--- url: 'https://dxpr.com/hc/documentation/learn-dxpr-builder/ai/models-and-providers' title: 'AI models and providers | DXPR' author: name: docs-sync url: 'https://dxpr.com/users/docs-sync' date: '2026-10-23T14:31:17+00:00' copyright: 'Copyright DXPR. All rights reserved.' type: article summary: 'The three Kavya models, where they run and how to restrict providers' tags: - 'Write and design with AI' image: 'https://dxpr.com/themes/custom/dxpr_com/images/dxpr-share-card.jpg' published: true og: site_name: DXPR updated_time: '2026-10-23T14:31:17+0000' 'image:alt': 'DXPR, powered by Drupal: AI-powered Drupal for marketers.' schema: '@context': 'https://schema.org' '@graph': - mainEntity: { } - '@type': TechArticle headline: 'AI models and providers' description: 'The three Kavya models, where they run and how to restrict providers' image: '@type': ImageObject url: 'https://dxpr.com/themes/custom/dxpr_com/images/dxpr-share-card.jpg' width: '1200' height: '630' datePublished: '2026-10-23T14:31:17+0000' dateModified: '2026-10-23T14:31:17+0000' author: '@type': Organization '@id': 'https://dxpr.com/#organization' name: DXPR url: 'https://dxpr.com/' logo: '@type': ImageObject url: 'https://dxpr.com/themes/custom/dxpr_com/dxpr-logo-dark.svg' width: '1348' height: '501' publisher: '@type': Organization '@id': 'https://dxpr.com/#organization' name: DXPR url: 'https://dxpr.com/' logo: '@type': ImageObject url: 'https://dxpr.com/themes/custom/dxpr_com/dxpr-logo-dark.svg' width: '1348' height: '501' mainEntityOfPage: 'https://dxpr.com/hc/documentation/learn-dxpr-builder/ai/models-and-providers' articleSection: 'Write and design with AI' - '@type': WebPage '@id': 'https://dxpr.com/hc/documentation/learn-dxpr-builder/ai/models-and-providers' description: 'The three Kavya models, where they run and how to restrict providers' breadcrumb: '@type': BreadcrumbList itemListElement: - '@type': ListItem position: 1 name: Home item: 'https://dxpr.com/' publisher: '@type': Organization '@id': 'https://dxpr.com/#organization' name: DXPR inLanguage: en --- [Help Center](/hc)/[Documentation](/hc/documentation) /[Learn DXPR Builder](/hc/documentation/learn-dxpr-builder) /[Build pages](/hc/documentation/learn-dxpr-builder/build) /[Write and design with AI](/hc/documentation/learn-dxpr-builder/ai) # AI models and providers. The three Kavya models, where they run and how to restrict providers Learn DXPR BuilderBuild pagesWrite and design with AI ## In this guide Related guides - [Configuration](/hc/documentation/learn-dxpr-builder/ai/configuration) - [AI API reference](/hc/documentation/learn-dxpr-builder/ai/api-reference) - [AI troubleshooting](/hc/documentation/learn-dxpr-builder/ai/troubleshooting) DXPR Builder's AI features run through the Drupal AI module and the DXPR provider module (`ai_provider_dxpr`). This page explains what a provider is, how the DXPR provider fits in, which models exist and what the "service providers" and weights on the settings pages mean. The video chooses the EU model in the editor and shows it in the AI Agent panel. ## Providers in the Drupal AI module In the AI module, a provider is a plugin that connects one AI service. Each provider declares the operation types it supports, such as chat, text to image, image to image and translate text. It also lists its models and holds its own API key settings. Installed providers are listed at **Configuration** > **AI** > **AI Infrastructure** > **AI Platform Providers** (`/admin/config/ai/providers`). This needs the `administer ai providers` permission. The default provider and model for each operation type are set at **Configuration** > **AI** > **AI Infrastructure** > **Default Models for AI Operations** (`/admin/config/ai/settings`). They are stored in `ai.settings` as `default_providers`. ## The DXPR provider `ai_provider_dxpr` registers the provider with ID `dxpr` and label "DXPR". It supports `chat`, `text_to_image`, `image_to_image` and `translate_text`, streams chat output, and talks to `kavya.dxpr.com/v1` with the API key selected from the Key module. Its settings form is at `/admin/config/ai/providers/dxpr` (**DXPR AI Provider**): - **DXPR API Key**: the Key entity that holds your DXPR API key (created at `/admin/config/system/keys`; on install the module selects a key whose label contains "DXPR") - **AI Usage Tracking**: credit balance and use - **Default model** and **Allowed models**: see Models below - **AI Service Providers**: Automatic or Manual Override, with a drag table of upstream services - **Em-dash processor**: post-processing level, with per-language overrides Saving the form also registers `dxpr` as the default provider for each operation type that has no default yet. Chat variants and translate text get `kavya-m1`. Text to image and image to image get `kavya-image`. Saving also sends one short test request to check the key and the account's rate limit. DXPR Builder's own AI page then shows your credits and a link to this form: **Configure AI service providers**. ## Which provider DXPR Builder uses DXPR Builder sends every AI request to the DXPR provider by name. So text commands, page creation, image generation and image editing always use the DXPR provider. The defaults at `/admin/config/ai/settings` do not change this. The routes behind these features are on [AI API reference](/hc/documentation/learn-dxpr-builder/ai/api-reference). Other providers in the AI module, such as OpenAI, Anthropic or Mistral, cannot serve DXPR Builder features. They can still be the defaults for other AI features on the same site, for example AI Image Alt Text or AI Translate. ## Models | Builder key | API model | Used for | |---|---|---| | `dxai_kavya_m1` | `kavya-m1` | Text commands and page creation (default) | | `dxai_kavya_m1_eu` | `kavya-m1-eu` | Text, routed to MistralAI (France) | | `dxai_kavya_m1_fast` | `kavya-m1-fast` | Text, lower latency | | (none) | `kavya-image` | Image generation and editing, always | A model that is not on **Allowed models** is refused. DXPR Builder's chat route answers `403` with "The selected model is not allowed. Please refresh the page to update your model options." For other AI module features, the provider answers "The selected model is not allowed by the administrator." Where the default text model is set: 1. With `ai_provider_dxpr` installed: **Default model** at `/admin/config/ai/providers/dxpr`(`ai_provider_dxpr.settings`, key `ai_model`). The DXPR Builder AI settings page then only shows a **Configure AI service providers** link 2. Without it: **Default model** and **Allow users to choose model** at **DXPR Studio** > **DXPR Builder** > **AI**(`/admin/dxpr_studio/dxpr_builder/ai_settings`, `dxpr_builder.settings`) ## Choose a model in the editor Editors can pick the model for their own AI requests: 1. On the DXPR Builder toolbar, click **More** and choose **AI settings**. 2. Under **Model**, choose a model, for example **DXAI Kavya M1 EU**. 3. Click **Save**. 4. Click **AI Agent** on the toolbar. The panel's header names the model you chose. The choice is stored in the browser (`ck5-ai-agent:model`). With **Allowed models**limited to one model, no selector is shown. ## Service providers and weights Kavya M1 is served by upstream AI services. The settings forms call these "service providers": OpenAI, MistralAI, Anthropic, Google Gemini and XAI. They are unrelated to Drupal AI module providers; they only steer routing inside the DXPR service. Two config keys control this, stored in `ai_provider_dxpr.settings` when that module is installed and otherwise in `dxpr_builder.settings`(schema in `config/schema/dxpr_builder.schema.yml`): | Key | Values | Effect | |---|---|---| | `ai_provider_selection_mode` | `automatic` (default) or `manual` | Automatic: no provider list is sent and DXPR routes on live performance. Manual: the enabled providers are sent with the request | | `ai_providers` | Per provider: `enabled`, `weight`, `region` | `weight` is an integer; lower means higher priority. `region` is `enabled` or `disabled` (the table section the row sits in) | In manual mode the field formatter passes the enabled providers to the browser. The browser sorts them by weight, lowest first, and sends them as a comma-separated `providers` string with each chat request, for example `openai,mistral,anthropic`. The DXPR service uses the first provider and falls back to the next enabled one when it is unavailable or slow. Weights are only sent when the model is `kavya-m1`; the EU and Fast models have fixed routing. Default weights: OpenAI 0, MistralAI 1, Anthropic 2, Google Gemini 3, XAI 4. Dragging a row up in the table lowers its weight. To change the order: 1. Go to `/admin/config/ai/providers/dxpr` (or `/admin/dxpr_studio/dxpr_builder/ai_settings`without the provider module) 2. Set **Default model** to DXAI Kavya M1; the provider options are hidden for other models 3. Choose **Manual Override** (**Manual priority**on the DXPR Builder page) 4. Drag providers between the Enabled and Disabled sections and into the order you want 5. Click **Save configuration** (**Save settings**on the DXPR Builder page) ## AI credits AI features consume DXPR credits from your subscription. Check remaining credits at **DXPR Studio** > **DXPR Builder** > **AI** under **Usage and Credits**, or under **AI Usage Tracking** on the DXPR AI Provider form. Top up at dxpr.com/user/me/subscription/credits. ## What's next? - [AI configuration](/hc/documentation/learn-dxpr-builder/ai/configuration) - [AI API reference](/hc/documentation/learn-dxpr-builder/ai/api-reference) - [AI troubleshooting](/hc/documentation/learn-dxpr-builder/ai/troubleshooting)