Skip to content

Sifter v1.0.1

Latest
Latest

Choose a tag to compare

View all tags
@bfortunato bfortunato released this 18 Jun 08:44
· 7 commits to main since this release
v1.0.1
162347f

Sifter v1.0.1

What's new

markitdown OCR via LiteLLM

When SIFTER_PREPROCESSOR=markitdown and SIFTER_MARKITDOWN_OCR=true, Sifter now wires the markitdown-ocr plugin to a configurable vision model via LiteLLM. Images embedded in PDFs, DOCX, and PPTX files are OCR'd at extraction time using any vision model you specify.

New config fields:

  • SIFTER_OCR_MODEL — vision model for markitdown image OCR (falls back to SIFTER_EXTRACTOR_MODEL when unset)
  • SIFTER_OCR_API_KEY — API key for the OCR model (falls back to SIFTER_DEFAULT_API_KEY)
  • SIFTER_OCR_BASE_URL — base URL for the OCR model provider

Standalone image bypass

PNG, JPG, WEBP, and TIFF files uploaded directly (not embedded in a document) now bypass markitdown and are sent as base64 image_url blocks directly to the vision model, preserving full visual fidelity.

Dynamic supported extensions

GET /api/config now returns supportedExtensions based on the active preprocessor. In markitdown mode this includes .xlsx, .pptx, .mp3, .epub, and more. The frontend file picker updates automatically.

Packages

Package Install
Python SDK pip install sifter-ai==1.0.1
Server pip install sifter-server==1.0.1
Docker ghcr.io/sifter-ai/sifter:1.0.1
Assets 2
Loading