the corpus instrument · specolia.ai
Point Specolia at any collection — a library, an archive, your own files — and it becomes a space you move through. Cited answers, living maps, an output studio. Local-first. Voice-guided.
What it is
Search gives you a list. Specolia gives you the shape of what you know, and a way to move through it — from the whole sky of your sources down to the exact page, without losing your place.
Ask in plain language or by voice. Every answer streams in with source markers — hover for the passage, click to open it. Nothing is a claim without its page.
Ask · ReaderThemes, lineages and concepts rendered as a space you fly through. Pick a lens, dive from geography to a place to its idea-graph to the source. The lens is the aperture.
Maps · MirrorTurn what you find into podcasts, cited papers, decks or a documentary — every claim chained back to the card and the source it came from. Your synthesis, kept honest.
StudioHow it works
You always know the cost before you spend it, and what leaves your machine before it does. The instrument is useful partway through — Ask goes live as soon as the first pages are in.
Drop a folder, a mail archive, a drive. Specolia surveys it — file types, scans needing OCR, languages, duplicates — before anything leaves your machine, and tells you so.
A matrix of routes — local, your cloud, API — each cell showing time, cost and whether data leaves. Set a hard cap. Two giant numerals: total cost, total time.
Dedupe, OCR, embed, caption. A live spend line rises toward your cap; pause and resume are one key away. Ask is live over the corpus as it fills.
Conversation with the corpus, by text or voice. Cited answers, with a confidence tick and a source glyph on every marker. Click to read the passage.
Open the map. Geography → a place → its concept-graph → the passage, one unbroken spatial dive, with breadcrumbs. Switch the lens; the space re-forms.
Send a selection to the Studio and build a podcast, paper, deck or documentary — the claim outline first, every claim chained to its evidence, then render.
The spaces
One hairline rail, no clutter. The Reader slides over whatever you're in, so you keep your context. The Ask line is always there, like a subtitle.
What's in this project — files, collections, ingest state, duplicates.
Cited conversation with the corpus, by text or voice.
Concept, theme, timeline, lineage and iconography, as a space you fly through.
Close reading, highlights and margin notes, anchored to the source.
Podcasts, papers, decks and documentaries, built from your notebook.
Your own attention reflected back — what you've asked, read and never opened.
Jobs, routes, spend and the queue. Pause and resume, always.
API keys, cloud accounts, local models and privacy routes — never displayed.
The lens is the aperture
Specola means a place to watch the sky. A lens is how you choose to look — by topic, by time, by source, and more. Switch it and the whole graph re-lays-out around that choice, with an animated morph. The choice is the instrument. It's the name.
A working set — the exact lenses evolve with your corpus.
Corpora & examples
A corpus with places opens on a map; one without opens straight into its ideas. Specolia keeps each project sealed and separate — the tool is shared, the data never is.
AI/ML research abstracts (cs.AI, cs.LG, cs.MA, stat.ML) from Cornell's arXiv snapshot.
one cited abstract per paper~500 books of world mythology, folklore, fairy tales and the occult, from Project Gutenberg.
~102k passagesThe full Request for Comments series — the standards and lore of the internet.
~9,500 documents · ~172k passagesWikipedia's curated Level-5 Vital Articles — a cross-section of human knowledge.
~28.5k articlesWorks of Greek and Latin antiquity from the Perseus Digital Library.
~1,140 works · ~60k passagesThe latest annual report for each S&P 500 company, from SEC EDGAR, sector-tagged.
~500 filings · ~57k passagesTop-cited 2024 open-access biomedical abstracts from Europe PMC.
~13k abstractsNTRS report abstracts across ten aerospace fields — propulsion to astrophysics.
~13k passagesArchaeoastronomy and the record of Teotihuacan. A sealed project — its contents never leave, and never reach a public surface.
41,088 passages · US-only routesA research library, a company archive, a lifetime of reading. Point Specolia at it and it becomes an instrument.
you set the capEight open corpora — arXiv, Wikipedia, Perseus, PubMed, NASA, SEC, RFCs, myth — rendered as a map you can click into. Real data, real citations.
Access & voice
The desktop is the server, because the product is local-first. Your phone reaches it over a private tunnel — never a public endpoint — so the instrument travels with you.
It runs on your machine. Your corpus never leaves unless you route it out, and the screen tells you the moment it does. One file is your backup.
Reach it from your phone or your car over a private tunnel to your own devices. Review, listen, ask — the heavy views follow when you're back at the desk.
Direct it by voice. It mostly just acts and acknowledges — a blink, not a monologue — and talks only when you want a conversation. The guide inside the space.
What it costs
No subscription for the work itself — Specolia quotes each ingest up front: time, cost, and exactly what leaves your machine, with a hard cap you set. Bring your own keys, or run local models for free. No surprise bills, ever.
Estimate · 1,284 files · 61,200 pages
~ 2h 10m · pause anytime · example figures
Specolia is in early access. Tell us what you'd point it at, and we'll set up your first instrument.
hello@specolia.ai