Privacy: what is stored, what leaves your machine, how to delete it

Job Radar is a single-user tool that runs on your own machine. It has no accounts, no telemetry, no update check and no hosted service, and the pages it serves load nothing from the internet at run time (fonts are bundled). This page lists what it stores, which outside services can receive some of your data, and how to remove everything. It is a technical description, not legal advice.

What is stored, and for how long

Data	Where	Kept
Your profile (me.yaml: skills, years, languages, summary; preferences.yaml: target titles, cities, salary floor, excluded companies and keywords)	data/profile/ on the host	until you delete it
Every saved version of your profile (the full YAML text)	database table profile_snapshots	until you delete the database; there is no clean-up job
Configuration, including your own scoring criteria (ai.yaml)	data/config/, editor backups in data/config/.backup/	editor backups are capped; migration backups are not cleaned up
Applications, next steps, notes, favorites and their notes	database	until you delete the database (a favorite note goes when you remove the favorite)
Raw postings as received from each channel, and unparsed job-alert emails (whole message)	database table raw_postings	until you delete the database
Email metadata (message id, sender address, IMAP uid)	database table email_messages	until you delete the database
AI results (scores with reasons, translations, breakdowns, signals, trap marks, weekly suggestions with quotes) and AI usage records (model, tokens, duration)	database	until you delete the database
Telegram messages that were sent (payload and message id), control requests and their results	database	until you delete the database
Runner heartbeats	database	24 hours
Database backups	data/backups/	automatic: 14 daily and 8 weekly dumps; manual backups are never removed. A deleted record stays in older backups until those are removed.
Job description exports for resume-kit	data/exports/jd/	until you delete them
Logs	container logs (rotated), data/logs/ (rotated), ~/Library/Logs/job-radar/ on macOS	rotated
Secrets: Gmail app password, Telegram token and chat id, API keys	.env	until you delete it
LinkedIn login state (only if you use the logged-in session)	.auth/	until you delete it
Temporary files of an AI CLI call	$TMPDIR/jr-codex-*, jr-claude-*	removed after each call (a killed process can leave one behind)

There is no automatic retention for the database tables above yet: raw postings, unparsed emails, notifications, control requests and profile snapshots stay until you remove the database. Back up and keep data/ out of cloud-synced or shared folders: it holds all of the above. Backup files and the exported JD files contain personal data.

What leaves your machine

Who	What	When	How to stop it
The AI provider you use: OpenAI (codex CLI), Anthropic (claude CLI or API), DeepSeek or Xiaomi MiMo (API, off by default)	Scoring: your profile summary, skills and years, language levels, target titles, target cities, work modes, remote scope, seniority, salary floor, contract preferences, excluded companies, excluded keywords, your own scoring criteria and the job description. Weekly suggestions: the jobs you marked (interested, preparing, applied, interviewing, offer, ignored) with their status, plus your preferences. Translation, breakdown, signals, trap check: only the job text. Your home city and visa status (me.yaml), your notes and your application notes are not sent.	every cycle	switch a provider off (enabled: false in ai.yaml, or Settings → AI), do not install its CLI or fill its key; switch the weekly suggestions off in schedule.yaml (suggest.enabled)
Job sites and APIs: Arbeitsagentur, Indeed, LinkedIn, StepStone, XING, Google Jobs via JSearch, company career boards, Bright Data, Apify	your search keywords and cities, and requests from your IP address (LinkedIn session: requests as your logged-in account)	every cycle	switch the channel off on the Sources page
Tavily or Brave (board discovery)	a search query such as “keyword jobs city”	daily, only with a key	do not set the key
Google (Gmail over IMAP)	login with your app password and reading one label	every cycle	switch the Gmail channel off; revoke the app password
Telegram	job title, company, score, the one-line AI summary, reminders, health alerts and, with the “Tailor résumé” button, the command text	when something is pushed	Telegram is off by default (notify.yaml); leave it off
claude in a terminal (resume-kit /tailor)	your complete résumé and the job description, when you run the command yourself	only when you run it	do not run it

Nothing is sent to the project’s maintainers or to any analytics service.

Where the AI providers keep it

• Subscription CLIs (codex, claude) use the consumer terms of your ChatGPT or Claude plan. Job Radar starts them without a saved session, but how the provider uses the content (for example to improve models) depends on your account settings: check “Improve the model for everyone” in ChatGPT (Settings → Data controls) and “Help improve Claude” in Claude (Settings → Privacy), and switch them off if you do not want that.
• API keys are covered by the provider’s commercial or API terms, which usually do not train on your data by default; read the terms of the one you use.
• DeepSeek and Xiaomi MiMo are operated from China; data sent to them is subject to Chinese law. They are off by default; switch one on only if you accept that.
• Telegram stores bot messages on its servers. They are not end-to-end encrypted. Anyone with access to the chat, your account or the bot token can read them.
• A Gmail app password gives access to the whole mailbox, not only the label Job Radar reads. Prefer a dedicated mailbox that only receives job alerts, and revoke the password when you stop.

The providers that are switched on and reachable (Settings → AI, pnpm run doctor) are the services that can receive your profile summary and job text.

Remove everything

Do this when you stop using Job Radar or want to erase your job-search traces. It cannot be undone.

1. Back up what you want to keep (for example data/profile/ and data/config/), and copy it out of the project directory.
2. Uninstall the installation: stop the host AI worker, remove the containers, delete the database volume jobradar-pg (this is the step that deletes every stored job, application and note), delete data/, .env and .auth/, and the checkout. The commands are in Uninstall.
3. Backups and copies elsewhere: delete data/backups/ (it goes with data/), any backup you copied to another disk or cloud drive, and exported JD files and tailored résumés (data/exports/, your resume-kit output folder).
4. Logs and temporary files: ~/Library/Logs/job-radar/ on macOS, and $TMPDIR/jr-* if a killed AI call left something behind.
5. Accounts: revoke the Gmail app password (Google Account → Security → App passwords); delete the Telegram bot in BotFather (/deletebot) and clear the chat; log the LinkedIn session out; delete API keys you created only for this (Anthropic, DeepSeek, MiMo, JSearch, Tavily, Bright Data, Apify).
6. At the AI providers: the codex and claude CLIs keep no local session from Job Radar, but your provider account may keep a conversation history or usage log according to your plan: delete it there if you wish.

There is no function to delete a single job or application from the web interface in this version. Deleting the database volume (step 2) is the supported way to erase stored jobs and notes; start again with an empty one if you want to keep using Job Radar.