: We now support WebRTC, allowing users to connect their audio sources directly to Millis agents via WebRTC. This integration is perfect for a range of applications, including:

Learn more here: WebRTC Integration.

: We've also added SIP Trunking support. With this, you can easily upgrade SIP-based phone systems or call center platforms to connect with advanced voice agents. This includes popular systems like , and more. We support both and for dialing and routing calls. Reach out to us if you'd like to integrate!

: We've added the API, allowing you to obtain a that can be used instead of to define how Millis handles calls. With this, you can pass agent_id, override agent_config, apply temporary agent configurations, and attach metadata to each call, enabling per-call customization for both inbound and outbound calls.

: You can now override agent config (I.e. assign different agent voices) for each phone number, even if they link to the same agent, providing greater flexibility for personalized experiences.

: Enable the agent to decide when to terminate the call itself without waiting for the user. This feature is currently experimental and recommended for use with larger language models like GPT-4o.

: You can now tag phone numbers to better organize and manage your phone list.

: You can now include metadata when using the API.

: Metadata is already forwarded to both the Prefetch Data Webhook and End of Call Webhook. With the new option, agents can now use metadata directly during calls. Additionally, you can pass metadata parameters directly into function calls by using the same parameter names.

Learn more here: Metadata Documentation.

: You can now bring your own ElevenLabs API key, and we won’t charge you when using ElevenLabs as the voice provider. Plus, you can easily import all custom voices from your ElevenLabs account to use with your Millis agents.

: The trending voice list from ElevenLabs is now available in popular languages like English, French, German, Dutch, and more.

:

We’ve added Neets TTS, the on the market at just ! Neets.ai offers the most in the industry with decent quality and latency. Supporting over 80+ languages, we're excited to include Neets in our growing list of TTS voice providers. This addition unlocks even more use cases for conversational voice AI across diverse languages and markets.

We've integrated Deepgram TTS into Millis! From our testing, it offers decent quality with good latency and is a very affordable option for voice synthesis. Currently, it supports English only, but we're hopeful for more language options in the future.

: We're excited to expand language support with , the latest state-of-the-art STT model. We've added support for 40 more languages, including . In total, Millis voice agents now support !

: Also based on Whisper Large v3, this new experimental feature allows agents to auto-detect the language the user is speaking and respond accordingly. No more fixed language settings—agents can seamlessly switch between languages during a conversation.

: We've added a new option to handle delayed responses from custom LLMs. Agents can now automatically say predefined phrases if a response takes longer than expected. Configure it like this:

: You can now manage your agent's knowledge base via API! This includes the ability to upload files, add them to the agent, delete files, and more—giving you greater control and flexibility in managing your knowledge assets.

We’re excited to announce the addition of as a supported telephony provider on Millis! With this update, we now support a total of , further expanding our global coverage. Exotel is available in and is a popular option in , offering even more flexibility for your telephony needs.

: You can now initiate outbound calls directly from the dashboard, no API required! Plus, an outbound campaign system for mass users is coming soon.

: Agents can now switch languages mid-call if the user requests it (e.g., "Can you speak Spanish?"). Set this option in the agent config with: .

: If you know a custom LLM response will take time, you can stream an initial phrase for the agent to say first with the option , then process and stream the rest later. With agent will generate phrase audio right away to keep the conversation smooth and responsive.

: We've made further optimizations to ElevenLabs, ensuring even faster streaming for a better real-time experience.