Amelia 7.1 advances our agentic AI platform with major improvements in four key areas: accuracy, latency, transparency, and user experience. This release delivers:

• More accurate and flexible AI agents for increased control and better end-user experiences
• Faster conversations with reduced latency on responses
• New UI visualizations for more intuitive end‑user interactions
• Expanded control and transparency with deeper insights into agent performance and history

These updates not only significantly elevate the conversational experience for end users, but they also provide platform managers with more intuitive tools, better usability, and richer data to optimize AI agents.

More accurate and flexible AI agents

AI agents can be built with greater flexibility and are now more context‑aware, thanks to process‑specific instructions, improved knowledge retrieval, and fine‑tuned matching algorithms. These updates make every AI agent response more precise and adaptable to user needs.

For illustrative purposes

Reduced conversational latency

Conversations feel faster and more natural with reduced response latency, through upgrades like real‑time response streaming, optimized built-in prompts, improved retrieval, and streamlined orchestration. This is especially impactful for voice channels.

For illustrative purposes

New UI visualizations

We’ve added adaptive card carousels in chat, turning lists of options (e.g., itineraries, hotel listings, appointment slots) into swipeable cards for a more intuitive, engaging browsing experience.

For illustrative purposes

Increased control and transparency

Support for custom on‑premise LLMs and new scripting tools to access AI agent execution histories give enterprises deeper control, visibility, and compliance in complex environments.

Specific 7.1 platform updates

Below, we’ll detail every important change in 7.1 you should know about.

UI enhancements

Visual cards within chats — for things such as flight itineraries, hotel listings, and appointment options — can now be viewed in a carousel format for simpler browsing. This enables more seamless, interactive, and visually appealing displays of information compared to previous stacked visuals or large buttons.

Agentic enhancements

You can now add detailed, scenario‑specific instructions to AI agents. This means that in addition to developing overall AI agent directions, you can now create separate instructions for each type of task it might handle — for example, scheduling, rescheduling, canceling, or updating an appointment. Process-specific instructions are still activated dynamically as required to achieve the user’s goal, making agents more organized, accurate, and efficient when managing different processes.

We’ve improved our matching algorithm and knowledge segmentation to allow for internal tagging and filtering of documents during retrieval. These enhancements enable more precise, context-aware filtering (for example, by channel, such as voice) for more accurate and relevant AI agent responses.

Amelia can now deliver answers as they’re being generated — word by word — so users don’t have to wait for the full response. This makes conversations feel quicker and more natural, especially in real‑time channels like voice.

To reduce the time taken for orchestration (especially in simpler scenarios), we implemented optimizations like parallelizing tasks and logic changes to cut down processing time.

Optimizations to built-in prompts enhance efficiency and improve the accuracy and response times of Amelia and other tasks like entity extraction and intent detection.

We added script-based services to enable access to agent execution information within scripts. This feature allows platform managers to manipulate and store this data, potentially for custom metrics or conditional checks, enhancing data transparency and enabling deeper processing.

Enterprise knowledge enhancements

Amelia now supports bulk uploads of up to 100 enterprise knowledge sources at a time — a significant improvement from the previous one-source-at-a-time limitation — providing significant management efficiencies.

Amelia can now process knowledge from a visual standpoint, extracting critical information from items such as images or tables within PDFs for more complete knowledge ingestion.

LLM enhancements

We have expanded options to use your own custom AI models (“LLMs”) for intent detection. This means organizations can choose the models that work best for them, including running everything on their own systems on-premises, instead of in the cloud — ideal for institutions that need full control over their data and operations.

With this release, you can adjust how creative or precise the AI’s responses are.

• Temperature and Top‑P are new controls that let you choose between responses that are more imaginative (good for brainstorming) or more focused and predictable (good for accuracy).
• A “seed” setting can also be used to make the AI’s answers more consistent, so the same question gets the same response each time if desired.

These options give you more control over how the AI behaves, depending on your needs.

Voice enhancements

Our upgraded voice gateway both lowers costs and improves streaming capabilities in all voice interactions, improving latency and efficiency for SoundHound AI’s customers. 

This capability ensures that text responses are streamed efficiently to text-to-speech providers for voice channels, with a logic implemented to flush messages at the end of each sentence to maintain natural-sounding speech.

Usability improvements

Users have the option to disable or toggle components within the Amelia platform — primarily those which have been deprecated — to streamline the UI.

Why 7.1 matters

Amelia 7.1 is a step forward in continuing to make AI agents more natural to talk to, more adaptable to enterprise needs, and more transparent for platform owners. By combining faster performance, greater control, and richer insights, 7.1 sets the stage for even more advanced agentic AI capabilities.

Explore these new features in your platform today, and stay tuned for what’s next.