Every mailroom needs a good hire. This one sorts documents before you finish your coffee. 🐾📑 Nutrient’s classify tooling sorts incoming documents by type so the right document reaches the right team, without anyone sorting by hand.

What is the Nutrient Decision Model, and why would you ever use it over a large language model? Easy answer: cost and speed

Most questions you ask about a document have a short answer. Is this an invoice or a contract? Is this extracted value actually on the page? Where does one invoice end and the next begin? Yet the usual approach is to send the whole file to a general-purpose LLM and wait while it writes a reply, one token at a time. So we built something different: the Nutrient document decision model. It reads a document, both the page image and the text, and returns a choice plus a calibrated confidence score. No generated paragraphs, no waiting. What that gets you: → About 200 ms per answer → Scores calibrated to how often the model is actually right → Visual understanding that text-only approaches miss It’s already live in Nutrient DWS, powering our classify and split endpoints and checking extraction results against the source document. Nutrient AI engineer, Austin Nguyen, explains why we built it, how we trained it, and where it falls short: bit.ly/4jpIZyF Try the demo on Hugging Face: bit.ly/4i8tEjE

Building on the Nutrient Data Extraction API is a breeze. Our customer came to us with a problem: they receive invoices from all across the globe. Different languages, different currencies, and each one checked by hand. Not anymore. In less than 30 minutes, we built them a custom app that identifies discrepancies, and notifies their accounts team when something needs their attention. All built inside of Claude using only natural language.

If you can't trust your OCR some of the time, can you trust it at all? OCR turns a scan or photo of a page into text your software can use. Clean, typed pages are a breeze. Real documents aren't. A rotated scan, small print, a long table, a handwritten form: that's where a number gets misread or lands in the wrong row. And the response still looks fine. Your agents can't tell. If you aren't looking closely, you won't either. But someone still has to sign off on it. So the question isn't whether OCR works on an easy page. It's whether you can tell when it got something wrong. We compared nine OCR APIs, including ours, and we didn't pull any punches. The results might surprise you. Nutrient links extracted content back to where it appears on the page and marks the values that need a second look, so you know what to check. Read more: bit.ly/4rQppOj

You don't hand an AI agent autonomy. It earns it. SpaceXAI Engineer Lauren Tan (@poteto), who ships thousands of code changes a month with AI agents, puts it simply: "Make loops autonomous only after it earns your trust." The agents have to show their work, and any step can stop the work when something looks off. The same is true when the agent works on your documents instead of your code. Before an AI agent fills a form or redacts a clause in a real contract, someone has to sign their name off. The trust comes down to three answers about your docs: who set it up, what permissions the agent has, and has anything changed? Nutrient AI Assistant 2.4.0 keeps those answers on file: • Every agent's setup is saved and versioned, so you know exactly which version ran on any given day. • Edits can wait for a person's approval, so someone can stop the line before anything changes. • Each session reaches only the documents and agents it was issued for, and narrow jobs can go to helpers that can only read. Start with relatively strict permissions. Loosen them as the agent proves itself, and keep records of every change. How it works: bit.ly/4rN1O0M

In AI-powered workflows, document infrastructure isn't just a supporting feature. It's fundamental. When AI prepares a tax return, accountants still need to check it. That means tracing every number back to its source, checking the calculation, and marking up what needs attention, all inside the platform instead of across a patchwork of tools. That's the approach Current took when building their AI-native tax platform on Nutrient. "This product wouldn't work without document editing," says @tuckerhaas, CTO at Current. Read the story 👇 bit.ly/3VYoY8u

AI answers are useful. But answers your software can actually act on exponentially more powerful. Meet DocSignals: Nutrient document extraction + Jev, TypeSafe’s new AI model. Ask a question about a document. Get a structured signal and a citation you can inspect. Then define what happens next: route the document, trigger an automation, or bring in a person to review. Probabilistic signals. Deterministic workflows. Get started for free.

A PDF conversion can return “success” and still produce a file that fails conformance validation. But completion rate only tells you that the process finished, not whether the outputs are compliant or accurate. Across 3,157 documents, Nutrient achieved SOTA results: • 97.2% PDF/A conformance • 96.5% PDF/UA conformance • OCR up to 9.3× faster than Apryse • Zero critical failures in the conformance runs Because an invalid document that produces no error doesn’t disappear. It just becomes someone else's problem. See the results and methodology: twp.ai/4huNDS

Agentic RAG is a clear upgrade over traditional retrieval for ambiguous, multistep questions that span multiple documents. We agree. But your agents are only as reliable as the context you give them. If extraction flattens a table, skips handwriting, or garbles a degraded scan, a smarter model will simply reason more confidently over bad data. Garbage in. Garbage out. Nutrient’s Data Extraction API turns PDFs, scans, images, and Office files into structured JSON or Markdown while preserving reading order, tables, layout, and source references. That gives your agents cleaner, traceable context to work with. Read why document extraction sets the ceiling for agentic RAG: twp.ai/4huN83