Skip to content

Guides

Set up & activate an agent

You turn a recurring task into your own AI agent, test it and release it for the organization.

An active Agent with clear instructions, a suitable model and exactly the tools it needs. It is usable in the chat, in workflows and via „Delegieren“ (Delegate).

AI & automation owner (design), Admin (activation, MCP approvals).

  • Admin or owner permissions for activation.
  • If external tools are to be used: a connected MCP server.
  1. Design – create a new agent under Aligia AI › Agenten › Agents. In the chat you describe what it should be able to do, for example “an agent that writes a weekly project status report”. Polaris writes the draft on the right. „Accept“ creates the agent. Alternatively you enter name, model, version and max. steps via „Manual“. A new agent is initially „Inaktiv“ (Inactive).
  2. Elaborate – On the detail page you maintain name, description and instructions. Under the properties you choose „Verwendung“ (Usage: „Haupt“ (Main), „Subagent“ or „Beides“ (Both)), the model and the max. steps.
  3. Assign capabilities – With „Skill hinzufügen“ (Add skill) you assign Agent skills. With „MCP-Server hinzufügen“ (Add MCP server) you assign servers. A custom agent reads Aligia data. It cannot change it directly. Externally it acts only via MCP tools.
  4. Approve tools – Which MCP tools the agent may use is defined by the tool permission on the server. Only „Allow“ is approved. „Ask“ and „Deny“ are currently both blocked.
  5. Activate – Only owners and admins may set the status to „Aktiv“ (Active). Aligia then checks whether a name exists and the model is permitted. Activation creates a dedicated agent account. Deactivating switches this account off again.
  6. Test – You try out an active agent with usage „Haupt“ or „Beides“ in Polaris via „Mit Agent chatten“ (Chat with agent). Alternatively a manual workflow starts it with a start.agent.<id> rule. You see every run under Executions.
  7. Deploy – Active agents are available in Issues and Tasks under „Delegieren“. Delegated agents work via a service account without admin permissions.

The agent is active, traceable and limited to the approved tools.

Agents · Connect an external tool via MCP · Check an execution & find errors