Enterprise semantic layer

Your semantic layer, without the YAML and the build queue.

Data Convo replaces hand-maintained dbt and YAML semantic models with a visual workspace you run yourself. Model joins, metrics and business definitions on an ERD canvas, then let three studios answer questions, watch for drift, and render the visuals your team actually asks about.

Self-hosted via Docker · Bring your own LLM key · Table and column level access control

What a semantic layer is supposed to do

A semantic layer is the governed definition of your business that sits between the warehouse and everyone querying it. The hard part was never the concept. It was keeping the definitions current without a codebase to babysit.

Manual dbt / YAML

  • Every metric lives in .yml files across a repository
  • A change needs a branch, a review, and a CI run before it is live
  • Joins are declared in code, disconnected from the schema they describe
  • Access rules are bolted on afterwards, in a different system
  • Nobody outside the data team can read or verify the model

Data Convo

  • Definitions live in one JSON model attached to the live connection
  • Edit in the browser and publish: no branch, no build, no deploy
  • Joins are drawn on an ERD canvas against real PK/FK constraints
  • Table and column permissions are enforced before SQL is generated
  • Business definitions are written in plain language your team can audit

Three studios, one governed model

The semantic layer is the foundation. Each studio is a different way to use it, and they run against the same definitions so answers stay consistent across all three.

01

Semantic Studio

Introspect your warehouse, then model it visually. Drag columns to link tables on an ERD canvas, and define aliases, synonyms, enums, custom metrics and business definitions in plain language.

  • ERD join modeling with FK constraints
  • Custom metrics and named business definitions
  • Column aliases, enums and synonym maps
  • BYOK: your OpenAI, Anthropic or OpenRouter key
  • Table and column level security per member
The model every studio reads from
02

Anomaly Studio

The semantic layer tells you what the data means. Anomaly Studio tells you when it stops meaning that. Baseline each monitored column, then get alerted the moment the shape of the data changes.

  • Per-column baselines: distribution, volume, type
  • Deviation flagged at roughly three sigma
  • Null spikes and volume shifts detected automatically
  • Alerts routed to email or a Slack webhook
  • Status and history on the dashboard
Catches broken pipelines before the dashboard lies
03

Interactive Dashboarding Studio

Ask a question, get a chart, then keep talking to it. Results render as charts and KPI cards, and follow-up questions run against the same dataset, so refining a view never means going back to SQL.

  • Natural-language questions over the semantic model
  • Charts and KPI cards generated from the result set
  • Chat directly about the visual on screen
  • Pivot, filter and drill without writing SQL
  • Self-healing SQL that corrects on error
Chat with the chart, not just the database

From warehouse to answered question

Four steps, no repository and no YAML file.

  1. Connect

    Point Data Convo at Snowflake, Databricks, BigQuery, Redshift, Postgres, SQL Server, MySQL or SQLite.

  2. Profile

    Tables, columns, primary keys and foreign keys are introspected into the ERD canvas automatically.

  3. Define

    Draw joins, then add aliases, enums, metrics and business definitions in business language.

  4. Publish

    The model is persisted and injected into every prompt. Ask a question, get a chart, keep chatting.

Frequently asked

The questions buyers actually ask before replacing a dbt or YAML semantic layer.

What is an enterprise semantic layer?

A semantic layer is a governed model of your business that sits between your warehouse and the tools that query it. It defines what each table, column and metric means in business language, so every answer is computed from the same definitions rather than re-derived by each analyst or language model.

How is this different from dbt or a YAML-based semantic layer?

dbt and similar tools describe the semantic model in code: YAML files, a repository, a CI pipeline and a build step. Data Convo models the same concepts visually on an entity-relationship canvas and stores them as JSON attached to your live connection. There is no repository to maintain and no build to wait for: you connect a warehouse, define joins, metrics and definitions in the browser, and publish.

Does Data Convo send our warehouse data to a third party?

No. Data Convo is deployed self-hosted via Docker inside your own environment and connects to your warehouse directly. You bring your own language model key (BYOK) and choose the provider, so the only outbound traffic is the schema context and the question you asked, sent to the provider you selected.

How does table-level security work for teams?

Administrators grant access per table and per column for member accounts. The permission list is applied before any query is generated, so a restricted table is never included in the prompt or the generated SQL for that member. Role and column restrictions are enforced server-side, not hidden in the interface.

What happens when the underlying data changes unexpectedly?

Anomaly Studio baselines each monitored column and flags deviations at roughly three standard deviations: null spikes, sudden volume changes and type mismatches. Alerts route to email or a Slack webhook as soon as a metric drifts, so a broken pipeline surfaces before it reaches a dashboard.

Can I ask questions about a chart I am already looking at?

Yes. The Interactive Dashboarding Studio renders results as charts and KPI cards, and you can keep chatting about the visual that is on screen. Follow-up questions run against the same dataset, so you can pivot, filter or drill in without writing SQL.

See it against your own schema

Connect a warehouse, model a few tables, and ask a real question. Most teams get their first governed answer before the demo call ends.