Teams · Data
Define metrics once in Agent Serve. Agents answer questions on them, watch pipelines and flag bad data, and every figure shows where it came from.
Agent Serve, the AI workspace where teams build, deploy, and manage AI agents, gives data teams a semantic model, governed metrics and agents that answer questions, check pipelines and watch data quality, connected to Snowflake, Databricks, BigQuery, PostgreSQL and Dagster.
At a glance
The problem it removes
The same questions come in by message every week. Each answer is a new query, and two teams quoting “revenue” rarely mean the same thing. Broken pipelines are found when a dashboard looks wrong.
Agents to build
Each agent is built in Agent Serve on the warehouses and tools the data team already runs.
People ask Agent Serve Chat about revenue or churn, and the answer uses the certified metric from the semantic model, with its source.
An Agent Serve Relay reads Dagster runs and table freshness each morning, and opens an issue for anything late or failed.
An Agent Serve monitor watches row counts and key metrics, and alerts the owner when one moves outside its range.
An Agent Serve Relay refreshes the dashboards, writes the commentary for each metric that moved, and shares the pack for review.
An Agent Serve agent reads Fivetran connector status and sync history, and flags connectors that failed or fell behind.
Examples show what teams can build in Agent Serve. They describe patterns, not named customer deployments.
How it runs
Agent Serve reads the metric definition before it writes any SQL, so the answer matches the dashboard.
A manager asks Chat how net revenue moved last quarter.
Agent Serve finds the certified Net revenue metric in the semantic model.
A read-only query runs against the warehouse where the data lives.
The answer comes back as a chart with the metric and source named.
Integrations
Agent Serve connects to warehouses, databases, orchestration, product analytics and BI tools.
Browse integrationsIn each tool
Agent Serve runs SQL where the data lives and turns read-only queries into datasets that dashboards and metrics read.
Daily loan book → dataset
Agent Serve reads runs and logs, and can launch or re-run a job when a person asks.
Failed run → logs read
Agent Serve runs Hex projects and reads their run status.
Weekly project run → status
Agent Serve connects to operational databases with SSL set per connection, and reads them with read-only queries.
Open claims · hourly
Agent Serve asks Databricks Genie spaces and Snowflake Cortex questions in plain language from a workflow.
Question → Genie answer
Agent Serve reads PlanetScale databases, branches and deploy requests.
Deploy request → reviewed
Agent Serve runs DAX queries against Power BI semantic models and starts a refresh when asked.
Question → DAX query
In the workspace
Data sources turn read-only queries into datasets on a schedule.
The semantic model says what each entity and measure means.
Metrics are defined once and read the same way everywhere.
Monitors alert the owner when a metric crosses its line.
In depth
Agent Serve dashboards read governed metrics, with drill-through, calculation groups and fiscal calendars.
An Agent Serve monitor watches a metric and alerts its owner, and a Relay can start the follow-up.
Agent Serve dashboards run scenarios on the same governed metrics, so a what-if starts from the real figure.
Before and after
| Work | Before | With Agent Serve |
|---|---|---|
| Ad-hoc questions | BeforeA message, a query and a screenshot, every time. | With Agent ServeChat answers on the certified metric, with the source named. |
| Definitions | BeforeThree versions of revenue in three dashboards. | With Agent ServeOne definition in the semantic model, read everywhere. |
| Pipeline failures | BeforeFound when a dashboard looks wrong. | With Agent ServeFlagged by a morning check before people look. |
Governance
Agent Serve accepts one SELECT or WITH statement per data source query and blocks write keywords.
Row-level policies narrow what each person sees in tables, dashboards, exports and scheduled reports.
Admin, write and read roles per workspace, plus permission groups that limit which models, blocks and integrations a team can use.
Proposals, approvals, declines and executed actions are written to the workspace audit log, which admins can export.
Every workflow run records each block’s inputs, outputs and errors, so a reviewer can trace how a result was reached.
Questions
Only if you choose to. An Agent Serve dataset is either saved into a workspace table on a schedule or queried live each time someone asks.
Agent Serve data source queries are read-only. A workflow can write through an integration such as Snowflake if you build it to, behind an approval step.
Agent Serve connects Snowflake, Databricks and BigQuery, and databases such as PostgreSQL, MySQL and SQL Server, among others in the integrations catalog.
Last updated
Enterprise access
Agent Serve is available to teams by invitation. Tell us what you want to build and we will set up your workspace with you.