Taggard is Your Memory EngineThink less.
Remember everything.
Taggard is a memory engine for your notes and documents. Every note gets two kinds of tags — Taggard suggests both: one set styled for browsing, yours to accept or reject, and a hidden layer for deep, cross-referenced recall you'll never have to search for by hand.
Taggard sits underneath whatever you already use to capture notes.
No new place to write. Drop in what you already have, and Taggard builds the second layer on top — automatically.
Add a note
Drop a document (.txt, .md, .pdf, .docx) or paste text directly.
Review the tags
Taggard suggests visible tags and a hidden layer underneath. Accept what's right, reject what's not — every call sharpens what it suggests next time.
Watch it connect
Entities get resolved and the note is automatically checked against everything else you've written — not just for shared keywords.
Ask, or follow the graph
Get a plain-English answer with citations, or click through the graph to see exactly how your notes link together.
Illustrative — every note in your account gets checked against everything else the same way, and this is the same graph you can click through on the Graph page.
Two kinds of tags. One for you, one for the machine.
Visible tags are Taggard's suggestions, styled the way you'd tag something yourself — a name, a project, a date — and yours to accept or reject. Underneath, Taggard builds a second, hidden layer: latent frames, tone, methodology, confidence — signals no one would bother tagging by hand, and ordinary search can't see.
Q3 planning notes
Team agreed to delay the senior hire until Q4 and reallocate budget toward contractor support. Acme renewal still open — revisit after the board review.
Entities resolved: Acme Corp, Q4. Confidence: 86%. Connected to two earlier notes Taggard found on its own.
Everything below runs on its own.
Entities, resolved
People, organizations, places, and products — recognized and matched across every note, so 'Acme' and 'Acme Corp' are the same thing. Everything else on this list builds on top of this.
Cross-references
Taggard finds when one note builds on, contradicts, cites, or overlaps in topic with another — automatically, not a backlink you draw yourself like in Notion or Obsidian — and shows the exact passages that triggered it.
Ask, with citations
Ask a question in plain English. Get an answer built from across your notes — including ones connected by meaning, not just keywords — with sources cited.
Idea evolution
Pick a topic and see it as a timeline: which note built on which, where you changed your mind, and when.
Theme discovery
Notes that share a latent pattern — not just a tag — get clustered automatically, and Taggard can name what connects them.
Knowledge graph
Every note and connection, visualized. Click a dot, follow a line, see your thinking as a map instead of a list.
A real AI stack, not a keyword index with a chatbot bolted on.
Every extraction step — tagging, entity resolution, cross-reference classification, theme naming — is a dedicated model call with its own schema, not one prompt doing everything.
Language models
GPT-class models handle tagging, entity extraction, cross-reference classification, and theme naming as separate, schema-constrained calls — plus a free-form call for Ask's actual answers, since those are prose for you to read, not structured data.
Vector search
Every note, chunk, tag, and entity gets an embedding. Postgres + pgvector (HNSW-indexed) power semantic search and similarity-based canonicalization.
Entity & tag canonicalization
New tags and entities are matched against existing ones by embedding distance before creating a duplicate — close variants merge automatically, and near-misses that don't quite clear the bar are flagged in your Insights digest instead of silently piling up as look-alikes.
Next.js + Postgres
Server-rendered throughout, backed by hosted Postgres (Supabase) with Prisma. No separate vector database to keep in sync.
Enforced by the database, not just the app.
Passwords are hashed, never stored in plain text. Reads are enforced with Row-Level Security at the database itself — not just trusted to application code — so Postgres refuses to return another account's rows even if the app had a bug. Export your full knowledge base as JSON or CSV whenever you want.
See the full security model →Questions worth answering up front
How is a hidden tag different from a regular tag?+
Visible tags are Taggard's suggestions, styled the way you'd tag something yourself — a name, a project, a date — and yours to accept or reject. Hidden tags are a second layer Taggard builds automatically: latent frames, tone, and methodology, each with a confidence score — signals no one would bother tagging by hand, and ordinary search can't see. They're what powers cross-reference detection and theme clustering.
How is this different from Notion or Obsidian's backlinks?+
Backlinks in tools like Notion or Obsidian are links you create yourself. Taggard's connections are found automatically — every note is checked against everything else you've written for five kinds of relationship, including contradictions — and each one shows the exact passage that triggered it, not just a link with no reason attached.
Does Taggard replace my note app?+
No — it's the layer underneath, not a new place to write. Keep using whatever you already capture notes in; drop the result into Taggard and it builds the tagging, connections, and recall on top.
Can I still browse using just the visible tags?+
Yes. Visible tags — the ones styled for ordinary browsing — are always there to filter by; the hidden layer runs underneath without changing how you search when you don't need it.
What happens to my data?+
It's isolated at the database level today via Row-Level Security, with a roadmap toward on-device embeddings and a fully offline desktop option. Taggard never sells your data and never trains models on it.
Feed it one note. Watch the second layer build itself.
Each note you add gets tagged, resolved against your existing entities, and checked for connections to everything else you've written — automatically. Start with anything: a note, a PDF, a Word doc.