Quick start
Python 3.10 or later is required. Run commands from the repository root. The daily flywheel currently requires macOS or Linux because it uses Unix file locks; native Windows is not supported for this entry point.
1. Install
python3 -m venv .venv
. .venv/bin/activate
python -m pip install -e .
cp -n .env.example .env
If .env already exists, add missing fields without overwriting it.
2. Choose a workflow
| Input or goal | Guide |
|---|---|
| Daily collection, visual processing, and training data | Daily data flywheel |
| Local financial PDF or existing OCR | Reference guides |
| Text export from v5/v6 releases | Existing release export |
3. Configure and run
Configure OCR, the vision model, the independent reviewer, and the target tokenizer as described in Model configuration. Confirm training admission in config/flywheel.json.
# Local configuration checks; no service requests
python -m workflow.flywheel_cli preflight
# Run once after configuration passes
python -m workflow.flywheel_cli run
python -m workflow.flywheel_cli status
See Configuration troubleshooting for preflight checks and missing fields.
4. Schedule daily runs
Complete a small real run and inspect visual descriptions and accepted candidates before installing cron or launchd.
Read and search
Use the sidebar on desktop or the menu on mobile. Each language has its own search index. Switch languages using the header; commands and environment variable names are identical across editions.