Advisor
The workspace-wide list of configuration issues across agents, knowledge bases, campaigns, numbers and contacts, and how to fix or ignore each one.
Advisor, under Observability in the sidebar, checks every object in the workspace against a set of rules and lists what fails. The app describes it as configuration issues detected across your agents, knowledge bases, campaigns and numbers. Each failed check is a finding: one rule, on one object.
Use it to catch problems before a caller does: an agent with no voice, a campaign with nobody to call, a number that can no longer connect.
How findings are grouped
The page is a board with one lane per severity, each headed by its count:
| Lane | What lands there |
|---|---|
| Errors | Something that stops the object from working, for example an outbound campaign with no phone number. |
| Warnings | Something that works but will cause trouble, for example two agents sharing a name. |
| Suggestions | Advice, for example a knowledge base that no agent uses. |
The severity comes from the rule, not from you: you can’t move a card between lanes. A lane with nothing in it reads Nothing here. A long lane shows its cards a page at a time as you scroll.
Reading a finding
Each card carries:
- an icon for the kind of object it concerns (agent, knowledge base, campaign, phone number, contacts or the workspace itself);
- the action to take as its title, for example Choose a voice or Select contacts;
- the name of the object, when the finding has one;
- one or two lines on why it matters, for example: a campaign with no contacts selected has nobody to call.
Below that, up to three actions:
| Action | What it does |
|---|---|
| Fix | Opens the page where the fix lives: the relevant section of the agent or knowledge base, or the campaign’s or number’s own page. Contacts and workspace findings have no page to open, so they show no Fix. |
| Fix it | Opens Atlas with a request to fix this finding already written, naming the object. Nothing is sent until you send it. Only rules that ship such a request show Fix it. |
| Ignore | Moves the finding out of its lane and into Ignored. |
When the checks run
Findings are kept up to date without you asking. The top right of the page says when: Next check followed by a time when a scheduled check is coming, Checking now… while it runs, or Checked followed by how long ago when no scheduled check applies.
Opening an agent’s Recommendations section checks that agent again, and Re-check there, or in a campaign’s Advisor section, checks that one object straight away. A finding whose problem is gone drops off the board after the next check.
Refresh, at the top right of this page, reloads the latest findings. It does not run the checks again.
Ignoring a finding
Ignore is for a finding you have decided to live with. It stays ignored when the checks run again, so it does not come back on the next check.
Ignored findings are hidden by default. When there are some, Show N ignored at the top right opens a fourth lane, Ignored, with those cards dimmed; Hide ignored closes it. The choice is kept in the page address, so a reload or a shared link opens the board the same way. Ignored cards have no actions: the board has no control to bring an ignored finding back.
When everything passes
With no open findings the board is replaced by All checks passing and a line saying no misconfigurations were detected across the workspace, with the same check time underneath. Ignored findings don’t count as open, so a workspace whose only findings are ignored also shows this state.
If the findings can’t be loaded, the page reads Could not load checks with Try again.
What it checks
The rules are grouped by the kind of object they look at. A selection:
| Object | Examples |
|---|---|
| Agents | Prompt is configured, Voice selected, Voice matches the language, Flow has nodes, Outcomes defined, Maximum call duration set, Knowledge base is usable, At least one test scenario, Agent name is unique |
| Knowledge bases | Knowledge base has content, Sources finished ingesting, Last ingestion succeeded, Knowledge base is in use, Knowledge base name is unique |
| Campaigns | Outbound campaign has a number, Campaign number is usable, Campaign has contacts, Schedule is valid, Campaign name is unique |
| Phone numbers | Phone number is active, Phone number has a concurrency limit |
| Contacts | No duplicate contact numbers |
| Workspace | Workspace has contacts |
The agent checks are the ones listed on Recommendations, where each is explained.
Other places the checks appear
Advisor is the workspace-wide view. The same findings also show up next to the object they concern:
- Recommendations, in an agent’s workspace, lists that agent’s open findings ranked by impact instead of by severity. It is the same finding underneath: dismissing a recommendation there ignores it here too.
- Advisor in a campaign’s page lists that campaign’s findings in the same Errors, Warnings, Suggestions and Ignored groups, with Re-check, Fix it and Ignore.
Fix a problem where you find it; the board catches up on its next check.