Gemini can now search your PDFs for charts and diagrams
Google's Gemini API File Search now handles images inside documents. RAG just got less annoying for technical docs.
Google announced multimodal file search for the Gemini API last week. The interesting bit is not that it searches text. Every RAG stack does that. The interesting bit is that it searches images embedded in PDFs and DOCX files. Most RAG pipelines strip images. You chunk the text, embed it, retrieve it. If the answer is in a flowchart or an architecture diagram, you miss it. This matters for technical documentation. A lot of enterprise knowledge lives in diagrams that no one bothered to transcribe into alt text. Gemini’s file search now indexes those images as part of the same retrieval step. You upload a PDF with charts. The model can answer questions about what is in the chart. You do not pre-process the images. You do not write separate parsers. It is part of the same API call. The catch is context window. Gemini 1.5 Pro has a two million token window, which is why this works at all. You can dump entire slide decks and still have room for the conversation. Smaller models would choke. I am curious how well it handles low-quality scans. Most corporate PDFs are not born-digital. They are scanned printouts from 2003, half-rotated, with coffee stains. If it can extract meaning from those, it is actually useful. If it needs clean vector graphics, it is a demo feature. The pricing is per-token, same as text. They do not break out what an image costs versus a paragraph. That makes cost prediction harder if you are building something that processes random user uploads. This is a pragmatic feature. Not flashy. But if you have ever tried to build RAG over technical manuals, you know the pain of stripping useful context because your chunking strategy cannot handle figures. Now you might not have to.