Show HN: ParseHawk – 100% Local Document AI with API, CLI, and Web UI
ParseHawk v0.1.0 is an Apache-2.0 licensed, fully local document AI platform that extracts JSON from PDFs and images. It builds on NuExtract3, enforcing provided JSON schemas with constrained decoding, and supports Apple Silicon and Linux with NVIDIA GPUs via API, CLI, and Web UI.
Background
- ParseHawk is a new open-source tool (Apache 2.0 license) that extracts structured JSON data from documents like PDFs and images — all running locally on your own machine, no cloud API calls needed.
- It builds on NuExtract3 (by NuMind), a small AI model fine-tuned specifically for extracting information from text. ParseHawk adds "constrained decoding" — meaning the AI is forced to output only data that matches a JSON schema you define, making results predictable and machine-usable.
- "100% local" is a key selling point for privacy/security: your documents never leave your computer. It works on Apple Silicon Macs (with a bundled Metal GPU backend) and Linux/NVIDIA machines.
- This sits in a crowded space alongside tools like Unstructured, LlamaParse, and Marker. The differentiators here: fully local, schema-enforced output, and a triple interface (API, CLI, web UI).