eagle-demi

DEMI (Document Extraction & Machine Intelligence) for EPIC. Deploys docling-serve as a cluster-internal PDF/DOCX extraction endpoint. Called by eagle-api on document upload; no custom application code — pure Helm deployment of the official upstream image.

Quick Start

Prerequisites: Create the API key secret in each namespace before deploying:

oc create secret generic eagle-demi-api-key \
  --from-literal=DOCLING_SERVE_API_KEY=$(openssl rand -hex 32) \
  -n 6cdc9e-dev

Deploy to dev:

helm upgrade --install eagle-demi ./helm \
  --namespace 6cdc9e-dev \
  --values ./helm/values-dev.yaml \
  --wait --timeout=10m

Deploy to test/prod: Use the GitHub Actions workflow_dispatch workflows in .github/workflows/.

Architecture

• Image: ghcr.io/docling-project/docling-serve-cpu:v1.21.0
• Port: 5001 (ClusterIP only — not exposed externally)
• Auth: X-Api-Key header from eagle-demi-api-key OpenShift Secret
• NetworkPolicy restricts ingress to eagle-api pods (role: api-eagle-epic) only

eagle-api calls http://eagle-demi:5001/v1/convert/file on document upload, stores extracted text as DocumentChunk records in MongoDB, and eagle-typesense syncs those to the document_chunks Typesense collection.

Configuration

All tunable limits live in helm/values.yaml — override per environment in helm/values-{env}.yaml.

Parameter	Default	Description
config.maxFileSize	104857600	Max upload size (bytes; 100 MB)
config.maxDocumentTimeout	300	Per-document timeout (seconds)
config.maxNumPages	500	Max pages per document
config.numThreads	4	Torch CPU threads
config.numWorkers	2	Engine worker processes

Related Repositories

• eagle-api — Calls eagle-demi on upload
• eagle-typesense — Syncs DocumentChunks to Typesense
• eagle-dev-guides.wiki — Architecture docs