Providers

Zammad offers AI-powered features to support your agents! To use AI features in Zammad, you must first configure and enable at least one AI provider. An AI provider processes your AI requests and is required for any AI feature. Add your providers in Zammad’s admin settings under AI > Providers and make sure to activate the toggle. Switching the toggle off stops all AI calls, even if one or more providers are configured.

Accessing this page requires the admin.ai_provider permission.

Note

Zammad’s AI features are completely optional. If you don’t want to see the AI section at all, read about how to hide it in the system docs.

Managing Providers

The AI provider configuration page shows a table of all configured providers. Each row has a status dot, the provider’s name and type, and badges for any special roles the provider currently has. The ︙ menu in the Action column lets you change these roles or remove the provider. See Default Provider and Capabilities for details. You can add one provider for all AI features you want to enable, or add multiple to assign different providers to individual features.

Add or Edit a Provider

Click New Provider to add a new provider or click on a row to edit an existing one. Both ways open a two-step dialog. The dialog’s first step lets you pick the provider type, enter a name and provider-specific information. Switching to the second step via the Next button verifies your entries, connects to the provider and fetches the available models. If there is a problem, you are notified at this point already. The second step allows you to select or add a model and to configure provider-specific settings. The fields shown depend on the provider you choose, so switching the type updates the dialog on the fly.

Type

Pick the AI provider. Available options:

• Zammad AI

• OpenAI

• Ollama

• Anthropic

• Azure AI

• Mistral AI

• Custom (OpenAI compatible)

Different fields appear depending on the type. Some providers also require configuration on their side before you can connect (for example Azure AI). Refer to the provider’s own documentation for those steps.

Note

The provider you need isn’t listed? We are going to add more over time. If you need a particular one or want to sponsor it, get in touch with our sales department.

Name

A human-readable label for the provider. This is what you see in the list and elsewhere in Zammad (for example when configuring a provider for a specific feature). The name must be unique.

Token

The API token issued by your provider. Don’t confuse it with the tokens a large language model deals with when processing a request.

Model

The base model used to generate text. For providers that offer a model list, the field is a dropdown. The default option shows the provider’s recommended model; pick a different one if you prefer. If the provider does not offer a model list, enter its name manually.

Azure AI and Zammad AI do not show this field. Azure AI identifies the model through its deployment URL, while Zammad AI chooses the model itself.

Tip

Need different models for different features? Add the same provider twice with a different model selected in each, then assign each connection to its feature via Configuring Provider per Feature.

Embedding Model

The model used to vectorize text for semantic search. The dropdown lists the models the provider offers. The default option shows the provider’s recommendation; pick a different one if you prefer. If the provider does not offer a matching model, enter its name manually. Required if you plan to assign the Semantic search capability to this connection (see Capabilities). Not every provider supports semantic search; the field appears only for the ones that do.

When you select or enter an embedding model, Zammad tries to determine values for the two options listed below. If Zammad cannot determine these values, you must provide them. Note that you should only adjust these values if you know what you are doing.

Embedding dimensions

The length of the vectors the embedding model produces.

Context window size

The number of input tokens the embedding model accepts at once.

Note

Changing the embedding model or the embedding dimensions triggers a full rebuild of the knowledge base index. This re-embeds every knowledge base article, which takes time and costs AI calls proportional to the size of your knowledge base. The rebuild runs in the background and Zammad notifies you with a toast message when it starts. Switching the semantic search capability to a provider that uses the same embedding model and dimensions does not trigger a rebuild.

OCR Model

The model used to extract text from images. The dropdown lists the models the provider offers. The default option falls back to the Model field. Zammad AI chooses the model itself and therefore doesn’t show this field. Not every provider supports image text recognition; the field appears only for the ones that do.

URL

The URL or IP address of the provider. Required for Ollama and OpenAI compatible providers.

URL (Completions)

Azure AI only. The endpoint where the text generation model is available.

URL (OCR)

Azure AI only. The endpoint where the OCR model is available. Leave empty to fall back to URL (Completions).

After filling in the fields, click Submit. Zammad tests the configuration before saving. If the test succeeds, the row’s status dot starts orange until the first successful request turns it green; see Status below for the colors.

Status

The colored dot at the start of each row shows whether the provider has been used successfully or whether it has issues:

Green

The provider has been used and the last request succeeded. The tooltip shows the timestamp of the last status update.

Orange

The provider is configured but has not been used yet. It turns green after the first successful request.

Red

The last request to this provider failed. Hover the dot for the error message from the provider. The tooltip always shows the most recent error, so the timestamp belongs to the last failed request.

Default Provider

The Default provider serves every AI feature that has no specific provider assigned. You set it from the ︙ menu in the Action column. A provider can also carry capability badges, so a single well-equipped provider can cover everything.

When you mark a provider as the default, Zammad clears that flag from whichever provider held it before, so the new default takes over immediately. The default provider always exists: when you delete the provider that holds the flag, Zammad promotes the oldest remaining provider.

Capabilities

Semantic search and image text recognition (OCR) are capabilities. You assign them from the ︙ menu in the Action column; the badges mark which provider currently covers each capability.

Semantic search

The provider used to turn text into numerical form (vector embeddings) so the knowledge base can find answers by meaning, not just keywords.

If no provider covers semantic search and a feature needs it, Zammad shows a warning on the corresponding feature page. Assign semantic search to a provider that supports it and that has an Embedding Model set to clear the warning. Note that switching this assignment may trigger a full index rebuild, see Embedding Model above.

Image text recognition

The provider used to extract text from images (OCR).

This capability is optional. To save on AI-related costs, you can remove it from all providers entirely; features then simply skip the OCR step.

When you assign a capability to a provider, Zammad clears that assignment from whichever provider held it before, so the new assignment takes over immediately. Unlike the default provider, capabilities are not promoted: when you delete the provider that covers a capability, the capability simply stays unassigned.

Actions

The ︙ menu in the Action column offers the following actions:

Set as default

Marks the provider as the default provider for AI features.

Use for semantic search / Do not use for semantic search

Assigns the provider to the semantic search capability, or clears that assignment. Only available for providers that support it.

Use for image text recognition / Do not use for image text recognition

Assigns the provider to the image text recognition capability, or clears that assignment.

Delete

Removes the provider. Not available for the Zammad AI provider on SaaS.

Configuring Provider per Feature

By default, all AI features use the provider marked as the default. For finer control, you can pick a specific provider for individual features.

Every AI feature settings page shows a Provider button in the page header. It opens a modal with a dropdown of all configured providers. Pick a provider to send that feature’s text generation requests to, or pick the Default (…) entry to clear any override and fall back to the default.

The choice only affects text generation. Semantic search always goes to the provider that covers the semantic search capability, and image text recognition always goes to the provider that covers the image text recognition capability. The Provider button is only visible to admins who hold the admin.ai_provider permission, since routing touches the providers you set up on this page.

Additional Provider Information

If you want to know more about Zammad AI or Ollama, read on. Otherwise, head over to Feedback & Logs or pick any of the AI feature pages from the menu on the left.

Zammad AI

Using Zammad AI as a provider is the easiest way to get started with Zammad’s AI features. It has several key advantages:

• You don’t have to set up an AI system or its configuration.

• Hosted in the EU and compliant with GDPR.

• No configuration in Zammad for SaaS customers, or just an API key for self-hosted customers.

• Requests with sensitive information aren’t used for training.

• One place to go for support when something isn’t working as expected.

SaaS Customers

Using Zammad AI requires a “V2” plan. Check your subscription settings and consider switching your plan. When a “V2” plan is active, you can buy AI calls for AI processing. A Zammad AI provider is added automatically and you can enable it as you like. Check your remaining AI calls from time to time or activate the notification to be alerted when few calls remain.

On SaaS, the Zammad AI provider is provisioned by the platform and you cannot delete it or change its type.

Self-Hosted Customers

Get in touch with our sales department to obtain a Zammad AI API key and to buy AI calls. After receiving your API key, add a new provider of type Zammad AI and save the key.

Ollama

Ollama lets you run your own AI server on a machine you control, so your data is not transferred to a third party. This approach only makes sense if you have a powerful GPU in your system!

If you don’t have Ollama running yet, their documentation is a good starting point. Available models are listed in Ollama’s model library, with useful information and additional links in the Ollama GitHub repository.