---
title: "AI-powered wizards"
manual: "LLM Foundation"
version: "0.44"
permalink: "https://docs.typo3.org/permalink/netresearch/nr-llm:administration-wizards@0.44"
source: "Administration/Wizards.rst"
rendered: "2026-11-12T17:35:48+00:00"
---

# AI-powered wizards {#administration-wizards}

The extension includes AI-powered wizards that use
your existing LLM providers to generate
configurations and tasks automatically. This reduces
manual setup to a minimum.

## Setup wizard {#administration-wizards-setup}

The setup wizard guides first-time configuration in
five steps:

1.  **Connect** — enter your provider endpoint and
    API key.
1.  **Verify** — test the connection.
1.  **Models** — fetch and select models from the
    provider catalogue.
1.  **Configure** — generate and select configuration proposals
    for the chosen models.
1.  **Review & Save** — review the provider, selected models and
    configurations, then save their records.

![Five-step setup wizard with progress indicator showing Connect, Verify, Models, Configure, and Save steps](../Images/backend-setup-wizard.png)

The setup wizard is the first entry in
**AI > Setup**. While no provider is
configured, the **AI > Overview** page opens
with a notice and a button to it. On TYPO3 v14 the
**AI > Setup** overview shows the same
notice; TYPO3 v13 has no such overview and opens the
last used entry of **AI > Setup**, or the
setup wizard on the first visit.

## Configuration wizard {#administration-wizards-config}

The configuration wizard generates a complete LLM
configuration using AI. Instead of filling in each
field manually, describe your use case in plain
language and the wizard generates everything.

1.  Navigate to **AI > Setup >
    Configurations**.
1.  Click **Create with AI**.
1.  Describe your use case (e.g., *"summarize blog
    posts in three sentences"*).
1.  The wizard generates: identifier, name, system
    prompt, temperature, and all other parameters.
1.  Review and click **Save**.

![Configuration wizard form with a plain-language description field and generated configuration preview](../Images/backend-config-wizard.png)

## Task wizard {#administration-wizards-task}

The task wizard creates a complete task setup — a
task **and** a dedicated configuration — in one
step.

1.  Navigate to **AI > Authoring >
    Tasks**.
1.  Click **Create with AI**.
1.  Describe the task (e.g., *"extract the five most
    important keywords from an article"*).
1.  The wizard generates: a task with prompt template,
    a configuration with system prompt and parameters,
    and a model recommendation.
1.  Review and click **Save**.

![Task wizard form with description field and generated task preview](../Images/backend-task-wizard.png)

## Model discovery {#administration-wizards-model-discovery}

On the model edit form, use the
**Fetch Models** button to query the
provider API. This auto-populates available models
with their capabilities, context length, and
pricing metadata.

### What the capability checkboxes are seeded with {#administration-wizards-capability-seed}

Discovery can use provider-reported capabilities, a bundled model catalogue,
family assumptions and operator edits. The provenance matters: a seeded
checkbox is not a successful test of that capability. See
[Managing models](https://docs.typo3.org/permalink/netresearch/nr-llm:administration-models@0.44) for how provenance is shown. The following table
summarizes the sources used by discovery:

| Provider | Discovery source |
| --- | --- |
| Mistral | chat, tools, vision (per-model `capabilities`) |
| OpenRouter | chat, tools, vision (`supported_parameters`, `architecture.input_modalities`) |
| Ollama | chat, tools, vision, embeddings (`/api/show`, Ollama 0.6 and newer) |
| Gemini | chat, streaming, embeddings (`supportedGenerationMethods`); vision and tools come from the built-in table for known model ids |
| Anthropic | chat, vision, tools, streaming for every model the listing returns — it returns Claude chat models only, and they all have them |
| OpenAI | from the built-in table, keyed by model id; an id outside it is seeded from its prefix (`dall-e-`, `tts-`, `whisper-`) and otherwise chat alone |
| Groq | chat only — the listing carries no capability field at all |

Where the API reports nothing, the record is seeded
with the narrowest true statement rather than a
guess. **Check the capability checkboxes after
discovery** and tick what the model actually does:
the field is yours to edit, and configurations that
select models by criteria match against it.
**Fetch Models** in the model form keeps
those edits where discovery cannot say what a model
lacks: for a Groq model, an Ollama model without a
reported list, and an OpenAI or Gemini chat model
outside the built-in table, it adds what it knows
and leaves the other ticks of a record of the same
kind alone.

> [!NOTE]
> Models discovered by an earlier version carry
> capabilities that were partly guessed from the
> model name. An upgrade cannot tell an operator's
> deliberate edit from a stale seed, so it does not
> overwrite either. Running **Fetch Models**
> again replaces a guessed tick only where discovery
> states the list: the built-in tables, Mistral,
> OpenRouter, and Ollama models whose capabilities
> the host reported. For the models named above,
> including Ollama models it did not look up or
> could not read, untick a stale capability by hand.

## Recommended workflow {#administration-workflow}

For a fresh installation:

1.  Run the **Setup wizard** from the dashboard
    to create your first provider, select models
    and save the generated configurations.
1.  Use the **Configuration wizard** to create
    additional use-case configurations (one per
    use case in your extensions).
1.  Use the **Task wizard** to create reusable
    prompt templates for editors.
1.  Share configuration identifiers with your
    extension developers — they reference them
    in code via
    `$configRepository->findByIdentifier('...')`.

For ongoing maintenance:

-   **Add providers** when you need additional
    AI services or separate prod/dev keys.
-   **Fetch models** periodically to pick up new
    models from providers.
-   **Edit configurations** to tune prompts and
    parameters — changes take effect immediately
    without code deployment.