Lattice gets your annotations and ideas out of the per-paper silo. See every highlight across your whole library in one place, filter and export them with citations, and promote the ones that matter into a searchable, taggable, cross-linked idea layer — all while staying linked to the source. Outline your paper and Lattice gathers every quote and note you filed under each heading into a first draft. It also includes grounded AI Q&A that answers from your PDFs with clickable page citations.
The name: a lattice is a network of connected nodes — which is exactly what your ideas become once they're no longer trapped in the single paper they came from.
📖 New here? Start with the Getting Started guide — a 10-minute walkthrough of every feature.
- See all highlights in one place — a single library-wide window listing every annotation, not one PDF at a time.
- Filter by colour, tag, keyword, and type — combined live.
- Export the filtered set with citations — copy as Markdown or HTML, or save as a standalone Zotero note. Each source is cited once (using your Zotero citation style) with its highlights beneath.
- Jump to source — click any annotation to open the PDF at that exact spot.
- Promote any annotation into a first-class idea — a real Zotero note that's searchable and taggable on its own.
- Tag ideas in place, link ideas to one another, and link back to the source paper via Zotero relations.
- Browse the layer in its own tab — your thinking is reachable independently of the paper it came from, which is what breaks down past a few hundred sources.
- Outline first, then fill it in — build the heading structure of your book, paper, or literature review, then file quotes and notes under each heading as you read.
- A heading is just a tag — each heading is a real Zotero tag (prefixed
§), so you can file a highlight under a heading straight from Zotero's own annotation tag box while reading, and it's gathered automatically — no separate mode to switch into. - Generate a first draft — Lattice gathers everything under every heading, in outline order, with citations, into a Markdown/HTML document or a standalone Zotero note. No copy-pasting, no endless text shuffling.
- Single-paper Q&A from a side panel in the reader, and multi-paper Q&A across selected items.
- Grounded, clickable citations —
[Page N]/[Paper N, Page M]jump straight to the cited page. - Save answers as annotations so AI-surfaced insights flow into the annotation browser and idea layer.
- Bring your own model — Anthropic (Claude), OpenAI (GPT), Ollama (local), DeepSeek, Grok (xAI). Keys stay in Zotero's own preferences; run fully offline with Ollama.
- Download the latest
lattice.xpifrom the Releases page. - In Zotero, open Tools → Plugins (or Add-ons).
- Click the gear icon ⚙ in the top-right → Install Plugin From File…
- Select the downloaded
.xpifile. - Restart Zotero.
Note
Requires Zotero 7+. For Q&A, each PDF's text must be indexed — open a PDF in the reader once, or right-click the item → Reindex Item. The annotation browser and idea layer work on your existing highlights and notes immediately.
- Tools → All Annotations.
- On the 📑 Annotations tab: filter by the colour swatches, tag dropdown, type, and the search box.
- Export with Copy as Markdown / Copy as HTML / Save as note, or click a row to open the source.
- Click ★ Promote to idea on a row (or ★ Promote all to ideas) to send annotations into the idea layer.
- In the same window, switch to the 🧠 Ideas tab (or Tools → Idea Layer (Citavi)).
- ➕ New idea, or work with promoted ones: add/remove tags inline, 📄 Open source, 🔗 Link to… another idea, 🗑 Delete.
- Ideas are ordinary Zotero notes tagged
★idea— so they're fully searchable and taggable in Zotero itself and they sync.
- Tools → Knowledge Organizer (or the 🗂 Organizer tab in the All Annotations window).
- Build your outline: ➕ Add heading, then ✎ rename, ▲ ▼ reorder among siblings, and ⭲ / ⭰ indent / outdent to nest sub-headings.
- File sources under a heading two ways:
- In the panel — select a heading, then under "File a quote or note here" search your annotations and click File here.
- While reading — in Zotero's reader, add the heading's tag (shown under the heading, e.g.
§Introduction) to any highlight. It appears under that heading automatically.
- Click 📄 Generate draft → preview it, toggle Markdown / HTML, optionally tick Include empty headings, then 📋 Copy or 📝 Save as note.
- Rename a heading and every quote filed under it follows the rename; delete a heading and your items keep their tags — nothing filed is lost.
- Single paper: open a PDF → Grounded Q&A section in the item pane → ask (Ctrl/⌘+Enter). Click
[Page N]to jump; 📌 Save to annotations to keep the answer. - Multiple papers: select 2+ items → right-click → "Q&A: Ask across selected papers." Citations are
[Paper N, Page M]; 📌 Save all to annotations anchors the answer on each cited paper.
Open Zotero → Settings (⌘,) → Lattice in the left panel.
- AI Provider — Anthropic, OpenAI, Ollama, DeepSeek, or Grok.
- API Key — paste your key (not required for Ollama).
- Base URL — shown for Ollama; defaults to
http://localhost:11434. - Model — choose from the provider's model list.
- Test connection — confirms provider, key, and model before you rely on them.
| Provider | Get a key | Key format |
|---|---|---|
| Anthropic | console.anthropic.com → API Keys | sk-ant-api03-… |
| OpenAI | platform.openai.com → API Keys | sk-… |
| DeepSeek | platform.deepseek.com → API Keys | sk-… |
| Grok (xAI) | console.x.ai → API Keys | xai-… |
| Ollama | No key needed — install Ollama and run a model locally | — |
| Provider | Endpoint | Models |
|---|---|---|
| Anthropic (Claude) | https://api.anthropic.com |
claude-haiku-4-5-20251001, claude-sonnet-4-6, claude-opus-4-8 |
| OpenAI (GPT) | https://api.openai.com |
gpt-5.3-instant, gpt-5.4-pro, gpt-5.4-thinking |
| Ollama (local) | http://localhost:11434 |
llama3.2, llama3.1, mistral, qwen2.5, gemma3 |
| DeepSeek | https://api.deepseek.com |
deepseek-r1 |
| Grok (xAI) | https://api.x.ai |
grok-4.3, grok-4.20 |
All cloud providers are called through their OpenAI-compatible /v1/chat/completions endpoint, except Anthropic, which uses its native /v1/messages API.
- Annotations are read library-wide from Zotero's data model (each highlight is an
annotationitem under its PDF), indexed in memory, and kept fresh via Zotero's Notifier. - Ideas are native standalone notes tagged
★idea; links use Zotero's own item-relation system, so nothing lives in a private database. - The Knowledge Organizer stores your outline as a single native note tagged
★outline(the heading tree as JSON), and every heading is an ordinary Zotero tag (prefixed§). So the outline and everything filed under it sync like any other item, and filing works from Zotero's own reader — again, no private database. - Q&A extracts per-page PDF text via
PDFWorker, builds a context with[Page N]markers, asks your chosen provider to cite every claim, and turns the citations into clickable links. Answers can be saved as page-anchored note annotations.
Built on the Zotero Plugin Template and zotero-plugin-scaffold.
npm install # install dependencies
npm run start # dev Zotero with the plugin loaded + hot reload
npm run build # production .xpi (output in .scaffold/build/)
npm run lint:fix # lint / formatsrc/
hooks.ts # Plugin lifecycle + menu registration
modules/
annotationIndex.ts # Library-wide annotation index + Notifier
annotationPanel.ts # All Annotations window (Annotations + Ideas + Organizer tabs)
annotationExport.ts # Markdown/HTML/note export with citations
annotationWriter.ts # Create page-anchored note annotations
ideaLayer.ts # Citavi idea objects (notes + relations)
ideaPanel.ts # Idea Layer browser UI
outlineModel.ts # Knowledge-organizer outline tree + heading tags (native note storage)
outlineExport.ts # Gather-by-heading → Markdown/HTML/note draft
outlinePanel.ts # Knowledge Organizer UI (outline tree + draft preview)
qaPanel.ts # Reader Q&A panel + multi-paper dialog
pdfExtractor.ts # Per-page PDF text extraction
llmClient.ts # Provider API calls + context building
preferenceScript.ts # Settings pane logic
"No text found in this PDF." Zotero hasn't indexed the PDF — open it in the reader once, or right-click → Reindex Item. Scanned PDFs without OCR have no extractable text.
"API key not set." Add your key in Settings → Lattice, then Test connection. (Not required for Ollama.)
Q&A citations aren't clickable. The model didn't emit the expected citation format; try a more capable model (Claude Sonnet/Opus, GPT-5.4 Pro), which follow the instructions more reliably.
New colours/tags don't appear in the filter row. Those rows are built when the window opens — close and reopen All Annotations to pick up newly added colours/tags.
Nothing shows under a heading in the Organizer. Filing uses the heading's exact tag, shown beneath the heading (e.g. §Introduction). If you tag highlights in the reader, match that tag exactly — or rename the heading from the panel, which re-tags everything filed under it for you.
Distributed under the AGPL-3.0-or-later license. See LICENSE.
No warranties are provided. You are responsible for any usage costs incurred with your chosen AI provider.