Panel is an early-build research workspace where an AI agent works alongside users, integrating chat, files, PDFs, and Jupyter notebooks in a single dock. The agent can create custom viewers and apps—called Panes—when built-in options are insufficient. Setup requires Node 22.18+ (or 24.12+), pnpm, uv (which fetches Python 3.12+), and Claude Code installed and signed in. Installation uses `pnpm install`, `uv sync`, and `pnpm start`, launching at `http://localhost:4173`. Data persists in `~/Panel/panel.db` and `~/Panel/workspaces`, surviving repo deletion.

Current features include file-aware chat with tool confirmation, isolated Workspaces with saved layouts, panes for code/PDFs/notebooks backed by real kernels, background command monitoring, and agent-generated custom panes. A literature review module and a "Hypothesis" module for structured research are included. Limitations: full support only for Claude Code; modules launch only via chat; no OpenAI API support for literature reviews or Hypothesis modules (though a key enables basic chat/tools).

The architecture centers on a Module Protocol with typed Inputs, Outputs, and Intermediates (e.g., chain-of-thought) for observability and validation. A Data Abstraction Layer (DAL) maps URIs to in-memory or filesystem objects, decoupling modules from storage details. Licensed MIT.