Matta. From scattered records to cited answers.
Puts your business records into one checked data model, and answers questions with sources.
What it is
One meaning across many systems.
For data and platform teams whose systems disagree about what a customer, a supplier or an order is. Matta maps records from those systems into one shared vocabulary, holds back any batch that breaks its rules for someone to fix, and answers questions with the facts it used.
A customer in the CRM, a supplier in procurement and an employee in HR each have histories and relationships that a table alone cannot explain. Matta gives those records a shared vocabulary and keeps where each statement came from.
How it works · MTTA
A record's journey. Into one meaning.
Records are mapped, checked against the rules, resolved into one entity and kept in separate layers, so every answer can show where it came from.
01 / 06MAPPED
Connect and map
Database, file and API connectors sync on a schedule. Local AI can suggest how each field maps to the shared vocabulary.
Read the figure as text
- Connect and map. Database, file and API connectors sync on a schedule. Local AI can suggest how each field maps to the shared vocabulary.
- Check against the rules. Every incoming fact is validated against the SHACL shapes before it is loaded.
- Hold what fails. Violations go to a quarantine queue. Review them, edit the payload and revalidate: nothing bad is loaded quietly.
- Resolve to one entity. The customer in the CRM and the account in finance become one entity. Each assertion keeps the system it came from.
- Keep the layers apart. Asserted facts, inferred relationships, constraints and vocabulary live in separate named graphs, so a derived claim is never mistaken for a stated one.
- Answer with the evidence. Ask retrieves graph facts and cites them. Agents use the same evidence through MCP tools and context packs.
- CRM: A customer, as the sales system knows them.
- Procurement: A supplier, as purchasing knows them.
- HR: An employee, with the history HR keeps.
- Files & APIs: Spreadsheets, documents and other APIs, on saved connectors.
- Connect & map: Saved connectors sync on a schedule with incremental cursors; local AI can suggest mappings against the live ontology.
- Validation: Incoming facts are checked against the shapes before anything is loaded.
- Quarantine: Violations wait in a queue. A person reviews them, edits the payload and revalidates.
- Resolve identity: Records about the same customer, supplier or employee resolve to one entity, each assertion keeping its source.
- Vocabulary: The shared words: what a customer, a supplier and a contract mean here.
- Constraints: What must be true of the data, each rule with a plain-English explanation.
- Asserted facts: What the source systems said, and which system said it.
- Inferred: What follows from the facts and rules, kept separate and explainable back to its sources.
- Ask with evidence: Answers come from retrieved graph facts, with citations. Other agents get the same evidence through MCP.
Try it · MTTA
Three spellings. One supplier.
- CRMMzizima Supplies Ltdcontact Asha N.
- ProcurementMZIZIMA SUPPLIES LIMITEDterms Net 30
- FinanceMzizima Suppliestax id TIN 104-332-917
One entity · sameAs
Mzizima Supplies Ltd
- contactAsha N.from CRM
- termsNet 30from Procurement
- tax idTIN 104-332-917from Finance
Three systems, three spellings
A definition no two systems agree on?
Say what it should do, which data it can touch and where it has to run. A person replies by email.