Two hundred spines. Four of them are why you came.
Two other dealers working the same wall, and no time to open more than a handful. BiblioSignal reads every spine in one photograph and tells you which four are worth your hands — including the ones you would have walked straight past. Then it tells you why: what the edition is, what makes it scarce, and what to check before you believe it. Each one carries a price too — the most you can pay and still come out ahead. Walk past the other 196 knowing what they were.
Same at a library sale, a thrift store, or a shop’s back room.
The number on the card is the number on the shelf.Leads are ranked by what they pay — 1 is the best book in the bay — and each one is marked roughly where it sits, so you know which shelf and which end of it to work. Note that 1 is not on the shelf you would have started with. Tap any spine, including the dim ones and the hatched patches it could not read.
A working preview with illustrative figures. The real screen is your phone, in the room, with the book in your hand.
Buying to keep or buying to sell, every scan teaches you something.
A price on its own tells you nothing you can carry to the next shelf. So every scan comes with the reasoning behind it — which cluster on this wall is the interesting one, why the third printing is a four-dollar book when the first is not, which title on the shelf is the trap that looks right and isn’t.
Most of that is identification, not arbitrage. A collector asks is this the printing I think it is? to avoid a mistake; a dealer asks it to make a margin — and most of us are both, depending on the book. Read it or skip it; the verdict sits on top either way. But read a few dozen and you stop needing it for the easy ones. That is the point.
- What the book actually is. Publisher, printing, issue points and edition history — the documented points for that title, checked against the book in your hands, never a generic rule about number lines.
- Why this one is scarce and that one isn’t. How many copies exist anywhere, read live — not a feeling about how old a book looks.
- What dealers are actually asking. The full spread with sample sizes, so a single optimistic listing never sets your expectation.
- What you should not pay. A ceiling that moves with condition and confirmed points, so an auction room does not do your arithmetic for you.
- How this room prices its stock. What is displayed versus what is shelved, and what that tells you about the rest of the wall. The part that transfers to every sale after this one.
The chain that puts rare books in your hands.
Twenty minutes in the room is the visible part. What makes the next trip better is everything after it — each step feeds the next, and the last one feeds every future price the tool gives you.
- 01
Photograph the bay
One picture, about three shelves. No forms, no setup — scanning with no trip open starts today’s trip for you, and you can name the venue later or never.
- 02
Get leads, with reasons
A vision model reads the spines off the photograph and matches them against a researched watchlist; an algorithm prices what it finds. Each lead carries the spine text it read, what it believes the book is, why, and how sure it is — plus an honest count of the spines it could not read, and where they are.
- 03
Everything below your bar gets held, not hidden
Book-club bindings, mass-market, decoratives, obsolete editions — suppressed by structural rules you write, each suppression citing the rule by name. A title on your watchlist overrides every filter and comes straight back.
- 04
Confirm it, then cart it at what you actually paid
The identification points for that title and publisher, as a checklist, with the camera one tap away. Tick one and the algorithm re-prices before you have put the book down — confirming the printing raises what you can pay, failing to confirm it lowers it rather than quietly hoping. What you hand over goes in the cart, with the running average basis in the header and a warning when an add drags it past your target.
- 05
Then it sells — or it doesn’t, because you kept it
Every sale is fed back into the algorithm, so the prices it gives you are corrected by your own results rather than by an average. After a season it also tells you which sales are worth the drive — the number nobody can look up for you.
The hunt is the fun part. The rest of it should run itself.
Twenty minutes in a room is the part worth having. The hours after it — relisting, remembering what you paid, checking whether the eBay search you set up in March ever turned anything up — are the part that quietly turns a hobby back into work. So they are automated, and the time comes back to you.
It hunts while you are not hunting
Your watchlist is checked on a schedule, and when a copy appears at a price worth paying it reaches you as an alert instead of a search you forgot to run.
Priced by the same algorithm, with one difference it tells you about: buying online you pay shipping too. Some books are worth six dollars at an estate sale and worth nothing at any online price. It says which are which.
Every copy, from the shelf to the sale
What you paid, where you found it, how long it sat, what it made — carried from the cart into an inventory row with its cost and provenance attached. When it goes out the door, a printed accession card and authentication record travel with it, sharing one certificate number.
“Where did I get this one?”
Listed once, in three places
One record becomes an AbeBooks, an eBay and an Etsy listing, with the copy written for each marketplace and nothing going live until you confirm it.
One case still manual: a sale on eBay or Etsy takes the copy down everywhere within five minutes, AbeBooks included. AbeBooks reports its own sales by batch rather than by API, so a sale there is the one you still mark yourself.
Seven things set your number — and knowing them makes you better at this.
These aren’t “what the book is worth.” They’re what decides whether handing over the marked price is a good idea — the same seven things a dealer who has been at this thirty years weighs without writing them down.
This is the part a chatbot cannot do. Not because any single factor is exotic — several are public — but because the answer depends on how they are weighed against each other, and against your costs and your patience. Ask a language model twice and you get two confident opinions that do not agree; the same shelf on Tuesday and on Friday gives you the same answer here, because it is a model and not a mood. It also prices your business — your selling costs, your patience, your walk-away floor — not the market’s average. What it weighs is below. How much each one counts is ours — built and corrected against real trips and real sales.
One factor, worked through. A stated first, three copies on offer, dealers asking about $300.
A true first and a book-club copy are different books at different prices. Until it is confirmed, the doubt is priced in.
Confirming the printing with the book open is worth $20 on this copy. That is one factor of seven. The algorithm is how they combine.
- What dealers are askingread live from the open market, not a price guide
- How long this kind of book takes to sella year on the shelf costs you money
- How much dealers disagreewhen asks run wild, nobody knows — dealers included
- Whether you have sold one beforeyour own sales outrank every other signal
- How many copies exist anywhereread live on every scan
- The return you are aiming foryours to set — every price on every trip follows it
What it will not do.
This is the part most tools leave out, and it is the reason to trust the rest of the page. Every one of these is a live constraint, not a roadmap item.
It will not tell you it found a winner
Every result is a lead carrying its evidence and its doubt, and the price it gives you moves with how sure it is. UNCLEAR is a real answer — a false “first printing” costs money; a missed deal only costs a deal.
The AI identifies the book — and says so when it can’t
A vision model reads the spines and matches them against a researched watchlist it knows deeply, or against publicly available data. Unclear pictures mean it may fail on a third of them — you can try for a better photo, but vision models are not perfect. Where it cannot place a title, it says so rather than inventing a match.
It cannot grade a book from a spine
Condition, jacket state, the points that matter — those need the book in your hands. What it can do is tell you exactly what to check, and what each check is worth to the price.
It is not finished
It scans and prices books today. What is still accumulating is history: every scan and every sale adds to the record behind the prices, and we test against that record as it grows. We are also working on live sold data — for now it shows asks, labelled as asks, next to the sold prices recorded over time.
Want to try it on your own shelves?
The scanner runs in one shop today. The list decides when it opens wider, and in what order. Tell me what you hunt and where — estate sales, library sales, a shop — and that shapes what gets built next.