> For the complete documentation index, see [llms.txt](https://docs.ixhello.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.ixhello.com/ixhc2/integrations/expand-llm-catalog.md). # Expand LLM Catalog ### Overview Expand LLM Catalog introduces the latest generation of AI models to the iX Hello vNext model catalog, giving builders and administrators greater flexibility to select the most suitable model for their specific business and customer experience needs. The expanded catalog now includes support for: * Anthropic Claude 4.x family * OpenAI GPT-5.3 series text models * Next generation real-time audio models With a broader range of AI models available, organizations can optimize model selection based on performance, capabilities, and use case requirements. This expansion enhances support for: * Text-based experiences such as content generation, summarization, and question answering * Coding and agentic workflows that require advanced reasoning and task execution * Conversational AI applications for natural, context-aware interactions * Voice and real-time experiences that enable low-latency speech and multimodal interactions By expanding the model catalog, iX Hello Customer v2 provides greater flexibility, improved AI capabilities, and more options for building intelligent customer and employee experiences. *** ### What Problem Does It Solve 1. **Expands Model Options** * The expanded model catalog gives builders access to a broader selection of AI providers and model families, removing the limitations of a narrower model set. * This enables teams to: * Choose the most appropriate model for specific use cases * Balance performance, speed, and cost requirements * Optimize AI experiences based on business needs 2. **Reduces Single-Provider Dependency** * By supporting models from multiple AI providers, the platform reduces reliance on a single vendor. * This provides: * Greater flexibility in model selection * Improved resilience and future scalability * The ability to adopt new innovations as they become available 3. **Improves Model-to-Task Fit** * Different AI workloads require different capabilities. The expanded catalog allows builders to select models that best match the task at hand, whether it requires: * Fast, cost-efficient responses * Long-context understanding and reasoning * Advanced coding and agentic capabilities * Realtime voice and conversational interactions * This helps improve both application performance and user experience. 4. **Enables Next-Generation Voice Experiences** * Support for newer real-time audio models provides a modern foundation for voice-based applications. * Benefits include: * Improved support for low-latency voice interactions * Enhanced real-time conversational experiences * Reduced reliance on older voice models that may be deprecated over time * Overall, the expanded LLM catalog gives organizations greater flexibility, better model alignment, and access to the latest AI capabilities across text, coding, conversational, and voice use cases. *** ### Why Use the Expanded LLM Catalog The expanded LLM catalog provides greater flexibility, better performance alignment, and improved scalability by giving organizations access to a wider range of AI models optimized for different workloads. 1. **Better Model Choice and Quality** Different AI models excel at different types of tasks. The expanded catalog enables builders to select the model that best fits their application requirements, helping improve both performance and user experience. **Examples include:** * Claude Opus 4.7 for complex reasoning, advanced problem-solving, and agentic coding tasks * Claude Sonnet 4.6 for a balanced combination of performance, cost efficiency, and large-context processing * Claude Haiku 4.5 for fast, responsive, and cost-effective interactions * GPT-5.3-codex and GPT-5.3-chat for advanced coding, automation, and conversational experiences 2. **Increased Resilience and Vendor Flexibility** Supporting multiple AI providers reduces dependency on a single model vendor and provides greater operational flexibility. This benefits critical platform capabilities such as: * Runtime Engine workloads * Flow Summary generation * Debug Inspector system flows A multi-provider approach also helps organizations adapt more easily as new models and technologies become available. 3. **Cost Optimization** Not every use case requires the most powerful model. The expanded catalog allows teams to select a model that meets business requirements while optimizing operational costs. Organizations can choose from a performance-based model tiering strategy: * Haiku 4.5 / GPT-5-mini for lightweight, high-volume workloads * Sonnet 4.6 / GPT-5.3-chat for general-purpose intelligent interactions * Opus 4.7 / GPT-5.3-codex for advanced reasoning and coding scenarios * Realtime-2 for voice and real-time experiences This flexibility helps balance quality, performance, and cost. 4. **Improved Voice and Realtime Experiences** The introduction of gpt-realtime-2 provides a modern foundation for voice-based applications and real-time conversational experiences. **Benefits include:** * Enhanced voice interaction quality * Lower-latency conversational experiences * Support for future voice capabilities and innovations * Reduced dependence on older real-time model versions that may eventually be retired 5. **Greater Operational Scalability** The platform's provider abstraction layer simplifies the integration of new AI model families. **This enables:** * Faster adoption of emerging AI technologies * Easier expansion of the model catalog over time * Reduced integration effort for product and engineering teams * A more scalable and future-ready AI ecosystem *** ### Feature Goals 1. **Provider-Agnostic Runtime Architecture** The Runtime Engine uses a provider abstraction layer, making it easier to add and support new AI models without changing core runtime functionality. 2. **New Model Deployments** The platform will introduce: * Three Claude model variants * Two GPT-5.3 model variants These models will initially be deployed in non-production environments for evaluation and testing before being promoted to production. 3. **Realtime Voice Model Support** The platform can support gpt-realtime-2 for eligible environments, enabling newer voice experiences without requiring changes to the Channel Gateway. 4. **GPT-5.x Reasoning Enhancements** GPT-5.x models require explicit reasoning configuration to ensure optimal performance. **Key updates include:** * Support for the reasoning\_effort parameter * Availability of reasoning\_effort=xhigh for advanced coding and reasoning workloads * Updates to model-specific settings for reasoning-enabled variants 5. **Enhanced Model Catalog Service** The Configuration Service now returns detailed model metadata instead of only deployment names, helping applications and users discover available model capabilities more easily. Access to the catalog continues to be controlled through the flows:read permission. 6. **Builder Experience Updates** Model selection is available across all supported AI configuration surfaces, including: * Global Persona AI Model dropdown * Flow Designer AI-powered nodes: * Ask * Agent * Summary * Classifier 7. **End-to-End Model Validation** All newly supported models are validated to ensure they work correctly with supported platform capabilities, including: * Text conversations * Voice conversations * Streaming responses * Tool calling * Supported runtime behaviors *** ### Where Users Will See This Feature 1. **Global Persona Configuration** The AI Model dropdown in the Global Persona settings now includes the newly supported models. 2. **Flow Designer Nodes** Builders can select the new models directly within AI-powered flow nodes, including: * Ask * Agent * Summary * Classifier 3. **Environment-Specific Model Catalogs** Available models may vary depending on the environment and use case. Separate catalogs can be maintained for chat and voice experiences based on: * Environment configuration * Supported modality (chat or voice) * Model availability and deployment status **List of Model Catalog:**
*** ### How the feature works 1. **Model metadata is returned by configuration** * When the platform retrieves available models, it uses the Configuration service. * The service returns model metadata including: * Provider * Capabilities * Extended thinking * Reasoning effort * Context window * Output cap * Modality * Supported APIs * Tier metadata 2. **Builders choose models in the UI** * Users select models from supported dropdowns in persona settings or flow nodes. * The available set depends on the configured catalog for that environment. 3. **Runtime uses provider abstraction** * Once selected, the Runtime Engine invokes the model through a provider abstraction layer. * This avoids one-off integrations for each model family. 4. **Runtime behavior varies by model capability** * Some models are optimized for: * Reasoning * Coding * General conversation * Realtime audio * The platform uses catalog metadata to understand those differences. *** ### User Workflow: Selecting and Configuring AI Models The platform supports multiple AI providers and models, allowing you to choose the best option based on your channel, use case, and performance requirements. 1. Create or Open a Flow 2. Navigate to Flow Designer. 3. Create a new flow or open an existing one.
4. Select the channel you want to configure: * Chat * Voice
2. **Select a Language** * In the flow configuration, choose the desired Language. * The available model options will be displayed based on the selected channel and language.
3. **Choose an AI-Model for Chat** * The Chat channel now supports models from multiple providers: * **OpenAI Models :** Available OpenAI models include: * GPT-4 series models * GPT-5.x series models * GPT-5 Mini variants * GPT-5 Nano variants * Real-Time models * Codex (high-reasoning, code-focused model) and so on
* **Anthropic Models :** Selected Anthropic models are now available and can be used for chat-based experiences. * To configure a model: * Open the Chat Settings section. * Select an AI model from the available providers. * Choose the model that best matches your requirements for: * Response quality * Reasoning capabilities * Speed and latency * Cost efficiency * Available Anthropic models include: * Claude Haiku * Claude Sonnet * Claude Opus
{% hint style="info" %} **Note:** The Codex model is designed for code generation and advanced reasoning tasks. Because of its specialized purpose, it is available only for chat experiences and is not supported in voice channels. {% endhint %} 4. Choose an AI-Model for Voice * Switch to the **Voice** channel configuration.
* Select a language. * Choose between: * Standard (Non-Real-Time) Models * Real-Time Models
* **Standard Voice Models** * Most supported OpenAI and Anthropic models are available for voice interactions.
* **Real-Time Voice Models** * Real-time voice experiences currently support OpenAI models only. * Available options include: * GPT-4-based Real-Time models * GPT-4-based Real-Time 5 models * GPT-5-based Real-Time 2 models
{% hint style="info" %} **Note:** Real-Time 2 models leverage GPT-5 reasoning capabilities. While they may have slightly higher latency than GPT-4-based real-time models, they are optimized to maintain responsive conversational experiences. {% endhint %} 5. **Realtime model considerations** * Voice path forward * gpt-realtime-2 is identified as the recommended forward path for the voice pipeline. * No gateway code change required * The JIRA description states that making this model available does not require Channel Gateway changes. * Environment-specific availability * Realtime model availability may be controlled per environment. 6. **Understand Model Performance Considerations** * Different models can produce different results in terms of quality, reasoning ability, speed, and latency. * **OpenAI Models :** OpenAI models are generally recommended for: * Conversational AI * Customer-facing interactions * Voice experiences * Low-latency use cases * The Mini and Real-Time variants offer an optimal balance between performance, response quality, and speed. * **Anthropic Models** * Anthropic's Claude models, particularly Claude Opus and Claude Sonnet, are designed for advanced reasoning tasks. * While these models can provide strong analytical capabilities, they typically have higher response times compared to OpenAI models. * For some requests, response times may be significantly longer, making them less suitable for: * Real-time conversations * Voice interactions * Fast-paced customer service workflows {% hint style="info" %} **Recommendation:** Use OpenAI Real-Time or Mini models for production voice experiences and latency-sensitive chat applications. {% endhint %} 6. **Configure Model-Specific Settings** * Some models expose additional configuration options. * Review the settings available for the selected model. * Configure any reasoning or advanced parameters as required. * For GPT-5 reasoning models, specify the required `reasoning_effort` setting where applicable. 7. **GPT-5.x reasoning considerations** * **Explicit reasoning effort is required** * For GPT-5.x reasoning-capable models, calls must explicitly pass reasoning\_effort. * The default changed to none beginning with GPT-5.1. * **Risk of silent regression** * If flows rely on implicit reasoning behavior, output quality may regress without obvious errors. * **Reasoning-specific parameter handling** * Paths targeting 'chat' reasoning variants should remove: * Temperature * Other non-reasoning parameters that do not apply * **Higher reasoning levels for coding workloads** * reasoning\_effort=xhigh should be available for codex-class tasks. 8. **Test and Validate the Flow** * After selecting a model, thoroughly test the flow before publishing. * Run the flow using the selected model. * Validate: * Response quality * Accuracy * Latency * Streaming behavior * Tool-calling capabilities * Voice interaction quality (where applicable) * Compare model behavior across different scenarios to determine the best fit for your use case. 9. **Model Availability and Vendor Support** * New models can be added as they become available from supported vendors. * Currently supported vendors include: * OpenAI * Anthropic {% hint style="info" %} **Important:** Adding a model from an existing supported vendor is typically straightforward. However, supporting a completely new AI vendor requires additional platform integration and development effort before its models can be made available. {% endhint %} --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://docs.ixhello.com/ixhc2/integrations/expand-llm-catalog.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.