The infrastructure between African data owners and the labs that need it.
Texadata connects African data owners with AI companies that need high-quality, legally usable African data. We verify where the data comes from, who holds the rights, and how it can be used — then license it to the right buyers.
What sits behind the catalogue.
Material under management across our originating clients as of September 2026. These are portfolio figures reported by the holders, not per-asset verified counts — the catalogue below is the verified, playable subset, and every figure in it can be audited against a file. We keep the two apart deliberately.
African data is not missing because it does not exist. It is missing because nobody cleared it.
The continent's broadcast archives, production libraries and everyday environments are largely absent from training corpora. Not because the material is unavailable — because it is fragmented across thousands of small rights holders with no route to market, no rights infrastructure, and no counterparty who knows how to buy it.
Thousands of holders, no aggregator
Radio stations, production houses, ministries and independent videographers each hold small-to-mid volumes. Individually none is worth a frontier lab's procurement cycle. Aggregated and cleared, they are.
Possession is not title
The party holding the tape is frequently not the party holding the right. Commissioned work, staff output and acquired programming each carry different defects. Nobody has traced them.
Distribution licences do not cover training
Almost every existing African media licence was written for broadcast or distribution. None of them name machine learning as a permitted use, which means none of them are usable by a lab.
The coverage gap is measurable, and we measured it.
Buyers are told African data adds “diversity.” That undersells it. We ran open-weight Whisper over the speech in our own catalogue. It does not underperform — it fails outright, and the failures are reproducible from the licensed files.
Four functions, one chain of custody.
Find and onboard the holder
We work directly with African broadcasters, production houses, ministries, podcast networks and independent operators — and commission purpose-built collection where no archive exists.
Establish that title is real
Chain of title traced from creation forward. Layered rights identified. Where an asset cannot be evidenced, it does not enter the catalogue.
Execute a training-specific licence
The instrument names AI and machine learning training as a permitted use. Subject and speaker consent is collected where individuals are identifiable or audible.
Deliver to the buyer's standard
Masters in original format with per-asset metadata and rights records, to the volume, structure and annotation depth the buyer specifies.
What is available today.
The catalogue below is not the inventory — it is the proof of standard. 12 motion assets and 8 speech assets, probed, described, tagged and rights-stated, each carrying the record we produce for every asset we license.
Sample catalogue
12 motion assets (10:45) and 8 speech assets (29:04). Aerial and ground capture across Nigerian urban, waterfront and ceremonial environments. Playable previews with full technical and rights records.
Open the catalogue →Origination network
Broadcast and podcast archive, production libraries, commissioned conversational speech, and purpose-built real-world capture in environments absent from current corpora.
How we source →The full sample pack is released under NDA.
Masters, metadata sidecars and the rights record for each asset. Scope, volume and rate are set per buyer.