Turn natural-language questions into governed, deterministic SQL over enterprise data — grounded in a business-vocabulary semantic layer instead of raw schema.
This repository is the single home for ASK. It holds two complementary bodies of work — an open standard and the product that implements it. Pick your path:
| If you want to… | Go to | What it is |
|---|---|---|
| Adopt the open standard — describe your own AI-ready data products in vendor-neutral YAML | definition/ |
The open ASK specification: Bronze / Silver / Gold layers, resolution priority, and reference examples. |
| Use the Onibex ASK Platform — install it, author a semantic layer, publish it, and query it from chat | platform/ |
The complete, screenshot-driven product manual: ASK Admin, the Configuration app, and the Chat. |
Agentic Semantic Knowledge (ASK) is a way to describe enterprise data so that AI agents can understand it, reason over it, and act on it reliably.
LLMs are good at writing SQL and chaining steps, but bad at knowing which table
answers which business question, what a cryptic code like MATNR means, or which
join path is cheapest. Without that context they hallucinate or refuse. ASK closes the
gap by formalizing the business semantics of data — entities, grains, measures,
statuses, relationships, and intent — into a layered contract any agent runtime can
consume.
This repository expresses that idea at two levels:
- The standard —
definition/— the open, runtime-neutral YAML contract. It describes what a data product means, not how it is built. Any vendor or team can adopt it. - The platform —
platform/— Onibex ASK Platform, the product that implements the standard end to end: author the semantic layer in ASK Admin, wire up databases and models in the Configuration app, publish dev → prod, and let business users query it in plain language through the Chat.
Both halves of this repository speak the same vocabulary — a medallion model in which every data product sits in one of three layers, and an agent resolves a question by preferring the most business-ready layer first:
| Layer | What it is | Agent visibility |
|---|---|---|
| Gold | A business definition, pre-joined and semantically resolved (e.g. "Open Sales Order Tracker"). | Primary — preferred first |
| Silver | A reusable enterprise artifact (Customer, Product, Sales Order), composed from Bronze. | Fallback — used when no Gold fits |
| Bronze | A raw source table, mostly uninterpreted. | Avoided — lineage only, not agent context |
The definition/ folder gives the normative rules for each
layer; the platform/ manual shows how to author them in the
product.
agentic-semantic-knowledge-ask/
├── README.md ← you are here
├── definition/ ← the open ASK specification (the standard) — AGPL-3.0
│ ├── README.md ← spec overview + quick example
│ ├── docs/ ← Bronze / Silver / Gold layer specifications
│ ├── examples/ ← reference YAML data products
│ └── LICENSE ← AGPL-3.0
└── platform/ ← Onibex ASK Platform user manual (the product)
├── README.md ← manual index (read in order / by area)
├── 01-installation.md
├── 02-concepts.md
├── ask-admin/ ← semantic-layer authoring flows
├── config/ ← technical configuration (DB, LLM, OpenSearch, SAP, MCP…)
├── chat/ ← using the chat (end users)
├── reference/ ← glossary + troubleshooting
└── images/ ← manual screenshots
- New to the concepts? Read the standard overview →
definition/README.md. - Deploying or using the product? Start with Installation, then Concepts & Architecture, then the ASK Admin flows.
The entire repository — both the open ASK specification under definition/ and the
Onibex ASK Platform manual under platform/ — is licensed under AGPL-3.0. See
LICENSE. Onibex publishes ASK so any vendor, customer, or community member
can adopt, extend, or implement it without proprietary lock-in.
ASK is initiated and maintained by Onibex, Inc. — an SAP Silver Partner and Confluent Gold Partner building real-time SAP data hyperconnectivity for the enterprise.
"Tables don't think. Schemas don't reason. ASK is what an agent reads when it needs to know what your data means."