philterd/arbiter repository overview

⁠Arbiter

Arbiter is a human-in-the-loop deidentification platform. It runs documents through the Philter / Phileas⁠ PII detector and gives reviewers a focused workflow for confirming or correcting each detection before a document is treated as redacted.

Detection is probabilistic, so the human review step is the point: Arbiter routes low-risk documents past a reviewer automatically and puts everything else in front of a person. You are responsible for validating output against your own data.

⁠Quick start

Arbiter needs MongoDB for storage and a Valkey (or Redis protocol compatible) server for sessions. OpenSearch is additionally required for full-text search.

docker run -d --name arbiter -p 8080:8080 \
  -e SPRING_DATA_MONGODB_URI=mongodb://mongodb:27017/arbiter \
  -e SPRING_DATA_REDIS_HOST=valkey \
  -e SPRING_SESSION_STORE_TYPE=redis \
  -e ARBITER_CRYPTO_SECRET="$(openssl rand -base64 32)" \
  -e ARBITER_ADMIN_INITIAL_PASSWORD='ChangeMeOnFirstLogin!' \
  philterd/arbiter:latest

Generate and store ARBITER_CRYPTO_SECRET once, then keep it: it is the AES-256 key protecting stored API keys and data-source credentials, and rotating it makes existing encrypted values unreadable.

⁠First run

On any start where no admin account exists, Arbiter creates [email protected]. If ARBITER_ADMIN_INITIAL_PASSWORD is set to at least 12 characters, that password is used; otherwise a random one is generated and printed to standard output in the startup banner. Change it after signing in.

⁠Documentation and source

• Documentation: https://philterd.github.io/arbiter⁠ (also served by the container itself at /docs)
• Source: https://github.com/philterd/arbiter⁠
• More Philterd open source: https://www.philterd.ai/open-source-software/⁠

⁠License

Apache License 2.0.