Feature Requests

For LLMs that support that parameter* Additionally, it would be useful to modulate the parameter via session_settings request

When using the Hume API, you can view an audio-visual timeline to track the latency between the request and response in real-time. Additionally, a detailed timeline is available in the logs or a list format, providing precise timestamps for each step in the process. For example: Message Received: The exact time when Hume receives the message from the client. CLM Handoff: The time when Hume forwards the message to the CLM First Response: The time when Hume receives the first response from the CLM and begins streaming it back to the client. Assistance Completion: The time when the interaction concludes, marked by the assistance end message.

Please add word boundary timing information (in milliseconds) to assistant_message and/or audio_output . This is to support real time video avatar animation. Additional nice-to-have: in addition to word boundary timing information, supply oculus visemes and their timings, as supplied by Azure TTS. Hume EVI2 voices with word boundary timing plus the existing prosody output would put Hume a LONG way ahead of the opposition in this space.

My suggestion is have average result of each emotion after every conversation. Also a chat summary is good to show to user. And after several conversations. I hope to have an emotion change graph to let user know their emotion change

The ability to perform custom actions during a call, such as transferring the caller, scheduling an appointment, or sending a text message.

Set up EVI to transition between different prompts or configurations based on the moment in the conversation, giving users more granular control over how it responds at specific points in the conversation. This can help This feature is currently under review - if you have more information about what you're looking for, please contact jeremy@hume.ai .

Configuration-level data retention controls rather than just account-level toggle for zero data retention would be really helpful. This would let organizations enable/disable data retention and training permissions on a per-user or per-configuration basis. Especially helpful for healthcare providers and enterprise clients who need granular control over AI data privacy settings to comply with regulations, manage user consent, and handle sensitive data appropriately. The current account-level toggle system requires organizations to either create multiple accounts (which disrupts workflows) or apply blanket settings that don't accommodate individual user preferences or varying privacy requirements across different use cases.

I'd like to be able to update the voice via API. For example, if doing a conversation and the end user would rather speak to a female, the only current way is by duplicating the configuration.

The EVI API currently has a maximum conversation duration of 1800 seconds (30 minutes). This feature request proposes the ability to extend this time limit. Two potential implementation options are: Increase the global time limit for all users Allow customizable time limits on a per-user basis

Once we add webhooks, it will be possible to send info to the webhook URL after each EVI call. This webhook could be linked to the developer's server, or used to do things like post slack messages to the user's channel. For example, some things we could send over the webhook at the end of EVI calls: Generated call summaries Main conversational topics (potentially with the major expressions per topic) Structured data defined by the user (e.g. action items, conversation issues) Voice metrics (e.g. avg number of words per turn for user and assistant, silence/pause length, etc) The call's audio recording Data on the user's expressions, e.g. spikes in frustration, averages, top 3 expressions, satisfaction score Other things Community question: What other information would be helpful to send over a webhook after EVI calls?