What is SEMANTIX
SEMANTIX is a context layer. It sits between your data sources and the people (or agents) asking questions about them, and its job is to make sure every answer is anchored in context your organisation actually agreed on.
The distinction that matters: a text-to-SQL generator will happily produce a syntactically perfect query against the wrong table, at the wrong grain, using a filter that was never agreed. Generating SQL is easy. Producing a number someone will act on is the hard part.
The five parts
Section titled “The five parts”1. Data sources (connectors)
Section titled “1. Data sources (connectors)”Your databases and SaaS tools, connected as MCP servers. Databases are attached through SEMANTIX’s own read-only SQL server, so nothing can write to them. Any service that speaks the Model Context Protocol can be attached too.
See connectors and data source templates.
2. The Registry
Section titled “2. The Registry”The catalogue of what SEMANTIX found. Two kinds of thing live here:
- Entities — a registered table, its columns, and what they mean. Each entity has a description, lineage back to its source, and a trust level from 0–5.
- Metrics — a defined, reusable calculation with a label, a description, its query, a run history, and a certification state.
An entity that exists but has no description is a table name. An entity with a description, lineage, and a trust level is something an AI can answer from.
See Registry.
3. Context scopes
Section titled “3. Context scopes”A scope groups the entities and metrics belonging to one team, domain, or workstream, and carries the business rules and findings for that area. It is the unit of organisation and the unit of access.
See context scopes and creating scopes.
4. Chat
Section titled “4. Chat”Where people ask questions in plain language. In a scoped conversation, the agent works from that scope’s governed context — the registered entities and their columns, the certified metrics, and any business rules attached to the scope. It can look up the distinct values of a column when it needs to resolve a filter, and every query it runs is validated as read-only before execution.
See Chat.
5. Self-watching metrics
Section titled “5. Self-watching metrics”Certified metrics run on a schedule and watch themselves. Anomalies are detected automatically and broken down by dimension, so when a number moves you can see which slice moved — the difference between a real demand shift and a stale pipeline sync.
Vocabulary
Section titled “Vocabulary”These terms are used consistently across the docs.
| Term | Meaning |
|---|---|
| Context layer | The governed context every answer is anchored in |
| Grounding | Anchoring an answer in that context instead of generating from raw schema |
| Entity | A registered table with its columns, description, lineage, and trust level |
| Field | A column of an entity, with type, description, and role (dimension, measure, time) |
| Certified metric | A metric cleared for decision-making by an admin |
| System trust score | The machine-computed confidence on a metric, shown as a percentage |
| Context scope | A governed grouping of entities, metrics, rules, and findings for one team or domain |
| Primary scope | The scope a person works in by default; sets their Home view and Chat context |
| Living business knowledge | Business rules and investigation findings attached to a scope, curated and certified by its owners |
| Introspection | Scanning a data source to discover its tables and columns and register them |
| Read-only | Enforced at the SQL layer: only SELECT/WITH/EXPLAIN/SHOW/TABLE/VALUES, row-limited |
| Open mode | A non-governed conversation with custom agents and arbitrary MCP servers |
| Open mode vs governed | Governed chats stay inside a scope’s context; open-mode chats do not |
How the pieces fit together
Section titled “How the pieces fit together”- An admin connects a data source and introspects it.
- Introspection populates the Registry with entities and fields.
- An admin sets trust and certifies the metrics worth trusting.
- Entities and metrics are assigned to scopes, along with business rules and findings.
- People are granted a scope, with a primary one.
- From then on, their Chat and their board are grounded in that scope.
If a step is missing, the next one produces an empty result — which is why the quickstart orders things the way it does.