Your pattern PDF,
transformed into a tracker.

Your pattern. Your PDF.
Your place, never lost.

InterTwined isn't a PDF reader with a counter bolted on. It understands what your pattern says — sections, shaping, stitch counts — and keeps your place through every row, every session, every frog-and-restart.

We don't guess. We verify.

Before the AI ever sees your pattern, a deterministic parser has already extracted every verifiable fact — cast-on count, gauge, stitch checkpoints, shaping math, section headers. These become anchors. The AI structures the rows around facts it can't argue with.

Ready to stop losing your place?

3 free extractions. 50 free AI questions. 125+ curated patterns. No subscription.

Plus mittens, cowls, bags, and more added regularly.

Every pattern credits its designer

Our library only includes freely available patterns, always with full attribution and a link to the designer's original source. We're here to help you follow patterns, not take credit for them.

Don't see your pattern? Import any PDF — DROPS, Yarnspirations, Ravelry, Etsy designers. If it's a knitting or crochet pattern, InterTwined can read it.

We're building this for crafters, not shareholders

Every dollar from credit packs goes back into expanding the curated pattern library — more patterns, better extraction quality, and a better app for you. We want to build the best crafting companion we can afford.

No venture funding. No ads. No subscription. Just a small team that knits.

Exactly what we send — nothing more

This is the complete payload for a single extraction. What you see below is what we receive. There is no hidden data, no secondary collection, no device fingerprinting.

{
  "timestamp": "2026-04-15T22:30:00Z",
  "app_version": "1.0.0",
  "name": "Slip Knit Socks",
  "score": 94,
  "dimensions": {
    "stitch_math": 23, "section_integrity": 15,
    "size_consistency": 12, "chart_honesty": 9,
    "instruction_quality": 15, "shape_data": 10,
    "metadata": 10
  },
  "cost": {
    "input_tokens": 4241, "output_tokens": 14535,
    "cached_tokens": 0, "duration_seconds": 38.2,
    "text_length": 9266
  },
  "context": {
    "re_extraction": false, "size_requested": true,
    "gauge_override": false, "curated": false,
    "publisher_hint": "DROPS"
  },
  "craft": "knitting",
  "category": "sweater",
  "pieces": 4,
  "sections": 12,
  "total_rows": 198,
  "parser": {
    "cast_on_found": true, "gauge_found": true,
    "multi_size": true, "pieces_detected": 4,
    "checkpoints": 27, "shaping_blocks": 9,
    "charts_referenced": 2, "atst_detected": true
  },
  "errors": ["chart_honesty_warning"]
}

We include the pattern name so we can find and purchase patterns that struggle — that's how we improve the engine. Notice what's not here: no PDF content, no row instructions, no stitch counts from your pattern, no designer, no user ID, no device ID, no IP address. We can't identify you from this data. We can only see: "Slip Knit Socks scored 94/100, took 38 seconds, and the chart dimension was slightly low" — then go buy the pattern and fix it.

The academic attempts

In 2019, MIT's CSAIL published "Neural Inverse Knitting" — a system that could look at a photograph of knitted fabric and reverse-engineer the machine knitting instructions. Impressive, but it solved image-to-machine-code, not human-pattern-to-structured-data.

Other efforts formalized knitting as a topological problem — modeling yarn as curves in 3D space, building formal grammars for notation. Beautiful mathematics. None of it helps a knitter follow a DROPS pattern on a Tuesday night.

Why the apps gave up

KnitCompanion, Row Counter, StitchFiddle — every existing app requires manual entry. They're PDF readers with counters bolted on. One startup, Knitrino, attempted interactive patterns in 2020. Their founders said it plainly: "Each pattern has to be manually built by us." The problem beat them.

The seven hard problems

What changed

Large language models. Not as a magic wand — a raw language model makes mistakes on stitch math. What works is a hybrid approach: a deterministic parser extracts every verifiable fact from the PDF (cast-on counts, gauge, checkpoints, shaping math), then feeds those as anchors to the most capable language models available. The parser keeps the model honest. The model handles the ambiguity the parser can't.

We validate every extraction across seven quality dimensions and score it out of 100. Today, 98% of patterns in our library score A or A+. The hardest patterns — DROPS sweaters with 8 charts and simultaneous shaping — went from failing grades to 97/100.

Three ways to track — each one a safety net

Every extraction gives you all three modes. Start where you're comfortable. Step up when you're confident. Fall back when you'd rather just read the PDF. You're always covered.

Advanced

The full picture

Row-by-row extracted instructions with computed stitch counts, shaping events highlighted, and a live garment schematic that tracks your position.

Works great on most patterns. The most detailed knitting tracker available — when the extraction nails it, nothing else comes close.

Simple — Default

The reliable middle

Your original pattern text, organized by section with completion checkmarks and row counts. The designer's words — we just keep your place in them.

Battle-tested on every pattern we've thrown at it. Most crafters will live here. Your PDF is always one glance away.

Basic

The safety net

Your PDF on top, a counter and garment schematic below. The extraction organized your pattern into sections with the right row counts — tap Next Row and read along.

If anything's off, edit the row count with one tap. The simplest possible tracker. It always works.

All three modes use the same extraction and the same credit. Switch between them anytime — your progress carries over.

We think we're close. We need your help.

No one has shipped automated pattern extraction at this level before. We're confident in the engine, but every pattern is different — every designer has their own style, notation, and quirks. The more patterns our system sees, the better it gets.

That's where you come in. Import your patterns. Tell us when something doesn't look right. Every piece of feedback makes the extraction better for everyone.

You can also opt in to share anonymous quality metrics. Three things get sent, nothing else:

• Extraction metrics — scores, pattern name, category, parser hits, cost. So we know what's working and what isn't.
• Row flags — when you swipe a row and flag it as wrong, we get the section, row number, and that row's instruction text (first 200 characters). This is the only pattern text we ever see, and only for rows you specifically flag.
• Completion ratings — thumbs up or down when you finish a pattern, plus how many rows you flagged along the way.

No PDFs. No full pattern text. No personal data. We include the pattern name so we can go buy patterns that struggle and fix the engine — that's how this gets better. As a thank you, every credit pack includes a bonus extraction when sharing is on.

You're not just using an app — you're helping solve a problem that researchers and developers have been working on for decades.