A Cursor-style agent panel for the Context Engineering thesis project. Brings the full retrieval → prune → assess → inject → generate pipeline directly into VS Code as a sidebar chat.
Features
Live pipeline steps — watch each stage animate in as it completes (retrieve, prune, assess, inject, generate)
Streaming LLM output — tokens stream in character by character like Cursor
Model picker — auto-populated from your local Ollama installation
Pipeline mode — choose Auto-Route (SELF_ROUTE), RAG, or Long Context (LC)
Context chips — attach the active file or your current selection to the prompt with one click
Index workspace — trigger ChromaDB indexing directly from the panel
Auto server management — the extension can start/stop the FastAPI backend for you
Requirements
Python environment with the backend dependencies installed (pip install -r backend/indexing/requirements.txt)
Ollama running locally with at least one model pulled (e.g. ollama pull qwen3.5:9b)
VS Code 1.85+
Development Setup
cd vscode-extension
npm install
npm run compile
Then press F5 in VS Code to open a new Extension Development Host window.
Configuration
Setting
Default
Description
contextEngine.backendPort
8000
Port for the FastAPI backend
contextEngine.pythonPath
python
Path to Python interpreter
contextEngine.defaultModel
qwen3.5:9b
Default Ollama model
contextEngine.autoStartServer
false
Start backend automatically
Usage
Open a workspace containing the project
Click the ⬡ Context Engine icon in the Activity Bar
Click ▶ Start to launch the backend server (or start it manually: cd backend && uvicorn mcp_server.server:app)
Click ⊕ to index your workspace into ChromaDB
Start chatting — select a model and pipeline mode at the bottom of the panel
