OCR Models

Otoroshi LLM Extension provides support for OCR (Optical Character Recognition) models, enabling text extraction from images and PDF documents through a unified, Mistral-inspired API.

OCR is exposed as a dedicated, first-class entity type — the OCR Model — just like Audio Models, Image Models, or Embedding Models. Each OCR Model wraps a provider and its configuration, and can be called through a dedicated plugin, the unified OpenAI-compatible API, or a workflow function.

Supported providers

Provider	Models	Input
AlphaEdge 🇫🇷 🇪🇺	alpha-digit-max, alpha-digit-medium	image, pdf
Mistral	mistral-ocr-latest, mistral-ocr-2505	image, pdf

Features

• Unified API: All providers are exposed through a single, consistent OCR endpoint regardless of the underlying provider
• Multiple input formats: Pass the document as a remote URL, a base64 data-uri, an inline base64 string, or a raw multipart file upload
• Two transports: JSON body (Mistral-style) or multipart/form-data file upload
• Vault integration: API tokens support Otoroshi vault references (e.g. ${vault://local/my-token})
• Model constraints: Restrict which models can be used via allow/block lists
• Workflow integration: OCR is available as a workflow function (ocr_call) for use in agentic pipelines

How to use OCR

There are three ways to call an OCR Model:

Method	Description
Dedicated plugin	The Cloud APIM - OCR backend plugin exposes a single POST /ocr route. See Plugins.
Unified API	The OpenAI Compatible API plugin exposes POST /ocr alongside chat, audio, image and embedding endpoints. See OpenAI Compatible API.
Workflow function	The ocr_call function calls an OCR Model from within a workflow.

OCR can also be performed through a regular text/LLM provider instead of an OCR Model entity — by sending a chat completion with an image or PDF content part. See OCR through text models.

OCR model entity

An OCR Model entity wraps a provider connection and its options:

{
  "id": "ocr-model_xxxxxxxxx",
  "name": "My OCR Model",
  "description": "OCR model backed by AlphaEdge",
  "provider": "alphaedge",
  "config": {
    "connection": {
      "base_url": "https://api-endpoints.alphaedge-ai.com",
      "token": "${vault://local/ALPHAEDGE_API_KEY}",
      "timeout": 180000
    },
    "options": {
      "model": "alpha-digit-max"
    }
  },
  "kind": "ai-gateway.extensions.cloud-apim.com/OcrModel"
}

See Providers for the per-provider configuration, and Plugins for the full API usage.