Running Offline Batches

Deploy our desktop server for totally isolated, infinite batch processing without internet access.

Why local inference?

When operating on heavily restricted financial archives or local PII documents, sending bytes over the wire is not an option. The solveOCR Desktop Server brings 100% of our cloud capabilities down to your metal.

Air-gapped and fully independent. Requires ~4GB RAM per worker.

Running the Image

Using the official Docker image allows you to automatically map hardware accelerators (like NVIDIA CUDA or Apple Silicon Accelerate).

docker run -p 8080:8080 \
  --gpus all \
  -e WORKERS=8 \
  solveocr/server:latest

Batch Execution

Point your scraper scripts to `localhost:8080`. The SDKs seamlessly support local base URLs.

client = solveocr.Client(
  api_key="local", 
  base_url="http://localhost:8080/v1"
)

# Throws 1,000 PDFs at the queue
job = client.batch_extract(glob("archives/*.pdf"))