Reference & Data Model
Agent (agent)
An agent is the organization's own AI specialist: instructions, model, step limit, and assigned skills and MCP servers.
Purpose
Section titled “Purpose”An agent bundles a fixed working instruction with a language model and the tools it may use. It is run as its own chat, as a subagent of Polaris, as the target of a delegation, or from a workflow (start.agent.<id>). (Data type: configuration)
Where in the app
Section titled “Where in the app”Aligia AI › Agenten › Agents. The list shows status, model, version, max. steps, skills and servers. The detail page has name, description and instructions on the left, „Properties“, skills, MCP servers and tools on the right. New agents are created in conversation with Polaris or via „Manual“. See Guides › Agents.
Fields
Section titled “Fields”| Field | Technical | Type | Required | Meaning | Notes |
|---|---|---|---|---|---|
| Name | name |
Text | yes | display name, also the name of the service account | an empty name blocks activation |
| Description | description |
Text | no | what the agent is meant for; Polaris uses it to select subagents | Markdown |
| Instructions | initial_instructions |
Text | no | system instruction of the agent | Markdown; placeholder „Initial instructions“ |
| Status | status |
Selection | yes | „Aktiv“ (Active) / „Inaktiv“ (Inactive) | admin only; changeable only via server function, default „Inaktiv“ |
| Usage | usage |
Selection | yes | „Haupt“ (Main, main), „Subagent“ (subagent), „Beides“ (Both, both) |
default „Haupt“ |
| Model | model |
Selection | yes | platform model or model of an own LLM provider | models no longer allowed appear as „(unsupported)“ |
| Version | version |
Text | yes | free version label | default “1.0” |
| Max. steps | max_steps |
Number | yes | upper limit of steps per run | default 10 |
Relations
Section titled “Relations”| Related to | Kind | Required | On archive | On delete |
|---|---|---|---|---|
Agent skill (agent_skill_map) |
n:m | no | kept | mapping deleted along |
MCP server (agent_mcp_server_map) |
n:m | no | – | mapping deleted along |
Service account (app_user) |
belongs to (1:1) | no | – | agent deleted along when the account is deleted |
Conversation (agent_conversation) |
has many | no | – | reference kept; a deleted agent can no longer be run |
Mapping tables
Section titled “Mapping tables”agent_skill_map and agent_mcp_server_map contain only the two references. They are maintained via „Skill hinzufügen“ (Add skill) and „MCP-Server hinzufügen“ (Add MCP server) on the detail page. Removing a mapping requires an online connection.
States
Section titled “States”| Value | Label | Transition |
|---|---|---|
inactive |
Inaktiv (Inactive) | default after creation. Deactivating sets the service account to inactive. |
active |
Aktiv (Active) | owner/admin only. Prerequisite: name not empty, model allowed. Creates the service account the first time. |
A non-admin gets a message on toggling that they lack the permission.
Rules & automation
Section titled “Rules & automation”- Polaris offers only active agents with usage „Subagent“ or „Beides“ as subagents (at most 12, alphabetically).
- When loading an agent for a run, only skills with status „Aktiv“ and only tools with permission „Allow“ from servers with status „Katalog verfügbar“ (Catalog available) count.
- The „Tools“ section of the detail page lists all tools of the assigned servers, regardless of their permission.
- Agents cannot be archived, only deleted or duplicated.
See also
Section titled “See also”Agent skill · MCP server · LLM model · Conversation · Service accounts