AI Agents
AI Agents
An AI agent does a job for you, on its own. Where the AI Assistant
answers a question while you chat with it, an agent is a saved set of steps — look something up, ask the model,
run a script, decide, notify, remember — that runs whenever it is told to: on a schedule, when a condition becomes
true, when an automation button asks, or when someone types @ in the assistant chat.
Typical agents check overnight failures and email a summary, watch EDI documents and chase missing acknowledgements, digest a shift's downtime, or propose fixes for a failing script for a person to approve.
Who can use AI Agents
- Licence: your company's plan must include AI Agents. Without it the agent
screens, the automation action, and the
@agents in chat are hidden. Letting an agent change things without asking needs a further plan add-on; without it, Acts on its own cannot be chosen. - Building and running agents (Agents, Workbench, Catalog, Runs, Memory, Fixtures, triggers) needs the DataMagik - Builder role.
- Stopping all agents with the kill switch needs DataMagik - Security as well as Builder. Setting up Notification channels needs DataMagik - Security.
- Approving an agent's proposed change is done from the approvals inbox by anyone who holds the role chosen as the agent's approver.
Finding the agent screens
Open Admin → AI Assistant → Agents. The same menu holds Agent runs and Agent fixtures. Inside the module a tab bar switches between Agents, Catalog, Runs, Memory and Fixtures.
/datamagik/admin/ai-agents.The Agents list
The list shows every agent in your company with what it does, when it runs next (for example Next … or Not scheduled), how its last run went, its autonomy and approver. Search with Search agents… and filter with All agents / On only / Off only.
| Control | What it does |
|---|---|
| Start from a ready-made agent | Opens the Catalog of agents we built, ready to install. See Creating an AI Agent. |
| Create agent | Opens the Workbench to build a new agent by doing. |
| On / Off switch | When On, the agent's triggers can start it. When Off, triggers do not start it, but you can still run it yourself. |
| Run | Starts one run now — even if the agent is off, so you can try it. Progress shows in the row (Running…, then finished or failed with the reason); Follow it in History opens the run. |
| History | Opens Agent runs filtered to this agent. |
| Edit | Opens the agent in the Workbench. |
| Update available / What's new | Shown on catalog agents when a newer version has been released. |
Some agents need something before they can run. An EDI agent asks you to pick the document (Run it on this one), and an agent whose Plex data source has not been matched yet shows Needs a Plex data source with Choose a data source. You are asked in words rather than having a run start and fail.
Deleting an agent
Tick one or more agents and use the delete bar, or delete from the Workbench. The dialog lists everything that goes with it — its saved versions, triggers and schedule, run history, memory, fixtures and share links — and closes any pending proposals. The AI Audit Log is kept. Type the agent's name to confirm. This cannot be undone.
Who an agent acts as
An unattended run does not act as a person. It acts as a platform identity holding the DataMagik - Agent role, which starts with the Builder permissions; a security administrator can narrow it under Roles. The agent's own What it can use settings narrow it further — both limits apply.
Learn more
- Creating an AI Agent — the Workbench, the Catalog, and turning a chat into an agent.
- When AI Agents Run — schedules, conditions, automation buttons and
@in chat. - AI Agent Approvals, Safety & Monitoring — approvals, notifications, run history, memory and the kill switch.