# Show HN: PDF Insight – local-first AI that sorts your PDFs on-device > Source: > Published: 2026-06-30 00:48:08+00:00 Drop a folder of receipts, statements and slips into PDF Insight. It reads each page, sorts and orders everything the way you ask, then hands back a single merged PDF. It all runs on your own computer. Two kinds of people use PDF Insight. Pick the door that sounds like you. Tax preparers, bookkeepers and firms handling client files Self-employed, and anyone with a pile of PDFs to organize Point it at a folder. Write your sorting rules once. PDF Insight reads and classifies every document, even scanned pages, orders them your way, lets you review, and exports one clean merged file. Nothing leaves your computer. Curious how it works? It's all under the hood, just below. 🧭 On the roadmap: a guided assistant that asks the right questions as you go, so nothing's missed in a client's return. Not in the app yet — the tool sorts and merges today. The plain version is above. Here's the technical detail, for anyone who wants it. PDF Insight runs an open local language model on your own machine through Ollama, which is free to install. The model reads and classifies each document locally; no account, API key or internet connection is needed for the local tier. Scanned and image-based pages are read with on-device OCR using Tesseract, in both English and French, so scanned slips are sorted and ordered just like native digital PDFs. The OCR runs entirely on your computer. In the local tier, nothing is uploaded. No server, no cloud copy of your files. It works with no internet connection, so it keeps running on an air-gapped machine during tax season. If you want near-instant processing, there is an optional, clearly-labelled paid cloud speed lane powered by Cerebras. It is off by default and billed separately; only when you turn it on do documents leave your machine. TaxDome, SmartVault, Canopy and Dext lead with compliance badges precisely because client files sit in their cloud. Pasting documents into ChatGPT ships client SINs to a third party. PDF Insight reads, sorts and merges entirely on your machine — so there's no upload, no server, and no third-party copy to breach. It removes the risk instead of insuring it. Client files uploaded to someone else's servers. You're trusting their breach response. The AI runs on your machine. Files are read locally and never transmitted. Nothing to breach. Prices in CAD. 14-day free trial, no card required. Plans named for who you are, not for upsells. For one person organizing their own PDFs For you — a solo preparer or personal use Pay once, keep it forever — for early adopters No subscription. Best if you plan to keep using it. For your team — multiple preparers No credit card. Runs entirely on your computer — nothing to upload. macOS, Windows, and Linux. macOS — signed & notarized by Apple: just open it, no warnings. Windows — unsigned for now, so on first launch click “More info” → “Run anyway” (one time). Needs Ollama (free) for the local AI — setup link in the app. Tap a question to expand. Straight answers for accountants and bookkeepers evaluating a private, local document tool. Yes. You don't set up anything. Drop your receipts and statements in a folder, type how you want them ordered, and get one clean PDF to send your accountant. Nothing is uploaded. No. In the default local tier, PDF Insight runs entirely on your own computer. Your clients' tax PDFs are read, classified and merged on-device and are never uploaded to any server. An optional, clearly-labelled paid cloud speed lane (powered by Cerebras) can be turned on for near-instant processing; only then do documents leave the machine. The local tier is the default — nothing leaves your machine when you use it. Yes. The local tier works fully offline. The AI model runs on-device through Ollama and the OCR runs on-device through Tesseract, so PDF Insight can organize and merge a client's documents with no internet connection. The licence check is offline-tolerant with a grace window, so it won't break on an air-gapped tax-season machine. PDF Insight is built for Quebec and Canadian tax slips, including T4, T4A, T5, RL-1, RL-3, RL-31, RRSP/REER contribution receipts and FHSA documents. Because it uses a local LLM to read and classify documents, you can also write your own rules to handle any other document in a client's pile. Point PDF Insight at the client's folder, write your sorting rules once, and the local AI classifies and orders every slip the way you specified. You review the result and export a single merged, correctly-ordered PDF for that client. A real 11-document bundle is organized in about 100 seconds on a 16GB Mac. Yes. PDF Insight reads scanned and image-based pages with on-device OCR using Tesseract, so scanned slips are classified and ordered just like native digital PDFs. The OCR runs locally and scanned pages are never sent to the cloud in the local tier. Yes. PDF Insight is a desktop application that runs on both macOS and Windows. On a 16GB Mac it organizes a full client bundle in about 100 seconds entirely on-device. TaxDome, SmartVault and Canopy are cloud document vaults that store your clients' files on their servers and charge per seat. PDF Insight is a local organizer that runs on your own machine and pre-sorts and merges the pile before it ever reaches a vault. Because nothing is uploaded in the local tier, there is no third-party copy of client data to breach. It also speaks Quebec slip vocabulary (T4, RL-1, RL-31) in French and English, which the US cloud tools do not. A real 11-document client bundle is organized in about 100 seconds fully locally on a 16GB Mac. If you enable the optional paid cloud speed lane (Cerebras), the same work completes in roughly one second of compute.