Skip to main content

Semantic Librarian

A free AI-powered search engine for Australia's public archives.

One search across 11 Australian archive APIs simultaneously — Trove, PROV, Museums Victoria, NMA, and more. Semantic search finds records that keyword search misses, across hundreds of institutions and hundreds of millions of records. Free, non-commercial, attribution-first.

R&DResearch in progress — not available yet
Register Interest

How Semantic Librarian works

From question to discovery in one search.

What you can search

11 APIs. Hundreds of institutions. Hundreds of millions of records.

Each API is a gateway to dozens — sometimes thousands — of contributing organisations, datasets, and collections.

National Institutions

  • National Library of Australia
  • National Archives of Australia via Trove
  • National Museum of Australia
  • Australian War Memorial via Trove
  • National Gallery of Australia via Trove
  • National Film & Sound Archive via Trove
  • AIATSIS via Trove
  • CSIRO via Trove + ALA

State Libraries

  • State Library of NSW via Trove
  • State Library Victoria via Trove
  • State Library of Queensland via Trove
  • State Library of SA via Trove
  • State Library of WA via Trove
  • Libraries Tasmania via Trove

Museums

  • Melbourne Museum
  • Scienceworks
  • Immigration Museum
  • Royal Exhibition Building
  • Australian Museum via Trove + ALA
  • Powerhouse / MAAS via Trove
  • Queensland Museum via Trove + ALA

Government & Heritage

  • Public Record Office Victoria
  • Heritage Council of Victoria
  • Geoscience Australia
  • Dept of PM&C
  • State Records NSW via Trove
  • Parliament of Australia via Trove

Science & Biodiversity

  • Atlas of Living Australia
  • BirdLife Australia via ALA
  • iNaturalist Australia via ALA
  • Australasian Virtual Herbarium via ALA

Digital Humanities

  • ACMI
  • TLCMap / GHAP
  • AusStage via TLCMap
  • Colonial Frontier Massacres via TLCMap

Types of records

Newspaper articles (1803–present) Historical photographs & maps Museum objects & artefacts Government archives Heritage places & shipwrecks Aerial photography (1928–1996) Species & biodiversity records Historical placenames PM speeches & transcripts Film, TV & videogames Court & immigration records Aboriginal and Torres Strait Islander cultural materials

Plus hundreds more

1,500+ Trove contributing organisations 9,757 ALA biodiversity datasets Hundreds of TLCMap community layers 37+ university repositories

What Semantic Librarian includes

  • Semantic search, not keyword matching

    Search by meaning, not exact words. 'Photos of Melbourne in the 1950s' finds records described as 'aerial photograph, City of Melbourne, 1952' — even though the words don't overlap. Historical language, inconsistent cataloguing, and spelling variations are no longer barriers.

  • One search, eleven archives

    A single query searches Trove, PROV, Museums Victoria, NMA, VHD, ALA, ACMI, Geoscience Australia, GHAP, PM Transcripts, and IIIF collections simultaneously — from Trove's 1,500+ partners to ALA's 9,757 datasets. No more opening 11 tabs or learning 11 different search interfaces.

  • Attribution built into the data model

    Every result carries its source, licence, and attribution text. CC-BY compliance is structural — the _attribution array in search responses ensures downstream consumers can display proper credit automatically.

  • A card catalogue, not a library

    Semantic Librarian indexes metadata only — titles, dates, descriptions, coordinates, subjects, and thumbnail URLs (hot-linked, never re-hosted). It never stores image binaries, full-text, or audio. Every result links back to the original source for the actual document, image, or record.

Transparency

Metadata Only — Never Re-Hosting

Indexes titles, dates, and descriptions. Thumbnail URLs are hot-linked back to the source institution; no images, audio, or full-text are ever re-hosted.

Attribution-First

Every result carries its source, licence, and attribution text. CC-BY compliance is structural.

Non-Commercial — forever

Free research tool. No subscriptions, no paid tiers, no for-sale offerings — ever. If running costs ever justify it, the only commercial-adjacent move that's permitted is registering as a not-for-profit purely to receive donations toward infrastructure. The user-facing service stays identical either way.

Good Neighbour

Polite rate limiting, descriptive headers, and respect for institutional policies.

Takedown-Respecting

Honours per-record removal requests within 30 days, in line with source institutions' content policies. Cultural-sensitivity protocols apply for Indigenous content (per AIATSIS guidance and individual community wishes).

How we work with source institutions

Each source institution's API Terms of Use or open-data licence governs how we use their data. The commitments below apply across the board.

  • Apply formally where required. Trove (NLA), AWM (in flight 2026-04), and WA Museum Sandbox (in flight 2026-04) require API keys via formal application. Whatever rate-limits, attribution, content scope, and removal SLAs each institution requires, the project commits to them.

  • Index metadata only. No articleText from Trove (per Trove §4.4(e), enforced after the February 2025 Sherratt incident). No image binaries from AWM (per AWM §Copyright). No oral-history audio from anywhere. Thumbnails are hot-linked URLs, never re-hosted.

  • Honour takedowns within 30 days. Per AWM §Removal of content. The same policy applies as a courtesy across all sources.

  • Respect rate limits. Per-source sliding-window rate limiter, daily and weekly budgets, a 1100ms politeness floor, and exponential backoff on 429. A pre-commit linter blocks Trove-incompatible content classes from ever entering the index.

  • Attribute every record. Source name, licence, and a link to the original record on every search result. Many results carry the "Provided by [Institution]" logo where the institution provides one.

  • Cultural-sensitivity protocols. Indigenous content carries record-level advisories where applicable, in line with AIATSIS guidance and individual community wishes.

  • Non-commercial only — forever. Required by some source ToUs (notably AWM §Commercial Use); aligned with the project's own permanent commitment regardless. The only future commercial-adjacent move that's permitted is registering as a not-for-profit purely to receive donations toward infrastructure costs.

If you're an institutional partner with a question about how Semantic Librarian uses your data, contact nathan@littlebearapps.com.

Get notified when Semantic Librarian is ready

Leave your details and I'll let you know when it's available.

No spam, ever. Privacy Policy

"I kept falling down rabbit holes on Trove — reading old newspaper articles from the 1920s, browsing PROV government archives from 150 years ago, exploring heritage buildings on VHD. But finding things was painful. Each archive had its own search, its own quirks, its own pagination. I'd spend an hour searching one site, then start again on the next. The collections are extraordinary. The search experience is not."

Nathan
N
Nathan

Who It's For

🎓

Academic Researcher

"How do I find primary sources across multiple archives for my thesis?"

Researchers who need to search comprehensively across Australian archives for a specific topic, period, or region.

You need: Cross-archive discovery that finds records keyword search misses.

🔍

History Enthusiast

"What happened in my local area 100 years ago?"

History buffs who explore out of curiosity — browsing old newspaper articles, looking up heritage places, discovering photographs of their town.

You need: A single search that finds the interesting stuff hidden across multiple platforms.

🌳

Genealogist

"How do I find records of my ancestors across different archives?"

Family history researchers tracing lineage through immigration records, court documents, newspaper mentions, and government archives.

You need: Cross-referencing records across archives that would take hours to search individually.

📚

Educator or Student

"Where can I find primary sources for my Australian history assignment?"

Teachers and students looking for original documents, photographs, and artefacts to support learning and assignments.

You need: Easy access to verified, attributable primary sources without navigating 11 separate platforms.

Questions about Semantic Librarian

What is Semantic Librarian?
Semantic Librarian is a free, non-commercial research tool that gives AI a unified metadata library across 11 Australian historical archive platforms. It uses semantic search to help you find historical images, newspaper articles, documents, maps, and records — without manually searching each site one by one.
Why is it called Semantic Librarian?
Honestly, it's a working title — I'll likely change the name before any public release. For now it describes what the tool does: a librarian that understands meaning, not just keywords. 'Semantic' because it uses AI to search by meaning rather than exact word matching. 'Librarian' because it knows every collection, every catalogue, and can point you to things you didn't know existed. It doesn't create or store the works — it helps you discover what's out there, then sends you to the original source.
Which archives does it search?
Semantic Librarian connects to 11 archive APIs, but each API is a gateway to many more institutions. Trove alone aggregates content from 1,500+ contributing organisations including every state library, the National Archives, the Australian War Memorial, and dozens of universities. The Atlas of Living Australia draws from 9,757 datasets across hundreds of museums, herbaria, and citizen science platforms. In total, Semantic Librarian gives you unified search across hundreds of institutions and hundreds of millions of records.
Is Semantic Librarian available to use?
Not yet as a public tool. Semantic Librarian is in the workshop — I'm using it personally and validating that the search results are genuinely useful across all 11 sources. There's no public release planned until the tool consistently helps people find things they couldn't find before.
Does Semantic Librarian store the actual content from archives?
No. Semantic Librarian indexes metadata only — titles, dates, descriptions, coordinates, subjects, and thumbnails. It never stores full article text, high-resolution images, or complete documents. Every result links back to the original source for the actual content.
What types of records can Semantic Librarian search?
Newspaper articles from 1803 to the present, historical photographs and maps, museum objects and artefacts, government archives and correspondence, heritage places and shipwrecks, historical aerial photography from 1928 to 1996, species and biodiversity records, historical placenames with coordinates, Prime Ministerial speeches and transcripts, films, TV programmes, videogames, court and immigration records, and Aboriginal and Torres Strait Islander cultural materials.
How many records are searchable?
The 11 APIs provide access to hundreds of millions of records. Trove alone aggregates approximately 308 million records including 254.6 million newspaper and gazette articles. The Atlas of Living Australia holds 101.5 million species occurrence records. Geoscience Australia has 1.2 million historical aerial photographs. Museums Victoria holds 17.2 million items. The actual number searchable through Semantic Librarian's semantic index is smaller — currently around 3.5 million indexed records with metadata, growing weekly — but every result links back to the full record at the original source.
Why share this if it's not available?
The problem of siloed Australian archives is real and widely felt. Sharing the approach — semantic search across unified metadata, with proper attribution — might be useful to others thinking about similar challenges. And the principles behind it (metadata-only, attribution-first, non-commercial) are worth discussing openly.

Also by Little Bear Apps

channelo R&D
A unified marketing data warehouse — own your data, see your costs, catch what's broken.
Outlook Assistant
22 tools for email, calendar, contacts, and more — all through natural language.
PitchDocs
16 slash commands that generate READMEs, changelogs, roadmaps, and llms.txt files from your codebase.
← Back to All Builds