BulletJournal is a single-project notebook orchestration platform for reproducible data science. It layers explicit artifact passing, persistent graph state, stale detection, checkpoints, and managed execution on top of Marimo notebooks, and it can run standalone or behind BulletJournal-Controller.

DISCLAIMER: This project is part of an experiment to evaluate the potential of AI tooling for software engineering. Most of the code in this repo was produced by an LLM and may not offer the same security or robustness as human-written code. Please don't deploy it in a critical production environment without isolation, especially given that the project was made to run user-provided Python code.

• FastAPI backend with REST + SSE updates
• Project format rooted in graph/, notebooks/, artifacts/, metadata/, checkpoints/, pyproject.toml, and uv.lock
• Strict AST parsing for notebook inputs, outputs, assets, and notebook docs
• Managed notebook execution in subprocesses with persisted artifact lineage and stale propagation
• File input nodes plus built-in and provider-discovered notebook and pipeline templates
• Checkpoint create/list/restore flows
• Zip import and export flows
• A bundled ReactFlow-based web UI for browsing projects, nodes, artifacts, issues, and events

• Python 3.11+
• A dedicated environment is recommended (uv or venv)

pip install bulletjournal-editor
bulletjournal init testproject
bulletjournal start testproject --open

If you are already inside a project root, running bulletjournal with no subcommand starts the app. When --project-id is omitted, BulletJournal derives it from the target directory name.

bulletjournal init testproject
bulletjournal init testproject --project-id custom-id
bulletjournal start .
bulletjournal dev . --open
bulletjournal doctor .
bulletjournal validate-templates
bulletjournal rebuild-state .
bulletjournal mark-environment-changed . --reason "dependencies updated"
bulletjournal export . testproject.zip
bulletjournal import testproject.zip restored-study

project_root/
├─ graph/
│  ├─ meta.json
│  ├─ nodes.json
│  ├─ edges.json
│  └─ layout.json
├─ notebooks/
├─ artifacts/
│  └─ objects/
├─ metadata/
│  ├─ project.json
│  └─ state.db
├─ checkpoints/
├─ uploads/
│  └─ temp/
├─ pyproject.toml
└─ uv.lock

• docs/EXECUTION_PLAN.md
• docs/ARCHITECTURE.md
• docs/PROJECT_FORMAT.md
• docs/NOTEBOOK_AUTHORING.md
• docs/API.md
• docs/TEMPLATES.md
• docs/OPERATIONS.md
• docs/TROUBLESHOOTING.md

python3 -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"
PYTHONPATH=src python -m pytest

ruff.toml is consumed by ruff automatically from the repo root. To enable the git hook:

pre-commit install

This repository now contains a functional MVP with tested core flows for:

• single-project startup
• interface parsing
• graph persistence and validation
• managed notebook runs
• artifact state transitions
• checkpoints and restore
• controller-facing status and environment invalidation hooks
• zip import/export
• SSE event streaming and API-driven UI updates

Known MVP constraints remain around deployment hardening and background execution semantics, but the repository is no longer just a skeleton.