Agent Serve · Semantic model
Describe your entities, how they relate, and what “revenue” means. Dashboards, metrics and Relays then join and measure data the same way.
The semantic model in Agent Serve, the AI workspace where teams build, deploy, and manage AI agents, describes business entities with dimensions, attributes, measures and synonyms, the relationships between them, and conformed dimensions shared across sources.
How it works
Agent Serve builds the semantic model over your workspace tables, then every dashboard and metric reads the same definitions.
Choose New entity, point it at a table, and add attributes, dimensions and synonyms.
Add a relationship with its cardinality, then Check relationship to see how well the keys match.
Add measures as an aggregate on a column, a governed metric, or a formula built from other measures.
Save model, and dashboards and metrics use the same joins and measures.
In depth
Agent Serve checks each relationship against the data, lines up values from different sources, and lets measures build on each other.
Check relationship in Agent Serve reports the match rate, duplicate keys and orphan keys, and what the data says the cardinality really is.
An Agent Serve conformed dimension gives tables from different systems one shared field, with value maps that turn “EMEA” into “Europe”.
A formula measure in Agent Serve combines other measures with MEASURE("…") and shows as a number, currency or percent.
The editor
Each Agent Serve entity maps to one workspace table and carries its attributes, dimensions, measures and relationships.
Policy → policies table
Agent Serve synonyms record the words your teams use, so “client” and “customer” point to the same entity.
Customer · client, account holder
Agent Serve suggests relationships between entities with a score, and one click on Add puts them in the model.
Loans.customer_id → Customers.id
An Agent Serve measure is an aggregate on a column, a governed metric, or a formula over other measures.
sum · avg · min · max · count · count_distinct
The Agent Serve Data model diagram draws every entity and relationship, so reviewers see how the data fits together.
Accounts 1:N Transactions
An Agent Serve relationship can join on up to four key pairs when one column is not enough.
branch_id + account_no
Examples
Three ways financial services teams use the Agent Serve semantic model to fix a number once.
Agent Serve maps accounts, customers and branches to their tables, so every deposits dashboard sums the same balance column the same way.
An Agent Serve formula measure divides claims paid by earned premium. A formula metric reuses it, and a Metric threshold monitor flags drift.
A conformed region in Agent Serve lines up the processing and CRM tables, with a value map so both use the same region names.
Examples show what teams can build in Agent Serve. They describe patterns, not named customer deployments.
Under the hood
| Scope | One semantic model per team space, with up to 200 entities. |
|---|---|
| Data | Workspace tables. Data sources reach the model once they sync into a table. |
| Relationships | 1:1, 1:N, N:1 and N:N, with up to 4 key pairs and three key-matching modes. |
| Relationship check | Match rate, duplicate keys, orphan keys and the cardinality the data suggests. |
| Measures | Column aggregates (sum, avg, min, max, count, count_distinct), governed metrics and formulas. |
| Formulas | Up to 4,000 characters, formatted as Number, Currency (USD) or Percent. |
| Conformed dimensions | Up to 50, Text or Date, with 20 members each and 500 value-map entries. |
| Used by | Dashboards, metrics, monitors and Relays. |
Before and after
| Work | Before | With Agent Serve |
|---|---|---|
| What revenue means | BeforeEach dashboard carries its own formula, and they disagree. | With Agent ServeOne measure in Agent Serve, reused by every dashboard and metric. |
| Joining tables | BeforeJoins rebuilt by hand in every report, with silent duplicates. | With Agent ServeRelationships checked for match rate, duplicate and orphan keys. |
| Regions from two systems | BeforeManual mapping in spreadsheets before every board pack. | With Agent ServeA conformed dimension with a value map, applied every time. |
| Business words | BeforeAnalysts translate “client” into the right table name. | With Agent ServeSynonyms on entities and measures record the words teams use. |
Governance
Anyone in the team space can read the Agent Serve semantic model; only people with write access can change it.
Security and governanceSaving the Agent Serve semantic model and verifying relationships needs write or admin access to the team space.
Each update to the Agent Serve semantic model writes an event to the workspace audit log.
Admin, write and read roles per workspace, plus permission groups that limit which models, blocks and integrations a team can use.
Sign in through your identity provider with SAML 2.0 or OIDC, and provision people with SCIM.
Run Agent Serve in your own cloud or data centre, next to the systems it reads.
Works with
Each entity maps to a workspace table, and the model joins tables by its relationships.
Dashboards join by relationship, blend across conformed dimensions and use MEASURE() in formulas.
Formula metrics resolve MEASURE(), so a monitor watches the same definition a dashboard shows.
A Relay can use the semantic model as a resource when it works with your data.
Further reading
When finance, risk and the business each compute the same ratio their own way, the board pack argues with itself. Here is a setup where every dashboard reads one governed definition, and drift is caught before the next meeting.
Risk and finance data · 13 min readThe morning spreadsheet is out of date before the meeting starts, and two people bring two RTO numbers. This playbook moves daily decisions onto dashboards that refresh every 15 minutes or query live, with one definition per number.
Ecommerce leadership · 5 min readMost D2C teams still start the day by stitching a Shopify export, a courier report and two ad accounts together in Excel. Here is a dashboard that fills itself as orders land and tells the team when something moves.
Fashion & D2C · 6 min readQuestions
A semantic model is a shared description of business data: what an entity represents, how entities relate, and which dimensions and measures are meaningful. In Agent Serve it sits over your workspace tables.
The Agent Serve semantic model maps each entity to a workspace table. Data from connected sources reaches the model once it syncs into a table.
Check relationship in Agent Serve compares the keys in both tables and reports the match rate, duplicate keys, orphan keys and the cardinality the data suggests.
Yes. A formula measure in Agent Serve combines other measures with MEASURE("…"), such as claims paid divided by earned premium, and formats the result.
Anyone in an Agent Serve team space can read the model. Saving it and verifying relationships needs write or admin access, and each update is audited.
One. Each Agent Serve team space has a single semantic model, with up to 200 entities, so every dashboard and metric reads the same definitions.
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.