Talk to your data
Ask your business a question. Get an answer you can check.
A chatbot that guesses at your database is a liability. We model what your business means first, then let an agent answer from that model.
Your business, as a map an agent can read
Every point is something your business knows. Every line is how two things relate. Agents follow the lines before they answer.
- 01Ontology
- A map of the things in your business (customers, orders, depots) and how they connect. It is how the agent knows what you mean.
- 02Semantic layer
- One agreed definition of every number, written in the open Apache Ossie standard, so reports, dashboards and agents all agree.
- 03Talk-to-data agents
- Ask in plain language. The agent picks the metric, writes the query, runs it, and shows its work.
- 04RAG for documents
- For answers that live in documents, not tables: it finds the right passages and quotes its sources.
The standard we use
Your semantic model, in a format every tool can read.
We write semantic models in Apache Ossie (formerly Open Semantic Interchange, OSI): an open, vendor-neutral specification, incubating at the Apache Software Foundation, for describing what your data means.
One definition per number
Revenue, late deliveries, active customers: each is defined once, as a metric with its formula. Every dashboard and every agent reads the same file, so they give the same answer.
Plain files you own
An Ossie model is YAML or JSON. It lives in your own repository and is reviewed like code. It is not locked inside one vendor's tool, and it moves with you if you change tools.
Written for AI, not only for BI
A model and its metrics can carry AI context: instructions, synonyms and example questions. That is how an agent knows that “late orders” means the late_deliveries metric.
Meaning first, then numbers
Ossie also has an ontology specification: the concepts in your business and how they relate. That is the map at the top of this page, and the agent reads it before it writes a query.
What is inside an Ossie model
- datasets
- Your tables, and the fields in them that matter
- relationships
- How those tables join to each other
- metrics
- The numbers you report, each with one formula
- ai_context
- Hints for agents: instructions, synonyms, examples
The specification is public: github.com/apache/ossie. A real model is in the example below, under semantic-model.yaml.
See it work
Pick a question. Watch the agent answer.
An invented delivery company, with its ontology and semantic model written in Ossie. The tabs show the two files the agent reads.
Try a question
What the agent did
- Matched “late deliveries” to the metric late_deliveries
- Matched “depots” to the entity Depot
- Wrote and ran the query · 3 rows
query.sql
-- metric: late_deliveries (semantic-model.yaml)SELECT d.name AS depot, COUNT(*) AS late_deliveriesFROM orders oJOIN depots d ON d.depot_id = o.depot_idWHERE o.delivered_at > o.promised_at AND o.delivered_at >= date_trunc('month', now()) - INTERVAL '1 month' AND o.delivered_at < date_trunc('month', now())GROUP BY d.nameORDER BY late_deliveries DESCLIMIT 3;Answer
Germiston had the most late deliveries last month, followed by Midrand and Rosslyn.
| Depot | Late deliveries |
|---|---|
| Germiston | 41 |
| Midrand | 28 |
| Rosslyn | 17 |
Source: Semantic layer · late_deliveries
Have something in mind? Let us help you.
Tell us what keeps going wrong. We reply within one working day, with questions or a rough plan.