atherdlp-server

Local AtherDLP server: the DLP policy engine, REST API, and dashboard, served on one port with no external services.

AtherDLP detects sensitive data leaving your application. The SDK intercepts outbound httpx requests, classifies the destination, and reports what was in the payload. This package is the other half: it evaluates those events against your policies and shows them in a dashboard.

Quick start

pipx install atherdlp-server
atherdlp up

That applies migrations to a SQLite database, seeds a few demo events so the dashboard is not empty on first run, and opens http://localhost:8473.

No Postgres, no Redis, no Docker, no Node toolchain — the compiled dashboard ships inside the package.

Then point your application at it:

pip install atherdlp

import atherdlp

atherdlp.configure(endpoint="http://127.0.0.1:8473/events/http")
atherdlp.install()

# your app's httpx calls happen here

atherdlp.uninstall()

configure() is required for events to reach the server. Without an endpoint, the SDK still intercepts and classifies, but delivers nowhere and the dashboard stays empty.

What you will see

Every intercepted request is classified by destination — calls to OpenAI, Anthropic, Gemini, Mistral and Cohere are tagged as LLM egress, Stripe as PCI — then matched against sensitive info types and evaluated against your policies to an allow / alert / block decision. Activity and alerts are correlated by trace id, so one leaked field can be followed from the call that sent it to the alert it raised.

atherdlp up

State persists between runs in --data-dir. Delete that directory for a clean slate.

This server has no authentication yet. It binds to loopback by default, which keeps it private to your machine. Do not expose it on a public interface: the read APIs return captured payload samples, which are by definition the most sensitive data the tool handles.

Optional extras

pip install "atherdlp-server[ocr]"       # image OCR — also needs the tesseract binary
pip install "atherdlp-server[postgres]"  # Postgres instead of SQLite

OCR degrades gracefully: without these, image extraction is skipped rather than failing the request.

Running from source

git clone <org>/AtherDLP-backend && git clone <org>/AtherDLP-frontend
cd AtherDLP-backend
python3 -m venv .venv && source .venv/bin/activate
pip install -e ".[test]"

python scripts/build_frontend.py   # builds the dashboard, needs npm
atherdlp up

Run the tests with pytest (they use an in-memory SQLite database).

To develop the dashboard with hot reload, run atherdlp up and npm run dev in AtherDLP-frontend — the Vite dev server proxies /api to port 8473.

Deployments beyond local mode

atherdlp up sets ATHERDLP_LOCAL_MODE=true, which runs scan tasks in-process instead of dispatching to Celery. That is correct for a single process and wrong for a multi-worker deployment, where inline scanning blocks the request that triggered it. For that, leave local mode off and run Celery workers against Redis with a Postgres ATHERDLP_DATABASE_URL.

License

Apache-2.0.