Agent Serve · Relays
Relays are AI coworkers in Agent Serve. Give one a role, tools and approval rules, then hand it real work.
Relays are AI coworkers in Agent Serve, the AI workspace where teams build, deploy, and manage AI agents. Each Relay has a specialization, instructions, resources and per-action permissions, and takes on tasks and ongoing work, pausing for approval before external actions such as sending an email.
How it works
Agent Serve walks you through Basics, How it works, What it can use, Ongoing work, Permissions, and Review & publish.
Start from a specialization such as Compliance Reviewer, then write the instructions and the deliverable you expect.
Give the Relay up to 30 resources: knowledge bases, tables, dashboards, metrics, the semantic model and rules.
Run on a schedule, when a monitor alerts, or when a connected inbox, calendar or table changes.
Choose Observe, Assist or Operate, set limits and name who approves.
Check the whole setup on one page, then publish. The Relay moves from draft to active.
In depth
Agent Serve shows every Relay, what it may do alone, and each action waiting on a person.
Agent Serve starts each Relay from a specialization, such as Revenue Analyst, Compliance Reviewer, Support Triage or Data Steward, or from your own role.
The Agent Serve Permissions step sets autonomy, per-action rules, task limits, approval expiry and who can approve, before a Relay starts work.
When an Agent Serve Relay needs approval, it stops and shows the exact action, such as saving a briefing to the Library, with Approve and Decline.
Agent Serve Relays run ongoing work on a schedule, When a monitor alerts, or by watching Gmail, Outlook, Google Calendar or a workspace table.
Autonomy
| Read and draft | Allowed in Observe, Assist and Operate. |
|---|---|
| Save to the Library | Observe: blocked · Assist: needs approval · Operate: allowed. |
| Hand off to another Relay | Observe: needs approval · Assist: allowed · Operate: allowed. |
| Send email or book a meeting | Observe: blocked · Assist and Operate: needs approval of the exact message, every time. |
| Approvers | Up to 20 named people, or else the owner, a steward or a workspace admin. The first decision wins. |
| Review window | 1 to 168 hours before a review expires, 24 by default. |
| Task limits | 1 to 10 tasks at once (2 by default) and 1 to 200 tasks per day (25 by default). |
| Hand-offs | Chains up to 2 deep, to as many as 5 Relays at once. |
Controls
Every Relay has a purpose, the resources it may use, the actions it may take alone, and an approver for the rest.
Pick Observe, Assist or Operate in Agent Serve, and each action follows that level: allowed, needs approval or blocked.
Read data: allowed · Send email: needs approval
Agent Serve caps how many tasks a Relay runs at once and per day, so ongoing work never runs away.
3 tasks at once · 40 per day
An Agent Serve triage Relay can pass a case to an analyst Relay, with a limit on how deep a chain can go.
Support Triage → Revenue Analyst
Give an Agent Serve Relay a browser. It saves screenshots and downloads to the Library as evidence.
Portal export → Library → Review
Agent Serve Relays send from Gmail, Outlook or the workspace inbox, to internal people only, an allowlist or anyone you allow.
Recipients: internal only
Agent Serve Relays propose meetings in Google Calendar or Outlook inside the working hours you set, and a person approves each booking.
Mon–Fri · 09:00–17:00
Examples
Three Relays financial services teams run on Agent Serve, each starting from one of the built-in specializations.
An Agent Serve Compliance Reviewer runs when a monitor alerts, reads the breach register and drafts a finding for a person to approve.
An Agent Serve Revenue Analyst reads the revenue dashboard every Monday at 8:00 and saves a briefing to the Library after approval.
An Agent Serve Support Triage Relay runs every weekday at 9:00, sorts new disputes from the tickets table, and hands complex ones to an analyst Relay.
Examples show what teams can build in Agent Serve. They describe patterns, not named customer deployments.
By the numbers
4
Built-in specializations
30
Resources a Relay can use
20
Named approvers per Relay
168 h
Longest review window
Before and after
| Work | Before | With Agent Serve |
|---|---|---|
| Recurring analysis | BeforeAn analyst rebuilds the same briefing every week. | With Agent ServeAn Agent Serve Relay drafts it on schedule and waits for review. |
| Outbound email | BeforeAn assistant tool sends whatever it wrote. | With Agent ServeA Relay shows the exact message, and only the approved one is sent. |
| Responding to alerts | BeforeAlerts sit unread until someone has time. | With Agent ServeA Relay starts work When a monitor alerts. |
| Knowing what the AI did | BeforeOutputs pasted into chat with no record behind them. | With Agent ServeA Relay activity trail plus the organization audit log. |
Governance
Agent Serve records what each Relay proposed, who decided, and what ran, in the Relay’s activity and the audit log.
Security and governanceAgent Serve sends a Relay’s approvals to up to 20 named people, or else to the owner, a steward or a workspace admin.
An Agent Serve review expires after the window you set, from 1 to 168 hours, so stale actions never run.
Relays never send email or book meetings without an approval, under any autonomy setting. An approval covers the exact payload the reviewer saw and expires if nobody decides.
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.
Works with
Start ongoing work When a monitor alerts.
Give a Relay knowledge bases to search for policies and procedures.
Read tables, or watch a workspace table for new rows.
Bind a dashboard or a metric so briefings use the governed numbers.
Bind rules so a Relay checks records the way the team does.
Save briefings and evidence to the Library, after approval in Assist.
Further reading
Disputes that fail reconciliation tend to sit in a side queue until the chargeback window has closed. Here is an end-to-end setup that finds them early, ranks them by deadline and drafts the pack, with an analyst signing every filing.
Cards and payments operations · 12 min readPeriodic reviews pile up because every file starts from scratch, even when nothing has changed. Here is a setup that refreshes the data, drafts the file and routes it by risk, so analysts spend their time on the reviews that need judgement.
Retail and business banking · 12 min readSampled call reviews find a missing risk warning weeks after the client acted on the advice. Here is a setup that checks every recorded advice call within a day and puts the quote in front of a compliance reviewer.
Wealth and advice · 12 min readQuestions
A Relay is an AI coworker in Agent Serve. It has a specialization, written instructions, the resources it may use and per-action permissions. Teams assign it one-off tasks or ongoing work.
No. An Agent Serve Relay never sends email or books a meeting without approval, under any autonomy level. It shows the exact message, and only the approved version runs.
Agent Serve runs ongoing work on a schedule, When a monitor alerts, or by watching Gmail, Outlook, Google Calendar or a workspace table.
Up to 20 people you name in Agent Serve, or else the Relay’s owner, a steward or a workspace admin. The first decision wins, and a review expires after the window you set.
Not today. You work with an Agent Serve Relay in its conversation and task list. Workflows can use speech-to-text and text-to-speech blocks for audio files.
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.