Use case · Regulatory reporting
Agent Serve checks report data against your rules, keeps one definition per figure, and drafts regulator answers from your policies for an owner to sign.
A regulatory reporting use case for Agent Serve, the AI workspace where teams build, deploy, and manage AI agents: the semantic model keeps one definition per figure, rules check report data before submission, and a workflow drafts answers to regulator requests from a knowledge base, checks them against sources and waits for the question owner to sign off.
At a glance
The problem it removes
Each return is rebuilt from spreadsheets that pull from several systems, and nobody is sure every team uses the same definition. When a regulator sends a letter with twelve questions and a deadline, someone splits it up by email, chases owners, and searches old submissions for the evidence.
How it runs
Each question becomes a row with an owner. Agent Serve drafts the answer and the owner decides what is sent.
Each question from the letter becomes a row in the request tracker, with its owner and due date. A new row starts the workflow.
The Knowledge block searches policies, procedures and past submissions for the passages that answer the question.
An agent drafts the answer, and a Guardrails Hallucination Check tests it against the knowledge base before it moves on.
A Human step sends the owner the draft and its sources. The owner edits, approves or sends it back.
In depth
Agent Serve keeps the definition of each reported figure in one place and checks the data before anyone signs.
The Agent Serve semantic model maps each entity to a table and defines measures once, so every dashboard and metric sums the same column the same way.
Agent Serve rules flag rows that fail a check, such as a missing counterparty type or a value outside a range, and a dashboard table lists them.
Agent Serve row policies decide which rows each person sees, and a dashboard can be previewed as another role before it is shared.
The Agent Serve Activity audit lists workspace activity and exports it as CSV, and every workflow run keeps a block-by-block trace.
Before and after
| Work | Before | With Agent Serve |
|---|---|---|
| Definitions | BeforeEach team’s spreadsheet calculates the figure its own way. | With Agent ServeThe semantic model defines each measure once for every view. |
| Data checks | BeforeValidation errors surface after submission, in a resubmission. | With Agent ServeRules flag failing rows on a dashboard before anyone signs. |
| Regulator questions | BeforeA letter is split up by email and chased in reply threads. | With Agent ServeEach question is a tracked row with an owner and a due date. |
| Evidence | BeforeOld submissions are searched by hand for the right paragraph. | With Agent ServeDrafts cite the passages they rely on, from your own documents. |
What does the work
Entities, relationships and measures describe report data once.
Rules flag rows that fail a check before the return is signed.
Policies, procedures and past submissions sit in searchable knowledge bases.
A workflow drafts each answer, checks it against sources and waits for the owner.
A dashboard lists open questions and failed checks by owner.
Examples
Three reporting jobs financial services teams run on Agent Serve, each ending with a named person’s sign-off.
Agent Serve rules flag rows that fail validation, and the reporting lead reviews the list on a dashboard before sign-off.
Each question becomes a row in an Agent Serve table, a workflow drafts the answer from the knowledge base, and the owner signs off.
The Agent Serve Data model diagram shows the tables and joins behind a reported measure, for a reviewer to follow.
Examples show what teams can build in Agent Serve. They describe patterns, not named customer deployments.
Governance
Agent Serve keeps sign-off with named people, limits who sees report data and records what changed.
Security and governanceThe Agent Serve Hallucination Check tests a drafted answer against the knowledge base and flags claims the sources do not support.
Saving the Agent Serve semantic model needs write access, and each update is written to the workspace audit log.
Row-level policies narrow what each person sees in tables, dashboards, exports and scheduled reports.
Workflows run from a deployed version, so a change is reviewed before it reaches production runs.
Proposals, approvals, declines and executed actions are written to the workspace audit log, which admins can export.
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.
Zero data retention with supported model providers, backed by your own provider keys and agreements or self-hosted models. Some provider features keep data by design, such as Gemini Deep Research, so they are outside zero data retention.
Every workflow run records each block’s inputs, outputs and errors, so a reviewer can trace how a result was reached.
Further reading
The deep dive models the cost of a regulator request and what changes, with every figure labelled.
Read the blogA regulator’s letter becomes a dozen questions, a dozen owners and one deadline. Here is a setup where a workflow logs and splits the request, a Relay drafts from approved sources and a named person signs off every answer.
Regulatory affairs · 13 min readRegulatory returns are signed by people, but most of the weeks before sign-off go on checking, tying out and explaining. Here is how to choose which of those steps to give to workflows first.
Regulatory reporting · 5 min readSupervisors keep asking banks to show where a risk number came from. These field notes answer the questions risk and finance teams raise when they try to keep that evidence current, and where Agent Serve fits.
Risk and finance · 6 min readQuestions
No. Agent Serve checks data, drafts answers and tracks questions. Your team signs off and submits through its usual channel.
The Agent Serve semantic model defines each measure once over workspace tables, and dashboards, metrics and monitors all read that definition.
Agent Serve data sources sync data from your systems into workspace tables, and workflows can also read databases such as PostgreSQL or Snowflake directly.
The Agent Serve Hallucination Check tests each draft against the knowledge base before it moves on, and a named owner reviews every answer.
A team can start with an Agent Serve request tracker, a knowledge base of policies and submissions, and the drafting workflow, then add the semantic model and validation rules.
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.