Semantica
The semantic layer for analytics

Ask three teams and you get four answers. Semantica is the metrics store that sits between your warehouse and everything downstream — BI tools, notebooks, spreadsheets, and AI agents all read from one governed definition. Write the logic once in code; serve a single, consistent number to the entire company.

The data teams who got tired of arguing about the number

NorthbeamCobalt DataHelio LabsGreypineTesseraDriftwellVantage AnalyticsNorthbeamCobalt DataHelio LabsGreypineTesseraDriftwellVantage Analytics
The metrics store

Stop redefining the samenumber in twelve places.

Every dashboard, query, and model that computes “active users” its own way is a chance to disagree. Semantica moves the definition out of the BI tool and into one governed layer that every consumer reads from.

Metrics as code

Define metrics, dimensions, and entities in version-controlled YAML — the aggregation, the filters, the joins, the grain. Open a pull request to change “churn,” let finance review the diff, and the new definition propagates to every tool at merge. No more tribal SQL pasted between dashboards.

One semantic graph

Entities, joins, and relationships modeled once. Semantica resolves the correct join path for any metric-by-dimension question, so nobody fans out a sum across a one-to-many and ships an inflated number to the board.

Certified definitions

Promote a metric from draft to certified once its owner signs off. Consumers see the badge and know the number is blessed — not someone's Friday-afternoon CASE statement copied into a dashboard.

Lineage to the column

Trace any metric back through every join, filter, and source column it touches. When a number moves on Monday, you see exactly which upstream table changed over the weekend.

Governed by row and role

Row-level policies and column masking live with the definition, enforced on every query path — so a metric stays private the same way whether it's asked from Looker, a notebook, or an agent.

What one definition does to the numbers

1
Definition per metric
60ms
Median metric query
0
Conflicting versions downstream
9
Tools reading one layer
The Metrics API

Headless by design.Query it from anywhere.

Semantica doesn't own a dashboard. It exposes your metrics through open interfaces so the tools your team already runs — and the agents they're starting to trust — all ask the same layer the same way.

GraphQL + SQL endpoints

Request a metric by name, sliced by any dimension, over any time grain — in GraphQL or plain SQL. Semantica compiles it to optimized warehouse SQL and returns the number, not raw rows for you to re-aggregate.

BI tools, connected

Native connectors and a JDBC interface for Looker, Tableau, Hex, Mode, and Power BI. They read certified metrics instead of each rebuilding the same logic in a proprietary modeling layer.

Trusted by your AI agents

Point your LLM at the Metrics API, not the raw schema. Agents answer “what was Q3 net revenue in EMEA” from the governed definition — grounded in certified logic, never inventing a join that doesn't exist.

dbt-native, warehouse-pushdown

Import your dbt models and tests, compile metrics down to Snowflake, BigQuery, Databricks, or Redshift, and run the work where the data already sits. No extra cache layer to keep warm and reconcile.

Where the layer earns its keep

One number, every surfaceit shows up on.

From the board deck to the in-product chart to the question an agent asks at midnight, the same definition answers — so the company stops arguing about whose dashboard is right.

Finance

Executive reporting

The revenue on the board slide and the revenue in the ops dashboard come from the same certified spec. Quarter-end reconciliation stops being a debate about methodology.

Data team

Self-serve analytics

Analysts pick a metric and a dimension; the join path is handled. They stop writing the fortieth bespoke version of “monthly active accounts.”

Product

Embedded dashboards

Serve tenant-scoped metrics inside your app with row-level policies enforced in the layer — the same numbers, isolated per customer, no per-tenant SQL to maintain.

GenAI

AI & agent answers

Give copilots and agents a metrics endpoint instead of table access, so “why did NRR drop” returns governed math with its lineage, not a confident guess.

GTM

Reverse ETL & ops

Sync the certified metric — not a snowflake of CRM formulas — into Salesforce, HubSpot, and Slack so go-to-market acts on the same truth as finance.

Growth

Experimentation

Wire experiment readouts to the canonical metric so a winning test and the quarterly target are measured by the exact same definition, not two that almost match.

From the heads of data

The dashboards finally agree.

We had four definitions of “active customer” living in four BI tools, and every leadership meeting opened with a fight about whose chart was right. We moved the logic into Semantica, certified one version, and the argument just ended. The number is the number now.

P
Priya Nandakumar
VP of Data, Northbeam

Our analysts used to spend half their week reverse-engineering someone else's SQL to reproduce a metric. With definitions in Git and lineage to the column, onboarding a new analyst dropped from weeks to days. They trust the layer instead of auditing it.

F
Felix Marchetti
Head of Analytics Engineering, Greypine

The day we pointed our internal copilot at the Metrics API instead of the warehouse, the hallucinated revenue numbers stopped. Execs ask the bot a question and get the certified answer back. That one change is what finally made AI usable for our finance team.

A
Adaeze Okonkwo
Director of Data Platform, Tessera
Pricing

Priced per metric query, not per seat.

Give the whole company — and its agents — the certified numbers. You pay for the queries served, never for who's allowed to ask.

Developer

For a single team modeling its first metrics.

$0/mo
  • Up to 1M metric queries/mo
  • Metrics-as-code in Git
  • GraphQL + SQL endpoints
  • 1 warehouse connection
  • Community Slack support
Most popular

Team

For data teams making the layer the source of truth.

$0.0002/query
  • Pay-per-query, volume tiers
  • Certified metrics + approvals
  • Full column-level lineage
  • BI connectors + AI Metrics API
  • Row/role governance
  • 99.9% SLA + priority support

Enterprise

For multi-warehouse estates and regulated data.

Custom
  • Unlimited metric queries
  • VPC-isolated or self-hosted plane
  • SSO, SCIM & query audit logs
  • Data residency (US · EU · CA)
  • Multi-warehouse federation
  • Named solutions architect

Questions, answered.

Is a semantic layer just a fancy view layer?

No. A view freezes one query shape; a semantic layer holds the definition and composes it on demand. You ask for a metric by any dimension and time grain, and Semantica resolves the right joins, applies the governed filters, and compiles optimized SQL for your warehouse. One definition answers thousands of differently-sliced questions — you don't pre-build a view for each.

Do I have to move my data into Semantica?

No. Semantica is a query-time layer, not a warehouse. It compiles metric requests down to SQL and pushes them to Snowflake, BigQuery, Databricks, or Redshift, where your data already lives. There's no ingestion step and no second copy — only the definitions live in Semantica.

How does this work with dbt?

It builds on the modeling you've already done. Import your dbt models, sources, and tests; layer metric definitions on top; and let lineage flow end to end. Your transformation pipeline stays in dbt — Semantica governs and serves the metrics that sit above it.

Will my BI tools and notebooks still work?

Yes. Semantica exposes GraphQL and SQL endpoints plus a JDBC interface, with native connectors for Looker, Tableau, Hex, Mode, and Power BI. Tools read certified metrics by name instead of each rebuilding the logic, so every chart agrees by construction.

How does this make AI answers trustworthy?

Point your LLMs and agents at the Metrics API instead of raw table access. When an agent answers a metric question, it pulls the certified definition and gets governed math back — not a guessed join across columns it doesn't understand. The semantic layer is the grounding layer for analytics AI.

Who decides what “certified” means?

You do. Each metric has an owner; changes go through pull-request review just like code. When the owner signs off, the metric is certified and carries a badge everywhere it's served. Drafts stay visible but clearly uncertified, so consumers always know whether a number is blessed.

Settle the number once.

Define your first metric in YAML, certify it, and watch every dashboard, query, and agent line up behind the same answer. No data migration, no rip-and-replace — just one definition the whole company can finally agree on.