Your LLMs are hungry for data, but documents are messy 😩. DocStrange is the answer! Our open-source solution turns any document into clean, LLM-ready data with one command. Give your models what they need. 🔗 github.com/NanoNets/docstran… 🔗 pypi.org/project/docstrange/

75% of what you pay Claude for is your agent re-discovering things it already knew yesterday. Graft fixes it with an absurdly simple idea: the agent learns the codebase once, , instead of relearning it every time. it scans the repo, writes it out as linked markdown inside your git, and the agent reads that before touching anything. Here's how:↓

I've been testing Graft, and it fixes one of the biggest pain points with AI coding agents. Normally, every new task starts from scratch. The agent spends time grepping files, following imports, and rebuilding context before it can actually code. Graft changes that by creating a version-controlled knowledge graph as linked Markdown files inside your repo. The agent reads that context first and keeps it updated as the codebase evolves. In my testing, it noticeably reduced unnecessary tool calls and made coding tasks feel much faster, all without sacrificing output quality. No embeddings. No vector DB. No daemon. Just Markdown files that live in your repository. Works with Claude Code, Cursor, Codex, and Gemini. Definitely worth checking out if you're building with AI coding agents. #AI #Coding #OpenSource #DeveloperTools #ClaudeCode #Cursor #Codex #Gemini #LLM

Real-world instructions don't come one rule at a time. They tangle - budgets, exceptions, "never do X," and conditions that fire other conditions. SurgeAI created a benchmark for this - it's called Complex Constraints. We tested Nanonets on this ComplexConstraints benchmark. Result: #1, ahead of every frontier model. 🥇 45.0% of tasks fully solved (Gemini 3.1 Pro: 40.4%) ✅ 90.0% of individual constraints satisfied The trick isn't a longer prompt - it's a context graph.

What software has actually gotten better? OpenAI and Anthropic are racing toward $100B+ in ARR. The #1 use case, by far, is software. So here's the question nobody's asking: which software you actually use every day has gotten better? Your AP system, claims processing, compliance checks, back-office ops - the software that runs companies still breaks on the same thing: following lots of rules that depend on each other. Surge AI built a benchmark for this - ComplexConstraints. Real tasks where rules interact, override, and cascade, and one early mistake breaks everything downstream. The best frontier models solve under 41% of them. Nanonets ranks #1: 45.0% of tasks fully solved vs 40.4% for the strongest public model, and 90% of individual constraints satisfied. The difference isn't a bigger model - it's how we represent the rules. We parse every instruction into a context graph: each rule a node, every dependency an edge, so nothing gets dropped as the rules pile up. Frontier models made writing software easier. Context graphs are what make the software actually get better - the missing piece for enterprise automation: not more raw intelligence, but reliable rule-following at scale. Full results + how it works 👇 nanonets.com/research/comple…

Nanonets OCR-3 is the only OCR model you'll need in your agentic stack. The model API exposes five endpoints - /parse - structured markdown /extract - structured outputs in your schema /split - classify or split outputs based on content /chunk - context-aware chunks optimized for RAG /vqa - grounded answers with bboxes over sources We've specifically fine-tuned the model on edge cases where OCR repeatedly fails - complex tables, forms, non-trivial layouts.

With bounding boxes, you get exact coordinates for every extracted element. Use them for - 1. RAG citations 2. Feeding specific document regions to agents 3. Agent observability With confidence scores, you can measure reliability of every extraction. Pass high-confidence outputs directly, route low-confidence outputs to human review or a larger model. Use them to push your net accuracy to near 100%.

#changelog Introducing AI Agent Guidelines. 👉 changelog.nanonets.com/intro…

Introducing Nanonets-OCR2: a lightweight 3B VLM that transforms documents into clean, structured Markdown and is capable of VQA. We have trained the model on close to 3 million documents. It is multi-lingual and can handle handwritten documents.

@nanonets just shipped Nanonets-OCR2: new 3B VLM for OCR! LaTeX equations, tables, handwriting, charts, multilingual - it does it all! You can try it against your data with one command via @huggingface Jobs - no local GPU needed! The HF Jobs command/output from the model 👇

We're excited to share Nanonets-OCR-s, a powerful and lightweight (3B) VLM model that converts documents into clean, structured Markdown. This model is trained to understand document structure and content context, like tables, equations, images, plots, watermarks, checkboxes, etc

Just launched: The smartest AI-powered Resume Builder — 100% free, no strings attached! Try it now: resume.nanonets.com #jobsearch #resume #AItools #resumetips #careerdevelopment #nanonets

#changelog Flexible Model Training Options. Our “Improve Model” feature has evolved to provide greater flexibility and precision in selecting training files. 👉 changelog.nanonets.com/flexi…