Example: a Workspace Manager Teammate

This example sets up a Teammate that looks after one workspace. On each check-in it:

• reports workflows that failed or paused themselves;
• reports other Teammates in the workspace that are failing or blocked;
• warns the team owner before the plan's storage, record or seat limits are reached;
• runs your own checks, for example "approval records waiting for more than 7 days", and alerts when they go wrong.

Your own checks are watches. AnyDB, not the AI, measures each watch and decides whether it alerts, so the same check gives the same answer on every run and with every model.

What you need first

• A workspace the Teammate will work in.
• For each check you want, a saved report in that workspace that counts what should not be there. For example, an Approvals report filtered to Status:Waiting.
• To warn about plan limits, a team admin must allow it (see step 3).

1. Create it

Create a Teammate (AI Teammates → New) with:

• Name: Workspace Manager

• Workspaces: the workspace it looks after

• Role: Keeps the Operations workspace healthy and tells the owner what needs a decision.

• Instructions, for example:

  On each check-in:

  1. List the workspace's workflow failures since your last check-in. Report each failing workflow once, with its failing step and error, and the workflows that paused themselves.
  2. Read the Teammates' health for the workspace. Report Teammates that are failing or blocked, with the reason.
  3. Read the team's usage. Warn the owner when any limit is at 85% or more.
  4. Report your watches as AnyDB measured them.

  Change nothing: report, and name the decision someone needs to make.

2. Schedule it

On the Schedule & budget tab, give it a check-in, for example every 24 hours, with the instruction "Check the workspace's health."

3. Set its access

• It reads workflow failures for the workspaces it works in.
• To read the team's usage against the plan's limits, and other Teammates' health, a team admin opens its Access tab and turns on Can read audit and security data. Usage and limits are otherwise shown only to team admins, in Manage Team → Overview.
• It changes nothing, so it needs no permission to change records.

4. Add your checks (watches)

A watch is one thing the Teammate keeps an eye on. Each watch has:

Part	What it is	Example
Name	What it watches	Old approvals
Source	A saved report in its workspace, and which number of its result to use: how many records it matches, or one metric's total	The "Waiting approvals" report, its record count
Condition	When to alert: one of >, >=, <, <=, ==, !=, and a number	> 0
Severity	Urgent tells you at once; digest waits for its next summary	Urgent

To add one, ask the Teammate, for example: "Watch the Waiting approvals report and alert me when it has any records." It proposes the watch, and nothing changes until you approve it. The approval shows exactly which report AnyDB will run and the condition it will check.

On every check-in, AnyDB runs each watch's report as the Teammate, with its own access, then compares the result with the condition and decides:

• it alerts once when the condition is met;
• it alerts again only if the value gets worse, or if it clears and comes back;
• a report the Teammate is not allowed to run is reported as an error, never as a value.

The Teammate then tells you what was found and what you need to decide.

tip

To watch records that have been waiting too long, filter the report on a date, for example approvals still waiting whose created date is more than 7 days ago. Then watch its record count with the condition > 0.

Reviewing alerts

On the Teammate's Memory tab, each watch shows its recent values and the alerts it raised. Mark an alert useful or noise: the Teammate keeps your rating as a lesson, so its later alerts follow it.

Watches added before this release were described in words, and the Teammate judged them itself. They keep working. To have AnyDB measure one, ask the Teammate to replace it with a watch on a report.

What a check-in looks like

Workspace health, 28 Oct

• Workflows: "Approve orders" failed 3 times since yesterday at its Update record step: Field Status not found. "Nightly sync" paused itself after 5 failures. Someone needs to fix the field name, then re-enable Nightly sync.
• Teammates: Release Verifier is blocked: the team's AI credits ran out.
• Plan: storage is at 8.6 of 10 GB (86%). The owner should upgrade or clean up attachments.
• Watches: Old approvals: 4 waiting for more than 7 days (alerted now).

From an AI assistant (MCP)

The same information is available from Claude, ChatGPT or any assistant connected through the AnyDB MCP connector, with the same access rules:

Tool	Use
anydb_list_workflow_failures	A workspace's failed workflow runs and paused workflows
anydb_teammate_health	How the Teammates in a workspace or the team are doing: health, counts and recent failures (team admins, or a Teammate with audit access)
anydb_team_usage	Usage against the plan's limits (team admins)