🚀 v1.0 — Now with Deep NER

AI That Reads
Between the Lines

Osmomark automatically highlights what matters on any page — organizations, people, places, technical terms, and smart keywords. All runs 100% in your browser. Zero API calls. Zero data leaves your machine.

✨ Install Free

🔒 Local AI ⚡ 100% Offline 🔋 Zero API Calls 💰 Free Tier Available

OpenAI recently announced that Sam Altman will lead the development of GPT-5, a model capable of multimodal reasoning across text, images, and code. The announcement came during a keynote at San Francisco's Moscone Center, where thousands of developers gathered to discuss large language models and their impact on software engineering. Google and Microsoft have both announced competing AI infrastructure investments.

Entity Organization Person Keyword Location

Features

Highlight Engine That Doesn't Phone Home

Three tiers of intelligence, from instant regex to full transformer NER. All running locally in your browser via WebAssembly.

⚡

Tier 1 — Instant Regex

URLs, emails, dates, numbers, money, and taxonomy dictionaries (programming languages, crypto, frameworks). ~60% coverage in under 5ms.

🧠

Tier 2 — NLP via Compromise.js

Fast POS tagging for organizations, people, places. TF-IDF keyword extraction with n-gram phrases. Verb-boosted scoring. ~85% coverage.

🤖

Tier 3 — Transformers NER

Full distilbert-base-NER pipeline via transformers.js. Identifies entities with ~95% accuracy. ~18MB model cached in IndexedDB. Opt-in, 5 free scans/day.

🔒

100% Private

Everything runs in the page's main world — not even a CDN fetch for the model runtime. Your text never touches a server. Period.

📄

Read Any File

Upload DOCX, PDF, TXT, MD, or PPTX directly in the popup. Osmomark extracts and highlights text from documents alongside web pages.

🎨

Customizable

Per-category colors, highlight styles (underline, background, border), symbol packs, and custom taxonomies. Pro unlock: full palette + exports.

🏗️ Architectured for MV3

Chrome's Manifest V3 blocks eval() in content scripts — which breaks most transformer-based extensions. Osmomark works around this by injecting the model runtime into the page's main world via a <script> tag, then bridging back with postMessage. No eval, no CDN, no compromises.

Script injection via web_accessible_resources
PostMessage bridge between isolated + main world
Model cached in IndexedDB — works offline
No service worker overhead for inference

🧩

📊 Smart Keyword Extraction

Tier 2b uses TF-IDF scoring with domain-aware boosting. Prioritizes verbs (3.0× boost), extracts bigrams and trigrams (all words must be content), handles compounds ("fine-tuning", "domain-specific"), and filters ~200 NEVER_KEYWORDS including transitional adverbs, contractions, and temporal words.

15 keywords on Pro (8 on Free)
All occurrences highlighted per term (no cap)
n-gram phrases > single words for context
Apostrophe-preserving regex matching

🔑

How It Works

Install. Read. That's It.

No accounts. No sign-ups. No data leaving your machine. 100% local AI in your browser.

Install

Add Osmomark to Chrome from the Web Store. It's free — no permissions beyond what it needs.

Browse

Open any page — article, docs, blog, PDF. Osmomark automatically scans and highlights what matters.

Absorb

Named entities in color-coded highlights. Smart keywords for the gist. Deep NER for precision. Knowledge, osmotically.

AI That Reads Between the Lines