jellycell

A living catalogue for reproducible analyses. Plain-text notebooks, content-hashed output caching, live HTML viewer. Agent-friendly from day one.

What it does

Plain-text notebooks

.py files in jupytext percent format with PEP-723 dependency declarations. Diffable, greppable, uv run-able.

Content-addressed cache

Every cell’s output is keyed on (source, deps, env). Re-run is a cache hit; change a cell, only the affected subgraph re-executes.

Live HTML catalogue

jellycell view serves a project-wide report with side-nav, artifact browser, and SSE-backed live reload while you edit.

Agent-friendly

Every command supports --json. jellycell prompt --write drops AGENTS.md + CLAUDE.md at your repo root so Cursor, Codex, Copilot, Claude Code, and Aider pick up the canonical agent guide automatically.

Start here

  Getting started

Install, scaffold a project, run your first notebook end-to-end in under five minutes.

Getting started
  User guide

Project layout, notebook file format, jc.* API, tearsheets, artifacts, the CLI — everything you touch day-to-day.

Project layout
  Reference

The living architecture + §10 contracts + internals. Authoritative source for how jellycell is built and what it promises.

Reference
  Contributing

Dev setup, versioning policy, how to add commands + lint rules, how releases are cut. Read before opening a PR.

Contributing

Install

pip install jellycell            # CLI only
pip install 'jellycell[server]'  # with `jellycell view`

Requires Python ≥ 3.11.

Contents