tidyline

An offline, read-only command-line tool that audits a Things 3 database on macOS.

Unofficial. tidyline is not affiliated with or endorsed by Cultured Code. "Things" is a trademark of its owner. This project uses the name only to say which app it reads. It uses no logo or icon.

Your data is personal

The Things database, the JSON export, the text report and any decoded file contain your personal data: task titles, notes, checklists and dates. Never commit them to a repository, and never post them in a public issue.

Anonymised exports reduce the risk. They cannot remove all of it.

• light keeps titles but replaces emails, phone numbers, URLs, money amounts and long digit runs with placeholders. It removes notes. It is a best-effort scrub. It cannot catch names, addresses or alphanumeric account numbers.
• strict replaces every title and name with a label such as Task k3f9az. It removes notes and replaces checklist items with a count. Only strict removes all titles.
• At every level, ids are kept. Anyone who also has your database can map them back to titles. Dates, counts and findings stay too, and they reveal how you work.

What it does

tidyline reads your Things database and reports on it: counts, health checks, a decoded view of your repeating tasks, a JSON export, and an anonymised prompt you can give to an AI assistant yourself.

What it does not do:

• It never writes to the Things database.
• It never reads the database file directly. It makes a private copy (a snapshot) with the SQLite backup API and reads that, then deletes the copy. Things can stay open while it runs.
• It makes no network calls. None of the current commands needs a network.
• It never reads or exports the URL-scheme auth token.
• It never changes, completes or opens anything in Things.

Install

From source:

git clone https://github.com/rathodlaxman/tidyline.git
cd tidyline
uv sync
uv run tidyline --version

You need uv and Python 3.10 or later.

Once published: pipx install tidyline or uv tool install tidyline.

macOS may block Terminal from reading Things' data folder. If doctor says so, open System Settings, Privacy & Security, Full Disk Access, turn it on for your terminal app, then quit and reopen the terminal.

Quick start

Run these from the folder you cloned. Write outputs outside any git folder, so they cannot be committed by accident. The examples use ~/tidyline-out/, which must exist first:

mkdir -p -m 700 ~/tidyline-out
uv run tidyline doctor
uv run tidyline report
uv run tidyline check
uv run tidyline recurrence
uv run tidyline export --out ~/tidyline-out/export.json
uv run tidyline prompt --out ~/tidyline-out/prompt
uv run tidyline decode ~/tidyline-out/answer.md --out ~/tidyline-out/answer-decoded.md

• doctor finds the database and checks permissions, the schema and versions.
• report prints a text summary. It shows counts only, except for repeating templates, which are listed with their titles.
• check runs the health checks.
• recurrence lists every repeating template with its readable rule.
• export writes a JSON export. Add --anonymise light or --anonymise strict to anonymise it.
• prompt and decode are described below.

Every command accepts --db PATH for a database in a non-standard place. You can also set the THINGSDB environment variable. Add --verbose before the command for more detail when something fails.

Commands

Command	What it does	Options
doctor	Finds the database and checks permissions, schema and versions.	--db
report	Prints a text summary, or writes it to a new file.	--out PATH, --include-completed, --no-color, --db
check	Runs the health checks.	--only ID [ID ...], --pattern REGEX [REGEX ...], --json, --db
export	Writes a JSON export to a new file.	--out PATH (required), --anonymise {none,light,strict}, --db
recurrence	Lists every repeating template with its readable rule.	--raw, --db
prompt	Writes an anonymised payload and a filled audit prompt.	--anonymise {light,strict}, --context FILE, --out DIR, --copy, --db
decode	Replaces strict labels in an answer with your real titles.	ANSWER_FILE, --out FILE, --db

Files written with --out get mode 0600, and an existing file is never overwritten.

Exit codes:

Code	Meaning
0	Success. For check: no warnings.
1	Only check uses this: at least one warning-level finding.
2	Usage error, such as a bad option, a bad regular expression or an output path that already exists.
3	The run failed: the database could not be found or read, or a file could not be written.

Using an AI assistant without sending anything automatically

tidyline never sends your data anywhere. prompt prepares two files, and you decide whether to attach or paste them into an assistant yourself.

uv run tidyline prompt --context my-context.txt --out ~/tidyline-out/prompt

• The default is --anonymise strict. There is no none option for prompt.
• --context FILE is a plain text file with sections labelled Role:, Priorities: and Dates:. It is added to the prompt exactly as you wrote it. A section you leave out stays as a visible placeholder, and the preview warns you.
• --out DIR writes tidyline_payload.json (the anonymised export) and tidyline_prompt.md (the filled prompt). --copy puts the prompt on the clipboard with pbcopy. With neither option, nothing is written.
• prompt always prints a preview first: the level, the counts, what was removed or replaced, and your context text. The last line says nothing has been sent anywhere.
• The text report is never part of the output.

At strict, titles are hidden, so the assistant cannot judge how your tasks are worded. The prompt tells it to skip that part. Use --anonymise light if you want task wording reviewed, and accept that light leaves titles in place.

To turn the answer back into real titles, save it to a file and run:

uv run tidyline decode ~/tidyline-out/answer.md

decode recomputes each label from the current database, so it needs no mapping file. Labels it cannot match stay as they are, and it reports how many. It needs the same database the prompt was made from, and light output needs no decoding. Titles go in exactly as stored, so a title containing | can break a markdown table cell. The decoded text contains real titles, so treat it as personal data.

Health checks

tidyline check exits with 1 if any warning is found. Info findings do not change the exit code. Findings carry ids and fixed messages, never titles. The text output shows titles on your own screen only. check --json prints ids only.

Check	Severity	What it finds
open_in_trashed_project	warning	Open to-dos inside a trashed project. The other checks leave these out.
unfiled_open	warning	Open to-dos with no area and no project, outside the Inbox.
title_hygiene	warning	A stray · at either end, leading or trailing spaces, doubled spaces, and matches of your own --pattern regular expressions.
possible_duplicates	warning	Open to-dos whose titles match after Unicode normalisation, case folding, apostrophes removed and other punctuation treated as spaces.
dated_in_someday_project	warning	To-dos with a start date inside a Someday project. They will never surface.
active_project_no_open_tasks	warning	Active projects with no open to-dos and no live repeating template.
recurrence_not_recognised	warning	Repeating templates whose rule tidyline cannot read or does not recognise.
leftover_templates	info	Repeating templates whose project is trashed or missing.
someday_in_active_project	info	Someday to-dos inside active projects.
outside_heading	info	To-dos outside any heading in projects that use headings.

A finding for one of your own patterns says only "matches user pattern N". The text output on your screen shows the pattern next to it; the JSON does not.

Things does not document its database

tidyline reads columns that Things does not document, because there is no public API for this. These include:

• the repeating-task columns rt1_recurrenceRule, rt1_repeatingTemplate and rt1_instanceCreationPaused;
• startBucket, which marks This Evening;
• deadlineSuppressionDate, which keeps an overdue deadline out of Today.

Repeat rules are stored as property lists whose meaning was worked out by comparing them with what Things shows. A rule tidyline does not understand is reported as not recognised. It is never guessed.

Before reading, tidyline checks that the tables and columns it needs exist. If they do not, it stops with a message that names what is missing, the versions it was tested with, and what to do next. See docs/COMPATIBILITY.md for what has been tested and what has not.

Not yet built: the HTML report and the opt-in model integration.

Development

uv sync
uv run pytest
uv run ruff check .
uv run ruff format --check .

Tests use only a synthetic database built by tests/fixtures/build_fixture.py. The real Things database is never used in tests, and no database, export or report is ever committed. Network sockets are blocked while the tests run.

Golden files under tests/fixtures/ are regenerated with TIDYLINE_UPDATE_GOLDEN=1 uv run pytest. Review the diff before committing.

The design and the decisions behind it are in docs/DESIGN.md. How repeat rules are read is in docs/RECURRENCE.md.

Licence

MIT. See LICENSE. tidyline depends on things.py, which is licensed under Apache 2.0.