Agents
An agent is an AI model bound to a personnel record with its skills and policies. Every agent has a responsible human; the agent's outputs and approval requests go to that person.
Where: Structure → Agents.
Creating an agent
The New Agent button asks for the following:
| Section | Field | Description |
|---|---|---|
| Basic information | Agent name, slug, title/role, status | The slug is used as the agent's address when tasks are handed between agents |
| Model and connections | LLM model | Models of providers whose key is defined are listed. If a model without a key is selected, a warning appears in chat |
| Responsible employee | The human who oversees the agent | |
| Department | When selected, that department's policies are added automatically | |
| Job description | Free text | The Suggest Skills button analyses the job description and suggests suitable skills |
| Skills | Pick from library / add tool | See Skills and Policies |
| Policies | Department policies + extra policies |
On creation, the agent becomes a personnel record (type: agent) and an agent configuration; it appears on the organization chart like other employees.
Statuses
| Status | Meaning |
|---|---|
Draft (draft) | Being prepared, not yet open for use |
Active (active) | Ready for use; selectable in chat, included in flow and task routing |
Archived (inactive) | Hidden from the list; record and history are kept |
Archive keeps the record; Delete is permanent. If you want to keep the history, archive.
Sending a change for approval
Instead of changing an agent's configuration directly, you can open a request with the Change Request button on the row (type: agent configuration, skill or policy). The request goes through two-stage approval, is applied when approved and, if wished, is written to Git. → Change Requests
The agent's memory
When a chat is closed, the system summarizes that session and adds it to the agent's long-term memory; these summaries are used as context in later sessions. Details: Chatting with an Agent → Memory.