Get started with AI Smart Redact
Deploy AI Smart Redact using Docker Compose, then upload and redact your first document through the web application. The full stack includes the Manager API, a Worker for entity detection, the Orchestrator for authentication and job management, the Human-in-the-Loop (HITL) web application for browser-based review, and two PostgreSQL databases.
Before you begin
- Docker and Docker Compose v2+ are installed.
- You have a valid AI Smart Redact license key. For evaluation, refer to Get a trial license key.
- Windows only: bash is available through WSL2 (recommended) or Git Bash.
- Your host can allocate roughly 8.5 GB RAM and 9.5 CPU cores across the service containers. For the per-service breakdown, refer to System requirements.
Get a trial license key
Follow these steps to obtain a free evaluation license key:
- Sign up or log in to the Pdftools Portal.
- On the Products page, next to AI Smart Redact, click Get started or See product.
- Next to the AI Smart Redact Trial, click Copy to copy the license key.
The free trial license key has the following limitations:
- The redacted files contain a watermark.
- The license key expires after 90 days.
The trial license key is valid for 90 days starting at UTC time. The Pdftools Portal indicates the validity period starting at 89 days (as the consumption of the first day already started). This period may slightly vary depending on your time zone relative to UTC.
For a full production license, contact sales.
Quick start
These steps bring up the default CPU stack. For GPU or API-only deployments, refer to Deployment variants.
-
Clone the samples repository:
git clone https://github.com/pdf-tools/smart-redact-samples.git && cd smart-redact-samples -
Set up your environment. Replace
<LICENSE_KEY>with the key from Get a trial license key:./smart-redact.sh setup --license-key "<LICENSE_KEY>"This creates
docker-compose/cpu/.envwith your license key and auto-generatedENCRYPTION_KEYandORCHESTRATOR_JWT_SECRET.infoAll three services (Manager, Worker, Orchestrator) validate the license key. The Worker fails to start without one. The Manager and Orchestrator start but reject API requests with HTTP 403 if the key is missing or invalid.
-
Start all services and wait for health checks to pass:
./smart-redact.sh up -
Verify that all services are healthy:
./smart-redact.sh health
Once the services are ready, the following endpoints are available:
- HITL web application: http://localhost:3000
- Manager API (Swagger): http://localhost:9982/swagger
- Orchestrator API (Swagger): http://localhost:9983/swagger
Redact your first document
The repository includes a sample PDF under samples/ that you can use to test the workflow.
- Open the HITL web application at http://localhost:3000.
- Log in with the default credentials (email:
admin@example.com, password:Admin@1234!Tmp). - Change the default password when prompted. Never use the seeded credentials in production. For the full first log-in flow and password requirements, refer to Authenticate with AI Smart Redact.
- Upload a PDF document containing sensitive data. You can use the included sample:
samples/Sample Document — AI Smart Redact.pdf. - Review the detected entities in the document viewer. Each entity is highlighted with its type label.
- Add, remove, or adjust detections as needed, then redact the document.
- Download the redacted PDF.
For details on the web application workflow, refer to HITL web application.
Deployment variants
AI Smart Redact ships in three Docker Compose variants. The CPU variant is the default and powers the Quick start. To switch variants, pass --variant <name> to both setup and up.
Each variant writes its own .env file under docker-compose/<variant>/, so run setup once per variant before up.
CPU (default)
Standard deployment with the HITL web application.
-
Set up the environment:
./smart-redact.sh setup --license-key "<LICENSE_KEY>" -
Start all services:
./smart-redact.sh up
GPU
Higher detection throughput with a CUDA-compatible GPU. Requires the NVIDIA Container Toolkit. For installation and tuning, refer to GPU deployment.
-
Set up the environment:
./smart-redact.sh setup --license-key "<LICENSE_KEY>" --variant gpu -
Start all services:
./smart-redact.sh up --variant gpu
Minimal (API only)
API integration without the HITL web application. For endpoint documentation, refer to the API reference.
-
Set up the environment:
./smart-redact.sh setup --license-key "<LICENSE_KEY>" --variant minimal -
Start all services:
./smart-redact.sh up --variant minimal
Stop the services
Stop and remove containers:
./smart-redact.sh down
Stop and remove containers including all data volumes:
./smart-redact.sh clean --volumes
To also remove the downloaded Docker images:
./smart-redact.sh clean --all
Next steps
- Configure AI Smart Redact for production settings (S3 storage, encryption, scaling).
- Scale AI Smart Redact with multiple Workers and GPU acceleration.
- Set up observability with OpenTelemetry and Grafana.
- Review the Architecture to understand how the subsystems connect.