Agents guess when they don’t know your business. ekai is the semantic layer for AI agents: it captures your definitions and rules and pushes them to Cortex, Genie, BigQuery Data Agent and dbt MetricFlow, so conversational analytics returns the right answer.
Mo Aidrus, co-founder of ekai, on what ekai does and who it’s for.
Most AI teams start with Enable. The other two keep it accurate as your data changes.
New data and AI products
A new business need, turned into a trusted data product in hours
Accurate AI agents
Give AI agents the business knowledge they’re missing
Legacy to modern platform
Keep the business logic when you switch platforms
The data foundation under all three
AI readiness comes down to one thing: whether the output is accurate.
Real questions hit edge cases your definitions cover. ekai gives agents those definitions before they answer.
Every agent reads one governed semantic model, so the answer is consistent across tools and teams.
Each number comes with the definition behind it, so people can check the answer and keep using it.
Context engineering is giving a model the business knowledge it needs to answer correctly: which tables matter, how they join, what each metric means. Done by hand, it takes months and goes stale. ekai builds that context from your data warehouse and your experts, and publishes it as a semantic layer for AI agents like Cortex, Genie and BigQuery Data Agent.
Because the agent is reading raw tables and guessing at what they mean. Ask the same question twice and it can pick two different revenue columns. A verified semantic model gives conversational analytics one definition per metric, so answers are accurate and consistent, and each comes with the definition behind it.
A semantic layer is the agreed meaning between raw tables and the people asking questions: what counts as an active customer, which revenue figure is net, when a quarter closes. BI survived without one because a human read the chart and knew. An agent does not — it answers from whatever the schema implies, and a plausible guess reads exactly like a fact.
The dbt semantic layer is where metric definitions live once somebody has written them. ekai is what writes them. It profiles your data warehouse, drafts entities, keys and relationships with the evidence for each, asks your experts only what the data cannot explain, and generates standard dbt models, tests and docs. You get the semantic layer you were going to hand-author, in hours rather than months.
Yes, and keeping the output standard is the reason. Genie and Cortex Analyst both answer better when they are handed a verified model instead of raw tables. ekai builds that model in your environment and publishes it as dbt and YAML, so the same definitions feed Genie, Cortex, your BI tool, and whatever you adopt next.
Neither, quite. A catalog records that a column exists. A modeling tool gives you somewhere to write transformations. ekai does the part in between — working out what the data means, checking that meaning against the data warehouse, and emitting the model. Every inference arrives with its evidence, and a person approves it before it stands.
Standard dbt, YAML and JSON in your own repository, with the tests and documentation generated alongside them. Nothing is locked to ekai and nothing needs us to keep running. Stop using ekai tomorrow and the models still work.
No. ekai profiles data in place, inside your environment. Nothing is copied out, not the rows and not the metadata — which is usually the shortest route through a security review, because there is no new egress to assess.
