BantuNomics Enter the ecosystem
The catalogue

Everything we offer, and how to reach it.

The BantuNomics substrate is one program of 10 products across the Bantu-language AI stack. This is what each is, why it matters, what you can use right now versus what a license unlocks, and the access ladder that governs it.

10 products 5 access tiers 459 FSIs released One token · every subdomain

How the platform is built — the living foundation →

The access ladder
Five tiers, one substrate.

Public proves the gap. Evaluation opens a limited slice. A pilot licenses an agreed set with curation. A Founding Partnership licenses the whole program.

TierPriceLanguagesAudioCuration
public
Prove the gap for free — the Alphabet Test, teasers, agent discovery. No login.
$0 none
evaluation
Verify on a limited slice across every subdomain with one token. Invite, no cost.
$0 · invite Bemba + Nyanja cap 5
pilot
License an agreed set for a fixed term, measured on your own held-out data, with the curation moat.
≈$150K · 75d agreed set cap 25 ✓ moat
founding partner
Not a dataset — the whole living program. Every product and domain (syllables, verbs, tone, health, numbers, and each one added next), every language, the full consented audio corpus, all exports and APIs — and everything curated while you're subscribed. It's a subscription to a platform that grows, not a one-time file. The ≈$1.75M/yr is the FOUNDING rate, locked two years; it rises afterward — founding partners get in early.
≈$1.75M / year every language full corpus
showroom
Invite-only funder view of the whole ecosystem your support makes possible.
invite every language cap 15
The economics of the bundle
Every recording is six datasets.

One consented, time-segmented Bantu→English recording is not a single labelled example. It is the ground truth for six things frontier labs otherwise buy from six vendors — so a Founding Partnership licenses the whole substrate, not a file.

01

Code-switch ASR

Natural Bantu→English switching with a known switch point — pre-labelled train + eval data.

02

Bantu-accented English

The English half, spoken by native Bantu speakers — accent-robustness and fairness data.

03

FSI alignment target

Each word decomposes into its language's Full Syllable Inventory — the tone-bearing units an aligner needs.

04

Pronunciation & TTS

Consented native pronunciations keyed to meaning — a clean, tone-aware basis for synthesis.

05

Self-grading benchmark

A turnkey accent-robustness eval with a frontier baseline — the segmentation is already structured, so no labelling vendor and no drift.

06

Domain & homograph truth

Clinical terms, spoken numerals, tonal minimal pairs — the meaning flat text can't encode.

The products
What's in the catalogue.

Ordered by maturity. Each card separates what is open to use now from what a license unlocks.

Full Syllable Inventories (FSIs)

Syllabic Alignment Matrices
Production · licensing core
BTS-S100 / BTS-API-100

The operating alphabet of a Bantu language — the complete, closed set of its legal syllables. Infrastructure, not a dataset.

Why it matters — No frontier model can produce even one unaided; it stops BPE tokenizers from fracturing syllable boundaries and gives aligners the tone-bearing units.

Use now

  • Alphabet Test (/alphabet-test)
  • Operating-alphabet benchmark
  • coverage.json
  • why-fsi essay
  • llms.txt + MCP descriptor

Licensed

  • Full FSI matrices per language (/api/v1/fsi/<iso>)
  • Consented 48kHz audio (NSI+ASI) recordings
  • Bulk export (/api/v1/export)
  • Full Atlas
Public proves the gap; evaluation opens Bemba+Nyanja; pilot unlocks an agreed set with curation; partner licenses all 459.
Open Full Syllable Inventories (FSIs) →

BTS Standards

Governed Bantu standards
Released · versioned
BTS-S100 · BTS-API-100 · BTS-E100 · BTS-UCM100

The auditable, versioned standards the whole ecosystem is built to — syllables, API, equations, concord.

Why it matters — Enterprise adoption needs a governed, citable specification, not a moving target.

Use now

  • Standards hub
  • 4 standard overviews (scope, versions, section maps)
  • 2 reference abstracts

Licensed

  • Full downloads / exports
  • Governed change data
  • Admin
Overviews are public; the governed artefacts and change history are gated.
Open BTS Standards →

Equations

Formal language equations
Developed · registry live
BTS-E100

A formal equation registry for syllable, morphology, numeral and syntax construction across the family.

Why it matters — Turns linguistic rules into executable, testable equations a model can be graded against.

Use now

  • Teaser home
  • /try/lab (Bemba, 4 noun classes)
  • /api/try/compose-noun
  • 3 sample equations

Licensed

  • Full registry (/sets, /equations)
  • Full /api/equations (+ csv / jsonl)
  • Full lab
A live Bemba sample + 3 equations are public; the full registry and exports are gated.
Open Equations →

Tone and Flat Text

A12 codec and tone lab
Developed · demo + corpus
A12 codec

Tone as meaning: the flat-text problem, the ulebomba demonstration, and the A12 tone codec path.

Why it matters — Flat text drops the tone that carries meaning; the corpus proves the homograph gap models can't self-certify.

Use now

  • ulebomba demonstration audio (the public demo)
  • why-it-matters (6-in-1)
  • llms.txt + MCP

Licensed

  • The 81-homograph / 158-take consented corpus audio
  • Proof set
  • ASR benchmark clips
Only the ulebomba demo audio is public; the homograph corpus is licensed.
Open Tone and Flat Text →

Universal Concord Matrix

Agreement engine
Developed · matrices live
BTS-UCM100

The Universal Concord Matrix — the agreement engine mapping noun class to every concord across 31 dimensions.

Why it matters — Bantu agreement is the grammar frontier models miss; the matrix is the ground-truth answer key.

Use now

  • /welcome (concept + 31 dimensions + coverage)
  • /sample (Bemba, 6 of 31 dimensions × 8 classes)

Licensed

  • Full 31-column matrices
  • /compare, /evidence
  • All /api/ucm/* exports (json/csv/xlsx)
A Bemba slice is public; full matrices, comparison and exports are gated.
Open Universal Concord Matrix →

Nouns

The Bantu noun, aligned across the family
Integrated · English-anchored matrix live across the family
ABS-400-000

The Bantu noun as a class-driven equation — NOUN = AUG(cl) + PREFIX(cl) + STEM — with the full class + concord system, an interactive builder, and the English-anchored aligned matrix: one concept mapped 1:1 to each language's noun (singular and plural), 352 languages, 127 complete.

Why it matters — The ref_id concept spine gives labs an aligned multilingual noun matrix — not a bag of words — and the class system + consented 6-in-1 code-switch recordings encode the noun-class grammar (concord, sg/pl pairing) flat text can't. It's about the whole family; Bemba leads the trail.

Use now

  • Noun-class equation + interactive builder (/equation)
  • The concept spine (83 anchors) + the aligned-matrix teaser across the family (/matrix)
  • Class + concord system (/classes)
  • Per-language coverage (/languages)
  • llms.txt + MCP

Licensed

  • The full aligned matrix across all languages
  • The generative engine batch + /api/v1/nouns
  • The consented audio corpus + bulk export
Public gets the equation, builder, class system, concept spine, and a cross-family matrix teaser; the full aligned matrix, uncapped engine, audio corpus, and bulk export are licensed.
Open Nouns →

Numbers

Bantu calculator studio + native voice
Integrated · calculator live
BTS-S100

The Bantu numeral substrate + calculator studio, plus two-mode native numeral audio (clean Bantu AND the real Bantu→English code-switch).

Why it matters — Spoken numerals are where code-switch and ASR break; the calculator API is a no-key wedge, the audio is the scarce layer.

Use now

  • Calculator + corpus API (/api/bantu-calc/*, no key)
  • why-it-matters (6-in-1)
  • llms.txt + MCP

Licensed

  • Two-mode audio clips (bantu_only + bantu_english)
  • Studio
  • Bulk corpus
Public gets the full no-key calculator + coverage/consent; the audio clips need a key.
Open Numbers →

Verbs

The Bantu verb equation
Integrated · Bemba paradigm live
BTS-V100 / BTS-E100-115

The Bantu verb as a 9-slot generative equation — one root expands into its whole paradigm (53 verb types) through the slots — with the full morpheme inventory, an interactive builder, and a 37-chapter tutorial.

Why it matters — Verbs carry most of a Bantu sentence's grammar; the generative engine + consented 6-in-1 code-switch recordings give models the paradigm flat text can't encode and monolingual corpora lack.

Use now

  • 9-slot equation + interactive verb builder (/equation)
  • Tutorial preview (/tutorial)
  • Per-language VT structure + live coverage (/languages)
  • 6-in-1 recordings demo (/why-it-matters)
  • llms.txt + MCP

Licensed

  • The full 37-chapter interactive course
  • The generative engine + /api/v1/verbs (build/decompose)
  • The consented audio corpus
Public gets the equation, builder, structure, and a preview; the full course, engine API, and audio corpus are licensed.
Open Verbs →

Body and Health

Health language packs
Active collection · Bemba deepest
BTS-BH100

A consent-backed clinical-language substrate: body parts + health phrases across 20+ Bantu languages, each concept with two consented native takes.

Why it matters — Clinically-scoped, consented, aligned code-switch audio is the layer monolingual corpora lack — and mistranslation in care is dangerous.

Use now

  • Clinical prove-it (/clinical-test)
  • Body-figure preview (/explore, Bemba demo, audio locked)
  • why-it-matters (6-in-1)
  • Coverage counts (audio-summary)
  • llms.txt + MCP

Licensed

  • Consented audio clips (both modes)
  • Full recordings + /body figure
  • All languages beyond the demo
A prove-it + a Bemba, audio-free body preview are public; the clips and full figure are licensed.
Open Body and Health →

BantuOS

Operating system layer
Vision shell · runtime pending

The governance/vision shell that frames the ecosystem architecture, layer stack and cartridge plan.

Why it matters — It names where the substrate is heading — the symbolic runtime that will compose the primitives.

Use now

  • 9 declarative vision/architecture pages
  • 5 read-only /api/os/* status endpoints

Licensed

Fully public — declarative only; no licensed data behind it yet.
Open BantuOS →
Diligence
Two data rooms.

Diligence data room

Public (buyers / funders)

The public diligence room: what the substrate is, the tiers, the economics, and how to evaluate — no login.

Open Diligence data room →

Engagement data room

Pilot + Founding Partner

The per-engagement room: an org-scoped, encrypted GCS bucket with the agreed deliverables, sample sets and reports.

Open Engagement data room →

Start where you are.

01 · USE NOW

Prove the gap, free

Run the Alphabet Test, the clinical prove-it, or the calculator API — no login, no key.

See the mission →
02 · EVALUATE

Get an evaluation token

An invite opens Bemba + Nyanja with a small audio cap across every subdomain — one token.

API & MCP by tier →
03 · LICENSE

Pilot, then partner

A pilot licenses an agreed set with curation; a Founding Partnership licenses the whole program.

Access ladder & pricing →

Where this is going next — the domains, languages, and pipeline we're building — is set out in a published product roadmap, available to evaluation partners and above.