Memosk - AI Coding Assistant for VS Code

Memosk is your AI-powered coding sidekick that lives in your VS Code Activity Bar. Chat, explain code, fix bugs, run tests - all with privacy-first local Ollama fallback.

🚀 Quick Start

1. Install dependencies:

npm install

2. Build:

npm run build

3. Run in dev mode (F5)
4. Load in VS Code: Extensions > ... > Install from VSIX (package after build)

🧠 Recommended Ollama Setup (Free/Local)

Primary: cline + qwen2.5-coder:7b (best agent+model combo)

ollama pull qwen2.5-coder:7b
ollama pull cline

Alternatives:

• opencode + qwen2.5-coder:7b
• deepseek-coder-v2
• codestral
• qwen2.5-coder:14b (heavier/better)

Set memosk.defaultOllamaModel in settings.

🔌 Providers

📱 Features

• Activity Bar Chat (Memosk: Open Chat)
• Explain Selection/File (Ctrl+Shift+P)
• Improve Code (select → command)
• Terminal/Problems Inspection (auto-capture errors)
• Tagging (label files/terminals for context)
• Run Tests (memosk.testCommand)
• Streaming responses
• Privacy: No uploads unless explicitly enabled+confirmed

🛡️ Privacy

• Local first (Ollama)
• Workspace files never sent without:
  1. memosk.privacy.uploadWorkspaceFiles: true
  2. Per-session confirmation dialog
• Terminal outputs sanitized (paths/tokens stripped)

🧪 Test Checklist

• [ ] Activity Bar shows Memosk icon
• [ ] Chat view opens
• [ ] memosk.ask quick input works
• [ ] Commands: explain/improve work with selection
• [ ] npm run build succeeds
• [ ] F5 loads extension without errors
• [ ] Ollama responds (with model pulled)

Troubleshooting

• Ollama not responding: ollama serve, check http://localhost:11434
• No model: ollama pull qwen2.5-coder:7b
• CORS: Ensure Ollama allows browser requests

Recommended Combos (Settings → Model Routing)

1. Daily: cline + qwen2.5-coder:7b
2. Heavy: gpt-4o-mini (OpenAI)
3. Code-only: deepseek-coder-v2

Happy coding! 🚀