How is Strata priced?

There is a Free plan (1 developer, 25 users, 2 data sources per project) and a Team plan (3 developers, 50 users, unlimited data sources). Team and Enterprise require scheduling a demo; Enterprise adds airgapped deployment, analytics opt-out, and custom support.

How do I get started?

You can try it free on your own machine — setup is claimed at 15 minutes using Docker and your own coding agent — or schedule a demo to see it on your own data.

Is Strata open source?

The site does not state that Strata is open source. It emphasizes an open-OLAP architecture (bring any engine, no lock-in) and offers a self-hosted free plan, but the product itself is governed by a Beta License.

Where do users consume the data?

In Strata's dashboards, in Google Sheets via Strata's own add-on (with in-place refresh), or via scheduled/subscription delivery. More delivery surfaces are described as on the way.

Strata — AI-safe self-service analytics for humans and agents

Strata is a business intelligence platform that combines a governed semantic layer with self-service dashboards, aiming to serve both non-technical human users and AI agents. It models data via a YAML-based naming convention, refuses semantically invalid queries (preventing grain-crossing double counting), and federates queries across hot/warm/cold OLAP tiers (ClickHouse, Druid, Snowflake, AWS Athena) for sub-second answers. It describes itself as born at Netflix.

Key Features

Semantic layer as code

The entire model is defined by two YAML file types plus a naming convention. A coding agent can draft it, 'strata audit' validates it against the warehouse, and Git carries it to production.

Dynamic universes

Users and agents can query the whole model without picking a predefined view, explore, or cube; each build step is validated and previewed, and semantically invalid queries are refused before they run.

Grain-safe cross-domain blending

Facts are aggregated to a common grain and joined on conformed keys (never fact to fact), so blends cannot double count and agents receive a refusal instead of a wrong number.

Federated smart OLAP routing

Every query routes to the fastest engine that can answer it — hot tiers (ClickHouse, Druid), warm (Snowflake), cold (AWS Athena) — with fallback to the warehouse when filters fall outside the hot tier, delivering sub-second answers and lower warehouse load.

Beautiful-by-default dashboards

Drop views in the order they should be read and the layout engine sizes, places, and colors them automatically — no grid editing.

Five measure types plus segments

Standard, Complex, Snapshot, and LOD (inclusion and exclusion) measures computed at the right grain, plus segments that put a cohort beside its baseline in one view.

Agent building in the report editor

Agents are built in the same editor as reports, with live measures and dimensions in their instructions; they inherit row-level security and expose a validation-loop API (MCP/API) for grounded analytics.

Row-level security and masking in the model

Masking and auto-filtering live in the semantic layer, so a single report is safe for every audience.

How It Works

1. Model by naming

Define the semantic layer with two YAML file types and a naming convention; your coding agent drafts it and 'strata audit' validates it against the warehouse before Git promotes it.

2. Connect engines

Bring your own OLAP engines — hot tier (ClickHouse, Druid), warm (Snowflake), cold (AWS Athena) — and Strata routes queries to the fastest engine that can answer them.

3. Ask and blend

Humans and agents (via MCP/API) query the whole model; facts are blended at a common grain on conformed keys, and invalid queries are refused rather than returning wrong numbers.

4. Consume

Get answers on auto-layout dashboards, in Google Sheets via the add-on, or via scheduled/subscription delivery.

Pros & Cons

Pros

Grain-safe blending prevents double counting by construction, with explicit refusals instead of wrong numbers
Federated routing across hot/warm/cold tiers delivers sub-second answers and reduces warehouse spend
Works with any OLAP engine — swappable, with no lock-in
Semantic layer defined by naming convention and two YAML file types, validated by a tool and carried by Git
AI agents inherit row-level security and are validated through a governed loop, not bolt-on chat
Auto-layout dashboards require no grid/visual design work

Who It's For

Best for

Data and engineering teams drowning in ad-hoc reporting requests from non-technical business users
Organizations wanting to expose data safely to AI agents with grain-safe, governed answers
Companies with existing warehouses (e.g., Snowflake) that want sub-second performance via hot-tier OLAP engines
Analytics engineering teams who want a Git-based, code-defined semantic layer with minimal configuration

Not ideal for

SQL-fluent analysts who just want another advanced query tool
Teams needing delivery to BI surfaces beyond Google Sheets today

Pricing

Three tiers: a Free plan for evaluating and small projects, and Team and Enterprise plans that start with a demo on your own data (Team pricing is not publicly listed).

Free

Evaluate Strata and ship small projects. 1 developer, 25 users, 2 data sources per project, unlimited projects and branches. Email support (3-day response), Docs, Community.

FAQ

What does Strata do?

It provides high-performance, AI-safe self-service analytics: a governed semantic layer fused with auto-layout dashboards, serving both human users and AI agents, with federated query routing across multiple OLAP engines.

Who is Strata for?

Teams that want to put deep data access in non-technical hands, open data to AI agents safely, free engineers from the ticket queue, and get trustworthy, sub-second answers on the warehouse they already have. It explicitly targets people who could never write the query themselves, not SQL-fluent analysts.

How does Strata prevent wrong numbers?

Queries are validated against the semantic model. Invalid, grain-crossing combinations are refused before they run, and cross-domain blends aggregate each fact to a common grain joined on conformed keys, so double counting is impossible by construction.

Which data engines does Strata support?

Strata