A local, self-hosted data-broker removal assistant — a subscription-free reimplementation of Incogni's core workflow that keeps your personal data on your own machine.
Incogni is a paid service that acts as your authorized agent to get your data deleted from ~400+ data brokers under CCPA/GDPR. This tool does the same job locally. Because you are the data subject (not a third-party agent), your requests carry full legal weight with no authorization paperwork, and your PII never leaves your computer.
- Broker directory — 130 curated high-priority brokers (124 US, 6 GDPR/UK-GDPR;
people-search, marketing, risk, recruitment) with opt-out URLs/emails, contact
method, legal basis, and difficulty. Seeded from
data/brokers.json; easy to extend. Brokers whose sites have been taken down (currently Radaris, Rehold, Verecor) are flaggeddefunct: kept with their history, but excluded from the dashboard count, sending, and scanning. - Local profiles — one or more people's identifiers (name, aliases, emails, phones, addresses, DOB, state), stored in SQLite, used only to fill request templates.
- Request generation — per-broker legal deletion + opt-out requests from Jinja2 templates. GDPR/UK-GDPR brokers get the GDPR template; US brokers get CCPA if the profile lives in California, otherwise a generic multi-state one.
- Sending — each broker gets an action page: for email brokers, a
mailto:link and copyable text; for form brokers, the opt-out URL plus step-by-step instructions and your details to paste. Email-capable brokers can also be sent automatically in bulk from the Send page (opt-in, see below); form brokers always stay manual. - Status tracking — per-broker pipeline (
not started → sent → confirmed / rejected / needs verification) with history, plus a dashboard with counters and a "due for follow-up" list. California profiles also get a pointer to the state's DROP one-shot deletion platform. - IMAP reply monitoring — point it at a mailbox/label; it correlates broker
replies (via a
[PIR-<id>]subject tag), classifies them, and auto-advances statuses. Ambiguous mail goes to a review queue; it never sends, replies, or deletes. - Recurrence — sent requests get a follow-up date (60 days people-search, 90 otherwise) surfaced on the dashboard when due.
- Exposure scanning — a local Playwright browser session checks a
broker's public search page for your profile and scores any listing
against it (name, age, location, alias, phone), never on a bare
name-appears-on-page match. Four outcomes: found / not found /
possible match (low-confidence, you confirm or dismiss) / couldn't
check (blocked/timeout, falls back to a manual search link). Go to the
Scan page and click Scan (single broker) or Scan all; each
scan launches a real browser and is rate-limited, so expect ~10-25s per
broker, more for a bulk run. Brokers in the same
network(shared owner or database) inherit a derived likely exposed verdict when one sibling is found. Coverage is partial. 19 brokers have a search URL (scankey indata/brokers.json). 10 of them are auto-scanned, but only 3 (USPhonebook, Advanced Background Checks, AnyWho) have selectors verified against a live page; the rest use unverified selectors or the generic fallback. The other 8 are manual-link only (JS wizards / bot walls). Every other broker shows up under "Not publicly searchable" until someone adds and verifies ascanconfig for it (python -m scripts.verify_scan).
python3 -m venv .venv
.venv/bin/pip install -r requirements.txt
.venv/bin/playwright install chromium # one-time, for exposure scanning
.venv/bin/python -m scripts.seed_brokers # load broker registry
.venv/bin/uvicorn app.main:app --port 8137 # http://127.0.0.1:8137Then open the dashboard, add a Profile, and work through the brokers. Open it
via 127.0.0.1 or localhost; other hostnames are rejected.
Copy config.example.toml to config.toml, set [imap] enabled = true, and
fill in host/username/app-password. Use a dedicated mailbox or Gmail label
and an app-specific password. Send your removal emails from that address so
replies land where the poller can see them, and keep the [PIR-N] subject tag
intact. Click Check inbox on the dashboard to poll.
Set [smtp] enabled = true in config.toml with the same mailbox as
[imap] — broker replies must land in the mailbox being polled, so auto-sent
requests use no Reply-To and rely on the From address being that mailbox.
On the Send page, Preview renders and stages every eligible
not-yet-sent email/both broker without sending anything; Send for real
opens one SMTP connection and works through the (editable) selection with a
jittered pause between messages. Only brokers still not_started and not
verdicted not_found are eligible; follow-up re-sends and form-only brokers
always stay manual. A full run of the ~73 email-capable brokers takes several
minutes and is well under Gmail's ~500 recipients/day limit.
To track sites that aren't in the shared registry, create
data/brokers.local.json (gitignored) as {"brokers": [...]}, with entries in the
same schema as data/brokers.json, then re-seed. A local name that duplicates an
existing broker makes the seed fail. Removing an entry later leaves its DB row in
place, same as removing one from brokers.json.
IMAP and SMTP verify TLS certificates, so self-signed mail servers won't connect.
The app accepts only 127.0.0.1/localhost as Host and rejects cross-origin
POSTs, which stops other websites from submitting forms to it (CSRF and DNS rebinding).
.venv/bin/python -m pytestCovers template rendering, status transitions / follow-up dates, reply classification, SMTP sending, routes (including the Host/Origin checks), broker data and seeding, and the scan pipeline (match scoring, HTML extraction, rate limiting, outcome persistence) — all against fakes and saved HTML fixtures, no live network calls.
app/ main.py (routes) · config.py · db.py · models.py · templater.py · inbox.py
sender.py · send_service.py
scanner.py · fetcher.py · extract.py · matcher.py · ratelimit.py · scan_service.py
templates/ (HTML) · static/ (style.css, fonts)
data/ brokers.json · brokers.local.json (optional, gitignored)
request_templates/ (ccpa/gdpr/generic + _identity)
scripts/ seed_brokers.py · verify_scan.py
tests/ fixtures/ (synthetic scan pages)
mockups/ ui-preview.html
- Opt-out URLs and privacy emails change often. Verify a broker's current
process before relying on it; update
data/brokers.jsonand re-seed. - Some brokers require phone/ID verification or re-list data quickly (flagged by
the
difficultyfield and notes) — those need manual follow-up. - Scanning uses the first address line only.
scanner.search_context()takes the city from it ("street, city, ST zip"and"street, city ST zip"both work). A profile with no parseable city can't scan brokers whose search URL needs{city}; those show as couldn't-check rather than searching nationwide. - This is a personal tool, not legal advice.
Automatic email sending shipped (see above). Remaining seams:
- Web-form automation. For
contact_method = formbrokers, drive the opt-out flow with Playwright behind a shareddeliver()call, dispatched bybroker.contact_method. Expect friction from CAPTCHAs and phone/email verification — keep the manual fallback. - Scheduled follow-ups. A background scheduler that re-sends due requests on their cadence instead of surfacing them for a manual click.
- Broker list expansion. Grow toward the full ~400+ using the public Big Ass Data Broker Opt-Out List and the California DROP registry.