For builders shipping search, matching, and discovery products

A theorem, not a conjecture

Embeddings: 6,144 bytes per item, approximate, changes when the model updates, needs a GPU. SBS: 64 bytes, exact, deterministic forever, runs on any CPU. 178 lookups to search a billion items — not a linear scan of millions of vectors.

96× smaller storage. 56× less energy. $0 GPU cost. Same answer every time you ask.

67 end-to-end tests passing. Buy credits. No subscriptions. Your vault stays forever.

Get a free API key to start encoding.

Get API Key — 10,000 free credits

Text in. 8-byte address out.

One API call. Under 200ms. Same answer every time you ask. No model to train. No vectors to store. No GPU to rent.

1

Send any text

Documents, code, logs, emails — anything with characters

2

Get 8 numbers back

Deterministic spectral address — 96× smaller than embeddings

3

Same every time

No model drift. Run it tomorrow, next year — identical result

4

178 lookups, not millions

Flip-graph walk: O(N^¼). At 1M items = 31 lookups. At 1B = 178. Exact, not approximate.

Five levels of encoding. Each one compounds.

Every level is explicitly encoded — nothing “emerges”. Each level adds signal that makes your users' search, matching, and discovery measurably better.

L0

Surface

Included

Your text receives a deterministic spectral fingerprint — 64 bytes, identical every time

Content is analysed, understood, and encoded in a single API call. All AI processing included. 96× smaller than embeddings.

For your product: Ship an instant search feature. No per-query AI cost. Zero model risk. Same result in 2026 and 2046.

L1

Semantic

Included

5-axis meaning decomposition — what content IS and what it ISN'T

Content understood across intent, category, mechanism, context, and exclusions. Distinguishes "fatigue from exertion" vs "fatigue despite rest" automatically.

For your product: Your search distinguishes meaning, not just words. Users find what they actually need.

L2

Contextual

Included

Differential encoding — how this item differs from its neighbours

Catches anomalies and relevance shifts. The document that CHANGED relative to context surfaces, not every mention.

For your product: Anomaly detection, relevance ranking, change tracking — built into the encoding itself.

L3

Fold

Included

Topic coherence — items cluster into knowledge folds automatically

Your data self-organises. Related content gravitates into the same fold. Cross-fold discovery surfaces non-obvious connections.

For your product: Auto-categorisation. Your users' data organises itself. Onboarding cost drops to near-zero.

L4

Portfolio

Included

Cross-fold crystallisation — how knowledge structures relate across your entire corpus

The flywheel. Every new item enriches every fold. The knowledge space crystallises over time — more data = better results for EVERY query.

For your product: Compounding moat. Your product improves with every customer's data. Switching cost is real — they can't export the crystallised structure.

The numbers. Side by side.

Screenshot any card. Send it to your CTO, your CFO, your board. This is the entire business case.

Storage per item

Embeddings

6,144 bytes

OpenAI ada-002: 1,536 floats × 4 bytes each. 1M items = 6.1 GB. 1B items = 6.1 TB. Requires a dedicated vector database.

SBS

64 bytes

8 spectral bins × 8 bytes. 1M items = 64 MB. 1B items = 64 GB. Fits in RAM on one server.

96× smaller. Your entire vector DB bill disappears.

Search 1 billion items

Embeddings

~400 distance calculations

HNSW approximate nearest-neighbour: 400 comparisons × 1,536 dimensions = 614,400 FLOPs per query. Still approximate — misses 5–15% of true nearest matches. Index takes hours to build.

SBS

178 flip-graph lookups

O(N^¼) walk through 4D tetrahedra. Each flip = one orientation test (4 multiplies). At 1M: 31 lookups. At 100M: 100. At 1B: 178. Exact — zero misses.

863× fewer operations. Exact, not approximate. No index to rebuild.

Annual GPU cost

Embeddings

$50,000–$200,000/yr

A100 cloud: $2–$3/hr × 24/7 = $17K–$26K/yr minimum. Production cluster with redundancy: $50K–$200K/yr. Plus CUDA lock-in, driver updates, procurement.

SBS

$0

Pure DFT arithmetic runs on any CPU. Your laptop. A $500 server. A Raspberry Pi. An edge device. No CUDA. No drivers. No procurement cycle.

Delete your entire GPU line item. Redeploy that $200K.

Carbon per 1M encodes

Embeddings

~4.5 kWh

A100 GPU at 300W for inference + vector DB servers running 24/7 + cooling overhead. At scale: tonnes of CO₂ per year.

SBS

~0.08 kWh

CPU at 65W × 1ms per encode × 1M items = 65 Wh total. No idle GPU. No vector DB baseline draw. No cooling premium.

56× less energy. Put this number in your ESG report.

Vector DB hosting (1B items)

Embeddings

$72,000/yr

Pinecone p2: ~$6,000/mo for 1B vectors. Self-hosted: 6.1 TB index across 8+ servers with 768 GB RAM each. Either way — expensive, complex, fragile.

SBS

$0 extra

64 GB total index. One server. Your existing Postgres or Supabase. No new vendor. No new infrastructure. No new ops burden.

$72K/yr saved on vector DB alone. No new vendor contract.

Model update risk

Embeddings

Re-embed everything

OpenAI deprecated ada-002 for ada-003. New model = new vectors = re-embed your entire corpus. Weeks of compute. Broken similarity during migration. Explain the regression to your users.

SBS

Zero risk — mathematical proof

DFT is a theorem, not a trained model. Same input produces the same 64 bytes today, next year, and in 2046. No migration. No re-embedding. No search regressions. Ever.

Never re-embed. Never migrate. Never explain a regression.

Encode

1–5p

Gets faster as axioms freeze

Search

Free – 1.5p

First 1,000/mo free. Then volume-priced.

Surface

3–10p

Discovery worth more at scale

8 markets. One API.

You build the product. SBS powers the intelligence underneath. Each market shows where embedding APIs fail and SBS compounds.

Click any market to see the engagement, retention, and cost case for your product.

Ship in an afternoon.

Five steps from payment to production. No onboarding calls. No integration meetings. Your lead engineer can ship this before lunch.

1

Pay → API key in 60 seconds

Stripe checkout. Key provisioned automatically. No human in the loop. Works immediately.

Authorization: Bearer sbs_live_your_api_key
2

Batch encode your existing corpus

Send up to 100 items per request to /v2/batch. Each item gets a unique ID you provide. 12,000 contracts = 120 requests = under 30 seconds.

POST /v2/batch
{ "items": [{ "id": "doc-1", "content": "..." }, ...], "maxLevel": 4 }
→ 100 items encoded in ~2 seconds
3

Encodings stored & indexed automatically

Every item you encode is added to your private flip graph — searchable immediately. You also get the raw 64-byte encoding back to store in YOUR database for analytics, backup, or client-side filtering.

Server-side: instant search via flip graphYour side: raw 64-byte encodings returnedIsolated per API key — multi-tenant by default
4

Wire up real-time search

Your users' queries hit /v2/search. Results in <3ms. The flip graph walks 178 tetrahedra at 1B items — not a linear scan. Exact results, zero misses.

POST /v2/search
{ "query": "penalty clause for delayed shipment", "maxResults": 10 }
→ 10 ranked results in 2.3ms
5

Every new item makes search better

New content joins the flip graph and auto-folds into topic clusters (L3). Over time, the knowledge space crystallises (L4) — more data = better results for EVERY query. This is the compounding moat your competitors can't replicate.

L3: Auto-clusteringL4: Crystallisation flywheelData moat: can't export the structure

API Reference

Four endpoints. One API key. Full schema below.

POST /v2/encode
// Encode any content — 64 bytes out, deterministic, no GPU
const res = await fetch('https://api.sbs/api/sbs/v3', {
  method: 'POST',
  headers: {
    'Authorization': 'Bearer sbs_your_api_key',
    'Content-Type': 'application/json'
  },
  body: JSON.stringify({
    action: 'ingest',
    text: 'Indemnity clause for late delivery under FOB terms'
  })
})

const { binary, omega, bigOmega, level, foldId } = await res.json()
// binary:    "10110101"       — 8-bit spectral address
// omega:     [0.42, 0.87, 0.15, 0.63]  — Kähler moduli (IS/NOT twist)
// bigOmega:  0.034            — volume form (crystallisation)
// level:     4                — deepest encoding achieved
// foldId:    "shipping-law"   — auto-detected topic cluster

Response Fields

FieldTypeDescription
binarystring8-bit spectral address
omeganumber[4]Spectral moduli — the encoding's position in 4D space
bigOmeganumberVolume form — 0 = uncrystallised, >0 = knowledge crystallised
level0-4Deepest encoding level achieved (L0 surface → L4 portfolio)
foldIdstringAuto-detected topic cluster this item belongs to
similaritynumberSearch only — spectral similarity score (0–1)
flipCountnumberSearch only — tetrahedra traversed (not linear scans)

Authentication

Bearer token in Authorization header. Key provisioned on payment.

Rate Limits

Encode: 1K/sec. Search: 10K/sec. Batch: 100 items/req, 100 req/sec.

Namespaces

Content isolated per API key. Multi-tenant by default. One key per environment.

SDKs

Node.js, Python, Go — available post-checkout. Or raw HTTP: four endpoints, that's it.

Credits. No subscriptions. Ever.

Buy credits. Use them when you want. Stop when you want. Your vault stays forever. Credits never expire. The more you use it, the faster it gets.

No free tier. No sales calls. Same door for everyone.

Three operations. Three prices.

Encode

Crystallise knowledge

1 – 1,000/mo5p
1,001 – 10,0003p
10,001 – 100,0001.5p
100,001+1p

Includes L0-L4, OWL decomposition, IS/NOT ledger, fold assignment, axiom freezing, persistence forever. Gets faster as axioms accumulate.

Search

Find what you need

1 – 1,000/moFree
1,001 – 100,0000.5p
100,001 – 1,000,0001p
1,000,001+1.5p

All search types: multi-level, flip search O(N¼), compare, disambiguate. Gets faster as vault grows — sublinear algorithms + cached topology.

Surface

Discover what you didn't know to look for

1 – 1,000/mo3p
1,001 – 100,0005p
100,001+10p

Resonance check against entire vault. Cross-fold L4 mining. Gets smarter as folds crystallise. Price rises with vault value — discovery in a 100K vault is worth more.

Buy credits. Never expire.

Not a subscription. A deposit. Use them whenever.

£50

£50 credit

Buy Credits

£200

£230 credit

+£30 bonus

Buy Credits
Popular

£1,000

£1,300 credit

+£300 bonus

Buy Credits
Enterprise

£10,000

£15,000 credit

+£5,000 bonus

Buy Credits

What it costs in practice

Solo developer

200 encodes → £10

800 searches → Free

300 surfaces → £9

~£19/mo

Startup (10 devs)

15K encodes → £225

50K searches → £250

10K surfaces → £150

~£625/mo

Enterprise

700K encodes → £7,000

5M searches → £75,000

200K surfaces → £20,000

~£102K/mo

The rules

No subscriptions. Credits only. Top up when you want.
Credits never expire. Buy today, use in 2028. We don't care.
Vault never deletes. Stop paying = read-only. Resume = everything's there.
No free tier. Sandbox for testing. Real use = real credits.
Same door for everyone. Solo dev and enterprise buy the same credits, same API, same maths.
Price rises with vault value. Surface events cost more at scale because they're worth more at scale.

96×

smaller than embeddings

<200ms

encode latency

64

bytes per signature

67/67

E2E tests passing

Buy credits. Get your key. Ship.

No onboarding calls. No integration meetings. The API key works the moment payment clears.

1

Buy credits

Stripe checkout. Pick a credit package. Your API key is provisioned automatically — no human in the loop. Credits never expire.

2

Encode

POST your content. Credits deducted per operation. Store encodings in YOUR database — Supabase, Postgres, anything. You own the data.

3

Ship

Build search, matching, discovery into your product. Your users see better results. You keep the margin. The flywheel starts.

Credits Policy

Credits never expire. Your vault never deletes. Stop using = vault goes read-only (L0 search stays free). Resume = everything's there. No subscriptions. No lock-in. The maths works or it doesn't — we don't need to convince you.

Authenticated by BioCaptcha

Every API.SBS key is backed by BioCaptcha topology verification. Your vascular fingerprint — not a password — proves you're human. Zero PII stored. GDPR-proof by architecture.