okraPDF Server

Durable, parser-agnostic PDF parsing you run yourself.

A durable workerd orchestrator drives each parse job; parsers are swappable containers. No okraPDF cloud — your documents stay on your box.

MIT · self-hosted · bring-your-own parser

---

Quick start

git clone https://github.com/okrapdf/server.git && cd server/apps/self-host
docker compose up --build
# → open http://127.0.0.1:8787  (web UI + API)

Full docs, the parser contract, durability details, and configuration live in apps/self-host/.

What's inside

• apps/self-host/orchestrator — a workerd Cloudflare Agent. One Durable Object per document owns the parse run with native Agents-SDK durability (checkpointed to embedded SQLite; resumes after a restart). Serves the web UI + the /v1/documents/:id/{upload,status,graph} API.
• apps/self-host/parsers/liteparse — the reference parser container (@llamaindex/liteparse, native, real 0–1 bounding boxes) behind a tiny uniform HTTP contract. Add your own (GPU OCR, a cloud VLM) the same way — see the parser contract in the docs.

The architecture is one lean docker compose: orchestrator + parser, one durable volume, no database. The page-resume durability guarantee is documented under "How durability works" in the self-host docs.

Status

Public beta, single-tenant. The HTTP API is unauthenticated, so it binds to 127.0.0.1 by default — set HOST_BIND=0.0.0.0 only behind your own auth/ingress.

License

MIT.