts registers VS Code commands and providers.
• src/config.ts is the single place that reads extension settings.
• src/features/chat/ contains the chat webview and its lifecycle.
• src/features/completion/ contains inline-completion behavior.
• src/features/explain/ contains selected-code explanation behavior.
• src/provider/ contains the provider abstraction and Ollama implementation.
• src/gateway/ contains the HTTP request helpers.
• out/ is generated JavaScript. Do not edit it directly.

Develop

1. Install dependencies with pnpm install.
2. Start Ollama and ensure the configured models are available.
3. Run pnpm run watch, then launch Run Good Buddy Extension from VS Code.

Run pnpm run check and pnpm run lint before committing. Use pnpm run build to create a production bundle.

Testing

This project now includes a Vitest-based unit-test setup for fast validation and coverage reports.

• Run unit tests with pnpm test
• Re-run automatically while editing with pnpm test:watch
• Generate a coverage report with pnpm coverage

The test suite uses mocked fetch responses for Ollama HTTP calls so you can verify request/response handling without a live local model server.

License

Good Buddy is available under the MIT License. You may use it for personal or commercial purposes, modify it, and redistribute it, subject to the license terms.

Attribution or citation is appreciated but not required. If Good Buddy is useful in your research, software, or commercial work, you can use the metadata in CITATION.cff or cite it as Good Buddy by the Good Buddy contributors.

Troubleshooting chat

The chat view reports the response body when Ollama returns malformed JSON rather than presenting an opaque parse error. Check the Good Buddy output channel for model-listing failures, and verify goodBuddy.endpoint points to the Ollama server (normally http://localhost:11434).

Workspace tools

Ask the chat to inspect the project, read a file, make a change, or run a command. Every request begins with the opened workspace path and a compact project tree, so the model has project context before it chooses a tool. It can list the project tree and read workspace files directly. For small changes it uses an exact replace_in_file edit; full-file writes are also supported. Both appear as an in-chat unified diff with Approve and Reject controls, and approval checks that the file has not changed since the proposal was generated. Shell commands retain a VS Code confirmation dialog. Tool paths are limited to the opened workspace, command output is capped, and an agent request can use at most five tools.

Use Attach in the chat footer to add up to five text files (80 KB each) to the next message. Selected filenames appear beneath the composer. Attachments are sent only with that next message and are then cleared; binary and oversized files are not attached.