Skip to main content

Runtime PII leak detector for Django — find personal data leaking from your API responses and third-party calls

Project description

Wiregraph

What sensitive user data is leaving your Django app — and where is it going?

Most teams can't answer that. Your view returns more than you think. Your OpenAI call ships customer emails to a third party. Your Stripe webhook echoes an SSN into a log line. Static analysis won't catch it. APM won't flag it.

Wiregraph is a runtime PII leak detector that sits inside your Django app and watches the traffic you're actually serving.

The alarm moment

A user hits /api/v1/support/ticket/. Your view enriches the response with an OpenAI summary. Wiregraph sees this:

[egress] POST api.openai.com/v1/chat/completions
  └─ EMAIL_ADDRESS  (confidence 0.99)  asset: request.body.messages[1].content
  └─ US_SSN         (confidence 0.95)  asset: request.body.messages[1].content

[response] 200 /api/v1/support/ticket/
  └─ PHONE_NUMBER   (confidence 0.90)  asset: response.body.ticket.notes

You didn't write code to log that. You didn't know it was happening. Now you do.

What you get

  • Runtime visibility into every PII-bearing field crossing your app's boundary — inbound, outbound, egress
  • Audit trail of which endpoints leak what, by tenant
  • Compliance evidence — exportable PDF/JSON reports of observed flows
  • Zero raw PII at rest — detections are hashed, masked, or truncated before storage

APM tells you a request was slow. Logs tell you what you decided to log. Wiregraph tells you what your app actually sent.

Quick start

pip install wiregraph
# settings.py
import wiregraph

INSTALLED_APPS = [*INSTALLED_APPS, *wiregraph.INSTALLED_APPS]
MIDDLEWARE = wiregraph.setup(MIDDLEWARE)
WIREGRAPH = {"ENABLED": True}
python manage.py migrate
python manage.py wiregraph_doctor   # sanity-check config

# attach your admin user to a tenant so events get attributed
python manage.py shell -c "from django.contrib.auth import get_user_model; from wiregraph_apps.tenants.models import Tenant, TenantMembership; u = get_user_model().objects.filter(is_superuser=True).first(); t, _ = Tenant.objects.get_or_create(name='Demo Co', defaults={'slug': 'demo'}); TenantMembership.objects.get_or_create(user=u, tenant=t)"

Hit any endpoint, then open /admin/wiregraph/dashboard/ to see the flow graph. (For the JSON API at /api/v1/, install wiregraph[drf] — see below.)

Wiregraph skips the Django admin URL prefix by default — otherwise the DataEvent admin would re-detect its own contents on refresh. Override with AUTO_EXCLUDE_ADMIN=False.

Install extras

Wiregraph ships in slices so you only pull in what you need.

Install line What you get
pip install wiregraph Core middleware + bundled admin dashboard at /admin/wiregraph/dashboard/. No DRF.
pip install wiregraph[drf] Adds the JSON API at /api/v1/ (viewsets, JWT auth, OpenAPI schema). Required by the React wiregraph-dashboard consumer.
pip install wiregraph[presidio] ML-based detection. Also run python -m spacy download en_core_web_lg.
pip install wiregraph[export] PDF/JSON compliance reports.
pip install wiregraph[postgres] psycopg[binary] for Postgres backends.
pip install wiregraph[all] Everything above plus dev tooling.

If you skip [drf], the /api/v1/ routes are not registered — the admin dashboard still works.

Security guarantee

Wiregraph never persists raw PII. Every detection is redacted (hash / mask / truncate, configurable) before it touches storage. The matched value lives in memory only long enough to classify and redact.

Detection

Regex out of the box: emails, phones, SSNs, credit cards, and more. Add your own:

WIREGRAPH = {
    "ENABLED": True,
    "CUSTOM_PATTERNS": [
        {"name": "emp_id", "regex": r"\bEMP-\d{6}\b", "confidence": 0.9},
    ],
}

Optional Presidio (ML NER for names, addresses, IBANs, 50+ entity types) runs async via Celery so the request path stays fast:

WIREGRAPH = {"ENABLED": True, "ENABLE_PRESIDIO": True}

See Installing Presidio. Presidio matches that overlap a regex hit on the same asset are deduped (regex wins).

Classification

Not every PII hit is a leak. An email to api.stripe.com is the product working; the same email to api.openai.com is an incident. Wiregraph classifies every DataEvent on what × where × policy and writes the verdict to DataEvent.outcome:

Outcome Meaning
expected Asset flowed to a sink that accepts it (Stripe ← email)
acceptable Trusted sink, asset not on its accept-list but not dangerous
suspicious Unknown sink, new flow, or known sink receiving an unexpected asset
prohibited Sensitive asset to a sink that should never receive it

A built-in catalog (~15 vendors: Stripe, OpenAI, Anthropic, Bedrock, Twilio, SendGrid, Segment, S3, Auth0, …) means a fresh install gets meaningful outcomes with zero setup. Override per-host via SINK_OVERRIDES (settings) or SinkCatalogOverride (DB, tenant-scoped). LLM strictness is tunable: LLM_POLICY = "strict" (default — medium+ PII to LLM is prohibited) or "relaxed".

Tenant resolution

By default Wiregraph walks request.user.tenant_memberships. Point TENANT_RESOLVER at your own callable if your project stores tenancy differently (FK, subdomain, header).

Retention purge

Delete events older than DATA_RETENTION_DAYS:

python manage.py wiregraph_purge [--dry-run]

Or via Celery Beat:

import wiregraph.celery as wg_celery
CELERY_BEAT_SCHEDULE = {**wg_celery.schedule(hour=3, minute=0)}

API

Requires pip install wiregraph[drf]. Without the extra, /api/v1/ routes are not registered.

Versioned under /api/v1/, JWT Bearer auth (obtain via /api/v1/auth/token/). OpenAPI at /api/v1/schema/, Swagger at /api/v1/schema/docs/.

Method Path Description
POST /api/v1/auth/token/ Obtain access + refresh token
POST /api/v1/auth/token/refresh/ Rotate refresh token
GET /api/v1/detection/events/ List events (filter: direction, data_asset, endpoint, timestamp__gte, timestamp__lte)
GET /api/v1/detection/events/{id}/ Retrieve event
GET /api/v1/detection/assets/ PII categories seen in traffic
GET /api/v1/detection/stats/summary/ Dashboard counts

See docs/settings.md for all config keys.

Requirements

Python ≥ 3.10 · Django ≥ 4.2 · Celery + Redis (async detection) · PostgreSQL recommended

License

MIT

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

wiregraph-0.1.0.post1.tar.gz (67.0 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

wiregraph-0.1.0.post1-py3-none-any.whl (106.4 kB view details)

Uploaded Python 3

File details

Details for the file wiregraph-0.1.0.post1.tar.gz.

File metadata

  • Download URL: wiregraph-0.1.0.post1.tar.gz
  • Upload date:
  • Size: 67.0 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.10.12

File hashes

Hashes for wiregraph-0.1.0.post1.tar.gz
Algorithm Hash digest
SHA256 f4fc5145d6ec42700b53a0a97b28bb53f139eebe78cfe9a7ba44f815c4aae243
MD5 17ff4d98e4c8d9bd6fe4b399dd7a8a24
BLAKE2b-256 53ec005b81bae9650f9e96f3de7d57ff069ecdf9778d29017e016437ce019d2f

See more details on using hashes here.

File details

Details for the file wiregraph-0.1.0.post1-py3-none-any.whl.

File metadata

File hashes

Hashes for wiregraph-0.1.0.post1-py3-none-any.whl
Algorithm Hash digest
SHA256 ec63a9dff1adc5b563aed8918c05f4dcc6545b236f6b9228bf2c702e82809541
MD5 cac7c6a695bf1f48efe26e65a52dcea2
BLAKE2b-256 cffcd5b7780a9638189e9e7c765cef7c22720b1d82668a8de1cb3b01476bcfcf

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page