NDI 1.0 NDI 1.0 is a compact document-processing model that ranks first on Parse Index Normalized across six public benchmarks and matches GPT-6 Astra quality at one-tenth the cost. Trained on more than 15 million financial files, it parses and classifies documents with industry-grade accuracy and ships with easy integrations for coding agents.
Visit productFeatured media
Loading media…
View on XCosigns Received
“I’M DONE GATEKEEPING THIS. i built a full NDI 1.0 trading system and open-sourced the entire repo. the agent can pull in market data and financial research: 10-Ks, earnings reports, SEC filings, financial statements, research PDFs and more. NDI extracts the information, grounds it back to the original source, then evaluates the decision while a frontier model reasons over the context. so it’s not just looking at price charts. market data + financial documents + AI reasoning → trading decision. repo + NDI API link below”
“una IA acaba de revisar 3.319 pagos de una empresa el modelo de @NaceAI comprueba que cada pago coincide con lo que la empresa tiene apuntado. el 99,4% lo resuelve solo y te avisa de lo raro, como pagos repetidos. y sin subir tus datos a ningún sitio.”
“The cheap model did the heavy lifting. The big one only got the hard pages this is the most boring job in any finance team: -> open the page (scans included) -> work out what kind of document it is -> file it in the right place @NaceAI Document Intelligence read all 145 pages Drex sorted them into 5 categories and decided 94.5% alone, about 444 ms each Astra 6 only got the 5.5% Drex was unsure about Drex alone: 95.9% with Astra 6 on the unsure pages: 97.9% Drex cost for all 145 pages: $0.0116 the cheap model decides most of it and the big one only sees what's hard. that's how this runs at scale (synthetic finance docs)”
“We gave AI a brain. Now we're giving it eyes. I explored a workflow combining Nace, Drex, and Opus 5.5. Nace handles scanned documents. Drex handles decisions. Opus 5.5 handles deeper reasoning. Instead of forcing one giant model to do everything, each layer focuses on a specific task. This is what the next generation of document-processing agents could look like. Works right out of the box with Claude Code, Codex, Pi, OpenCode, Hermes, and Prime Agent through its CLI and SDK”
“A full day of SEC filings. Read, scored and alerted before your coffee gets cold this is the most boring job on filing day: -> read every 8-K and exhibit -> work out which ones matter -> write the alert @NaceAI Document Intelligence read all 334 documents in 116s Drex scored all 219 filings in 21s. Drex cost: $0.035 Opus 5.5 wrote 18 alerts, and Drex checked every line what it caught on the day: -> C.H. Robinson x RXO merger -> Leslie's Chapter 11 -> ESS Tech NYSE delisting -> Aterian auditor change the flagged alert is the one I'd watch. the writer said something the filing didn't, and Drex caught it CLI + SDK, works with Claude Code, Codex, PI, Opencode, Hermes and Prime Agent”
“I threw a messy PDF at Nace's new Document Intelligence model. And this is where things get interesting. Instead of just converting the PDF into plain text, NDI is built to understand the document's structure, extract useful information, and connect every extracted value back to its original source. It supports 36 file types, including PDFs, spreadsheets, scanned documents, emails, and more. It's a perception layer built specifically for AI agents. Pretty interesting direction for enterprise AI.”
“INSURANCE PAYMENTS ARE A MONEY PROBLEM, NOT A PAPER PROBLEM Every payer sends remittances in a different format. Someone has to read each one, match the claims and check the totals before a single dollar gets posted. This is @NaceAI's new Nace Document Intelligence doing that job: → 8 payers, 6 different layouts, some of them faxed scans → 29 claims pulled into one clean schema → 176/180 fields correct → 7 of 8 payments balanced and auto-posted → the 1 that didn't add up went to the exception queue instead of the books That last part is the whole point. It didn't guess. The balance check caught a misread line and stopped it. Every extracted value points back to the exact spot in the source doc, so anyone can verify it in seconds. Tied for #1 on the parsing leaderboard, 36 file types, runs managed, in your cloud or fully air-gapped. Top-tier accuracy at a fraction of the cost. For a back office processing thousands of these a month, that's pure margin.”