Codebuff Agent

JupyterLab extension for AI-powered code assistance with embedded agent services.

Architecture

This extension runs agent services in-process:

• Provides the JupyterLab frontend UI
• Handles Jupyter-specific integration (kernel communication, notebook manipulation)
• Runs LLM-based agent services directly via codebuff_core

Prerequisites

• JupyterLab: Version 4.0 or higher

Installation

Development Installation

# Install dependencies
yarn install

# Build extension
yarn build

# Install in JupyterLab (development mode)
pip install -e .
jupyter labextension develop . --overwrite

Starting the Development Environment

cd extensions/jupyter
jupyter lab

Configuration

Configure LLM settings in ~/.jupyter/codebuff_config.json:

{
  "provider": "vllm",
  "vllm": {
    "endpoint": "https://openrouter.ai/api/v1",
    "apiKey": "",
    "model": "openai/gpt-4o"
  }
}

Development

Watch Mode

# Terminal 1: Watch TypeScript
yarn watch:src

# Terminal 2: Watch labextension
yarn watch:labextension

Testing

# Run UI tests
yarn test:ui

# Run UI tests with browser visible
yarn test:ui:headed

License

MIT

Frontend lint

Run jlpm lint:check for ESLint (TypeScript/TSX under frontend/) and Stylelint (CSS under frontend/styles/). jlpm lint runs the same checks; jlpm lint:fix applies automatic fixes. Generated lib/, labextension bundles, and dependencies are excluded.

Prettier is a separate, opt-in formatting operation: jlpm prettier:check reports existing formatting differences and jlpm prettier rewrites frontend files. It is intentionally outside lint so restoring lint does not reformat the whole UI.

The existing codebase reports unused variables, redundant escapes, ANSI regexes, and CSS selector ordering/duplicates as warnings. Syntax and recommended correctness rules remain errors, including invalid React hook usage. Streaming while (true) readers and empty catch blocks are allowed intentionally. ESLint does not establish module reachability; that guard is tracked in issue #59.