Reference & Data Model
LLM model (llm_model)
An LLM model is a concrete model of an own provider with declared facts, prices and validation status.
Purpose
Section titled “Purpose”The LLM model makes a provider model selectable in Aligia. Context window, output limit, reasoning and image capability control how the engine addresses the model. Prices serve for recording; no billing takes place. (Data type: configuration)
Where in the app
Section titled “Where in the app”Detail page of the LLM provider, section „MODELLE“ (Models) with „Hinzufügen“ (Add), dialog „Neues Modell“ (New model) / „Modell bearbeiten“ (Edit model). Models are selected on Agents, in the Polaris chat („Modell“) and under Einstellungen › Arbeitsbereich › Automatisierung. See Guides › LLM provider.
Fields
Section titled “Fields”| Field | Technical | Type | Required | Meaning | Notes |
|---|---|---|---|---|---|
| Model ID | model_id |
Text | yes | ID sent to the provider | stored as text on agents |
| Label | label |
Text | yes | display name | – |
| Context window | context_limit |
Number | no | max. tokens per request | empty = unknown; the table shows „Kontextfenster unbekannt“ (Context window unknown) |
| Max. output tokens | max_output_tokens |
Number | no | upper limit per response | empty = provider default |
| Reasoning | reasoning |
Selection | yes | „Kein Reasoning“ (No reasoning, none), „Effort-gesteuert“ (Effort-controlled, effort), „Adaptiv (Modell entscheidet)“ (Adaptive – model decides, adaptive) |
declared, not validated |
| Accepts image input | input_image |
Yes/No | yes | images from the chat are sent as image content | declared, not validated |
| Input | input_price_per_m |
Amount | yes | USD per 1M input tokens | default 0 |
| Output | output_price_per_m |
Amount | yes | USD per 1M output tokens | default 0 |
| Cached input | cached_input_price_per_m |
Amount | yes | USD per 1M cached input tokens | default 0 |
| Status | status |
Selection | yes | validation state, see States | System |
| Last error | last_error |
Text | no | reason of the last error | System; tooltip on the badge |
Not visible: agent mode, time of the last validation.
Relations
Section titled “Relations”| Related to | Kind | Required | On archive | On delete |
|---|---|---|---|---|
| LLM provider | belongs to | yes | – | model deleted along |
States
Section titled “States”| Value | Label | Meaning |
|---|---|---|
unchecked |
Nicht geprüft (Not validated) | never probed; default |
validating |
Wird geprüft… (Validating…) | validation running |
connected |
Verbunden (Connected) | validation successful |
error |
Fehler (Error) | validation failed |
Organization task model (org_llm_task_model)
Section titled “Organization task model (org_llm_task_model)”Per task (task: chat, review, title, extractor) the organization stores at most one model ID (model_id). Visible under Einstellungen › Arbeitsbereich › Automatisierung as „Polaris-Standardmodell“ (Polaris default model), review model, model for chat titles and „Extraktions-Modell“ (Extraction model). Empty = platform model. Owner/admin only. See Guides › Workspace.
Rules & automation
Section titled “Rules & automation”- An agent whose model is no longer allowed cannot be activated (message
unsupported_model). - „Modell entfernen“ (Remove model) deletes the record immediately without confirmation.