Graceful failure when ontology tables are nonexistent

Merge pull request #241 from epflgraph/ontology-graceful-fail

Ontology graceful fail

Use generic filters for rag/retrieve

v0.17.1

bump elasticsearch-interface version to 0.11.1

Full PDF OCR support

• Full PDF support
• Gemini support for OCR

/rag/retrieve access management

Adds access management to /rag/retrieve, making it impossible for just anyone to access just any index. Two new tables are added to graphai_auth: User_Retrieve_Access and Retrieve_Index_Aliases, making it possible to give a user access to all or some of the existing indexes, including through their aliases (if any).

No-caching option

Adds no_cache flag for translation, embedding, and OCR (plus image upload/download) endpoints, which disables caching and is useful for sensitive data.

Secure image upload

Image uploads now require an additional privilege (upload) which should only be granted to admin accounts. Uploads now have a maximum size and will fail if that size is exceeded.

RAG and image upload

• Adds RAG endpoints (/rag/retrieve and /rag/chunk)
• Adds image upload endpoint (/image/upload_file)

OpenAI-based OCR

• OpenAI OCR with the extraction of math as LaTeX
• LaTeX verification
• PDF OCR will now by default return the results as a JSON containing page by page results

Fix image origin_token

Merge pull request #217 from epflgraph/image-origin-token-debug

Image origin token debug

Direct image retrieval and OCR

Enables the direct retrieval of image and pdf files, and allows for OCR to be performed on all of them.