Skip to main content
A custom agent is Primo AI working on a job you define. It starts when something happens in Primo or on a schedule you set, follows your instructions, and uses only the knowledge and connectors you give it. Start from scratch, or pick one of the Agent templates and adapt it.

Prerequisites

  • The Manage AI agents permission to create, edit, enable, or delete agents. The View AI agents permission is enough to see agents and their runs.
  • Knowledge base items published for any policy or procedure the agent should follow.
  • Connectors set up for any system outside Primo the agent should reach.

Create a custom agent

1

Go to Agents > AI Team

Click Browse templates. In the Templates catalog, pick a template to adapt, or click Start from scratch. See Agent templates.
2

Name and describe the agent

  • Name — up to 80 characters. It is how the agent appears in the list and on its runs.
  • Icon — click Shuffle colour to change the agent’s color.
  • Description — one or two sentences on what the agent does.
3

Add triggers

Click Add trigger and choose a type:
  • Event — the agent runs every time the selected Primo event occurs. Search the list and pick the event.
  • Schedule — the agent runs on a recurrence: Starting a date, every n hours, days, weeks, or months, at a time, in a time zone. Weekly schedules let you pick the days, monthly schedules the day of the month.
Click Save on the trigger card to keep it. You edit one trigger at a time, and an agent can have several triggers, events and schedules combined.
4

Write the instructions

Instructions are what the agent receives on every run: what to look at, what to do, what to leave to your team, and how to report. See Write good instructions.To draft them with Primo AI, click the assistant icon next to Instructions. It reads the name, description, and triggers on screen, then asks you questions or drafts from the setup. The draft lands in the field — nothing is saved until you click Create.
5

Pick the knowledge

In Knowledge, select the published knowledge base items the agent can read. The agent reads only the items you pick here. Unpublished items are not offered.
6

Choose the connectors

  • Web access — lets the agent search and read the web.
  • Connectors — turn on each connector the agent can use, including the Primo connector for your Primo data. Expand a connector to set a policy per tool:
An agent run has no one to answer a prompt, so there is no Require approval here. Each connector starts from its Agent permissions, set on the connector’s page in Agents > Settings > Connectors: by default, every tool is allowed. The count next to a connector, for example 12 of 40 tools enabled, is how many tools the agent can use. See Set tool permissions.
7

Set the model tier

In Advanced settings, pick the Model tier for every run of this agent:
8

Click 'Create'

The agent is created enabled. It runs the next time one of its triggers fires.
Grant the narrowest access that does the job. Set read-write tools to Always refuse unless the agent needs them, and check what it would change with a dry run first.

Write good instructions

The agent follows its instructions literally on every run. Cover four things:
  • The job — what the agent is for and what it looks at on each run.
  • The actions — what it may do on its own, and what it must propose instead of doing.
  • The limits — what it never does, and when it stops and hands back to your team.
  • The output — what it produces: a ticket, a comment, a report, and who it reaches.
Name the people, policies, and thresholds the agent should use. When a rule lives in a knowledge base item, pick that item in Knowledge rather than copying it into the instructions.

Test an agent with Dry run or Run now

You don’t have to wait for a trigger to see what an agent does. Open the agent: its page has Dry run and Run now. Dry run runs the agent on your real data without changing anything. Every email, ticket, comment or device action it would make is recorded instead of done. The run’s details show a What it would have done section before its report. Tools set to Always refuse stay out of a dry run, as they do on a real run. Run now runs the agent for real, after you confirm. Use it once a dry run shows what you expect. Both use the saved agent, so save your changes first. An agent that only runs on events can’t be started this way: it has nothing to work on. A disabled agent can still do a dry run, but needs Enable agent before Run now. A run you start shows as Running in the Runs tab until it ends.

Follow its runs

Open an agent and select the Runs tab to see every run, the trigger that started it, its Last activity, and its outcome. A run is everything the agent did from one start. On a ticket, the agent keeps one run: each time a ticket event wakes it, for example a new comment, that work joins the same run. Its duration and AI credits add up across each time the agent worked. Only a run started by a schedule, Run now, or Dry run ends with a Summary and, when the instructions ask for one, a run report. A run that waits on a ticket reads Waiting: the agent is idle, and the ticket is still open with the agent on it. Click a run to open its details:
  • Status, Trigger, Duration, and AI credits at the top.
  • A card when something is left to you: for a run that is Not done, the problem and its fix, with Edit agent. For a run that is done but leaves you an action, what to do, with a link to the page to act on when there is one.
  • The Summary of what the agent did, and the error if the run failed.
  • The Artifacts tab: each Run report, oldest first, then each file the agent wrote, with a preview. Hover a card to download it.
  • The Run tab: what the agent did step by step, with a line each time something woke the agent again. Click a step to see each tool call’s Input and Result. While the agent works, new steps appear as they happen.
  • The Tickets tab: the tickets the run opened or worked on.
A run with no artifact and no ticket shows its steps only. To stop a running agent, open Actions > Stop the agent in the run’s details. To read a run on a page of its own, open Actions > Open full page in its details. The Run page shows the whole conversation, with the status bar of the latest run at the bottom. While the agent works, the page updates as it goes, and you can stop the agent from the bar. When the run holds a ticket, Open ticket at the bottom of the page opens it. Primo writes the run report as a PDF only when the agent’s instructions ask for a report. In a dark cockpit, the report shows in dark. Its download is always the light version. Runs from before November 13, 2026 keep their HTML report. To see the runs of all your agents in one place, open Agents > Runs.

Manage your agents

The Agents list shows each agent with its type (Primo or Custom), triggers, status, and, over the last 30 days, its runs, the runs that need attention, and the AI credits it used. Use the tabs (All agents, Enabled, Disabled, Failing), search, and filters to narrow the list, and Save as to keep a view. From an agent’s Actions (…) menu:
  • Disable agent — stops all triggers. The agent keeps its configuration and run history. Enable agent turns it back on. A disabled agent keeps working on the tickets and tasks that already have it as their AI agent. You can’t put it on a new ticket or task: the AI agent picker shows it greyed out. To stop it on a ticket, set the ticket’s AI agent to None.
  • Delete agent — removes the agent. Past runs remain accessible.
Edits apply to the next run. While you have unsaved changes, the agent shows Pending changes.

Find your former workflows

On October 12 and 13, 2026, Primo moved the workflows you had set up to custom agents. Each agent keeps its workflow’s schedule and its instructions word for word, followed by a short note on how to deliver its reports and emails. Its report is now the agent’s run report instead of the workflow’s own PDF or CSV file. Primo didn’t copy past workflow runs. Some workflows came over as disabled agents, for example those that hadn’t run in the last 35 days. You find them in the Disabled tab of Agents > AI Team, and they don’t run until you click Enable agent in their … menu. A moved agent can use the tools of the Primo connector without asking, whether it came over enabled or disabled. Check each one’s connectors, and set read-write tools to Always refuse for any change it shouldn’t make on its own. See Create a custom agent. Workflows that weren’t active, or had no instructions, were not moved and no longer exist.