Skip to content

Latest commit

 

History

History
54 lines (35 loc) · 2.51 KB

AZURE_SETUP.md

File metadata and controls

54 lines (35 loc) · 2.51 KB

Azure Setup Guide

docproc CLI can use Azure for document extraction:

  1. Azure OpenAI — chat (for LLM refinement) and optional vision chat for PDF embedded images.
  2. Azure AI Vision (Computer Vision) — image extraction: Describe API (captions/tags) + Read API (OCR for text in images, e.g. equations and labels).

Environment variables

Set in .env or your environment (then use docproc init-config --env .env to generate ~/.config/docproc/docproc.yml, or set them when running the CLI):

Variable Description
AZURE_OPENAI_API_KEY API key (same key often works for OpenAI + Vision in one resource)
AZURE_OPENAI_ENDPOINT Azure OpenAI endpoint, e.g. https://<resource>.openai.azure.com/
AZURE_OPENAI_DEPLOYMENT Chat model deployment name (e.g. gpt-4o)
AZURE_OPENAI_EMBEDDING_DEPLOYMENT Optional; used only if you had RAG (not used by CLI).
AZURE_VISION_ENDPOINT Computer Vision endpoint, e.g. https://<resource>.cognitiveservices.azure.com/

If Vision is in the same Cognitive Services resource as OpenAI, use the same key for both.

Credentials via Azure CLI

Use the script to fetch endpoint and key from an existing Cognitive Services resource:

./scripts/azure_env.sh [resource-name] [resource-group]
# Append output to .env: ./scripts/azure_env.sh myresource myrg >> .env

Requires Azure CLI and az login.

DeploymentNotFound (404)

If you see DeploymentNotFound: The API deployment for this resource does not exist:

  • Azure OpenAI requires deployments for each model. Create them in the Azure OpenAI portal.
  • Chat / vision: Ensure AZURE_OPENAI_DEPLOYMENT matches the deployment name (e.g. gpt-4o, gpt-4o-mini).

Using Azure AI Vision for images

If you have Azure AI Vision (Computer Vision) deployed:

  1. Set AZURE_VISION_ENDPOINT to your Computer Vision endpoint (e.g. https://<resource>.cognitiveservices.azure.com/).
  2. Use the same key as AZURE_OPENAI_API_KEY when both are in the same resource.
  3. Image extraction will use:
    • Describe API — image captions and tags
    • Read API (v3.2) — OCR for text inside images (equations, labels)

See CONFIGURATION.md for ingest.use_vision and related options.

Demo (Go app)

The demo uses OPENAI_API_KEY for RAG and grading. To use Azure from the demo you would need to extend the Go app to support an Azure provider; out of the box it uses OpenAI.