When does Voxtral TTS available for our agents? ?

Hi Retell Team! First of all, the voice latency and naturalness of Retell are game-changing. I'm currently implementing Retell for high-end clinics and law firms in Brazil, and the feedback is amazing. However, I’ve noticed a recurring demand from my clients: Native WhatsApp Text Integration. In markets like Brazil, India, and Europe, WhatsApp is the primary communication tool. Clients often want to start a conversation via text and escalate to a voice call (or vice-versa) within the same context. My suggestion: If Retell could offer a native way to handle WhatsApp text messages (perhaps via a Twilio or Meta API integration) within the same Dashboard/Agent logic, it would create a true Omnichannel AI Agent. Benefits for Retell: Higher Retention: Users wouldn't need to jump between Retell and N8N/Make to handle texts. Unified Context: The agent would 'remember' what was said in text when the client calls. Market Expansion: Massive growth potential in countries where WhatsApp is the 'OS' of business. Keep up the great work!" @Joshua - Solutions Director @Oscar @ Cream Digital @User

Is there any sort of ETA on when GPT Realtime 1.5 will have the ability to add custom voices or receive an Australian voice?

It would be really nice if we had the ability to edit and/or view knowledge base's file contents I've been creating MD files for our Knowledge Bases and it would be really nice if we could just directly change the file contents without having to delete the file and then reading. This would then apply the changes across all agents and nodes that have this knowledge base attached to it I know it's not much but this would be very useful to see what knowledge bases are using. It would also be nice if you do crawl/scrape a site then to see the information that comes back. I presume when you do a scrape, its getting the information from the page and creating some sort of JSON, Markdown or Text file that the LLM can understand. TLDR: Ability to edit file contents of a knowledge base without having to delete the file and reattaching the editted version. Also; Ability to see the information that gets returned when you do a scrape/crawl of a site for a knowledge base

Hey Retell team! 👋 Feature request: Would love to see support for

gemini-3.1-flash-live-preview

— Google's native audio-to-audio (A2A) model released on March 26. Right now, when using

gemini-3.1-flash-lite

in Retell, the detail logs show the standard ASR → LLM → TTS pipeline. That makes sense since it's a text-based model. But

gemini-3.1-flash-live

is architecturally different — it uses the Gemini Live API over bidirectional WebSockets, processes raw PCM audio natively, and skips TTS synthesis entirely, which is exactly how you already integrated OpenAI Realtime. The potential benefits for Retell users would be significant: • Lower end-to-end latency (no ASR/TTS overhead) • Native barge-in support • Better acoustic nuance handling (pitch, pace, background noise) • 90 languages supported natively Is this something on the roadmap? Would be a game-changer for production voice agents. 🚀

Currently the inbound webhook agent_override supports overriding voice config, LLM model settings, and various call behavior fields — but not general_tools. Our use case: we have enterprise customers with customer-specific tool-call endpoints (e.g., a CRM lookup tool that only applies to one customer). Today we have to pre-register these tools on the LLM in the Retell UI. This works but has downsides: - Can’t dynamically compile tools on our end based on caller context - Can’t store/manage tool definitions in our backend with our custom system - Can’t dynamically attach only tools needed for the call - prevent context rot Proposed change: Allow agent_override.retell_llm.general_tools in the inbound webhook response (and create_phone_call API). This would let us dynamically attach tools on a per-call basis — same pattern asdynamic_variables but for tool definitions. Example response shape:{“call_inbound”: {“agent_override”: {“retell_llm”: {“general_tools”: [{“name”: “lookup_banker”,“type”: “custom”,“url”: “``https://example.com/api/tool-calls”``,“description”: “Look up a banker by name”,“parameters”: { … }}]}},“dynamic_variables”: { … }}} Ideally this would be additive (merged with existing LLM tools) rather than a full replacement, with an option to replace if needed. Official request: https://community.retellai.com/t/allow-tools-override-on-inbound-webhook-agent-override/2323

To make it quicker, give us the ability to rename agents from the index page. I've just had to go through all of agents to rename them with a prefix of 'OLD:'. This would've been quicker if we have the option to rename from the index page and also move them into a new folder to try and keep things organised. Also; it would be nice on the name you have FOLDER_NAME/AGENT_NAME. OR just a badge on the rows with the name of the folder they're in. https://cdn.discordapp.com/attachments/1493542115593158736/1493542115924381758/Screenshot_2026-04-14_at_10.21.14.png?ex=69df58fd&is=69de077d&hm=c41cfae981751162f024177c8bf310d1c1996a1292cd438284e2f6fac32bb356&

Currently with all the new features Retell is pushing out. The agents behaviour can change as unintentional side effect. We have started getting this transition.

previous node: GLB-CRTR-CNV

new node: GREET-CNV
transition type: interrupt_go_back

This has caused our conversation flows to be worse, as they are not carrying out the same flow. This seems to be a new transition update which would be great if we could manually onboard or be alerted to the new versions and then correct any behaviour changes, when we manually upgrade

Voices are top tier and it's OPENAI protocol, easy add...FEAUTURE REQUEST

I've created a node function under my Welcome node called extract_customer_context. I can’t use it for transitions in the fine-tuning examples section, but I can use my custom-built functions there. Could you also add support for node functions in this section? https://cdn.discordapp.com/attachments/1505941775150350336/1505941775410528277/retell_.jpeg?ex=6a0c7512&is=6a0b2392&hm=20c30c0bf38387cfc0e303bc191301abe7714d958457cde3cdca77ccdf2cc7e2&

Gemini Models are crazy good in performance. Please be fast in adding this new LLM.

Ive been doing some work. We have an agent for our inbound calls and we're going to be doing outbound calls using a different agent. We're currently using another number for the outbound calls because we dont want to hurt the reputation on the inbound caller number as I know phone number health is important (years of working with a call centre) What would be really useful in the call_analyzed is that we can see under the tool_calls key that there is an array on various tools that were called. When it's in agent swap, you get something like

{
  "event": "call_analyzed",
  "call": {
  ...
  "tool_calls": [
    {
      "tool_call_id": "agent_swap_bf8a65",
      "name": "agent_swap",
      "type": "agent_swap",
      "start_time_sec": 0.072,
      "latency_ms": 21.000000000000004,
      "success": true
    }
  ],
  ...
}

This is great but could we add the agent's id (agent_1234567) and the agent's name (Inbound Agent). This will let us know which agent is transferred it to. Result:

{
  "event": "call_analyzed",
  "call": {
  ...
  "tool_calls": [
    {
      "tool_call_id": "agent_swap_bf8a65",
      "name": "agent_swap",
      "type": "agent_swap",
      "start_time_sec": 0.072,
      "latency_ms": 21.000000000000004,
      "success": true,
      "agent_id": "agent_1234567",
      "agent_name": "Inbound Agent"
    }
  ],
  ...
}

If you need a call_id, I can more than happy supply with you one.

What is being discussed in the image. When an retell agent is detecting an interruption to the converstation, in converstional nodes. Instead of repeating the entire node, or repeating the node from the start. It should be able to acknowledge the interruption, and let users customize on how the agent should deal with these cases. https://cdn.discordapp.com/attachments/1514791145379463268/1514791145891299418/image.png?ex=6a2ca6b0&is=6a2b5530&hm=575e0ae3edcbfc61e50df77fa9b86f321d9220ac3c578d6fc204de9a9e4a89e1&

I just saw, you have Contacts which is just all the information about a phonenumber (name, number email, dnc etc), it would be really nice to have an api endpoint (api/v1/contacts/{contact_id} GET, POST AND PATCH). Then we could set the dnc and other attributes like the name via API This would be pretty easy to implement as the Contact looks super simple and its just a way to store all the information about the user from a call/chat

We would like to get the new GPT 5.6 language model added as soon as possible, considering how higher quality, PLEASE BRING IT AS SOON AS YOU CAN AS MY CLIENTS ARE REQUESTING IT

The new LLM is really good, can we please get it please

We tested a few Cantonese pronunciation that we found it is not working as expected for IPA and Jyutping work as charm in our local test while we could not find related setting in retell.

We need the GPT Live 1 mini language model. It sounds really amazing and it will help us build a better AI receptionist

Would be very nice, if you implement Inworld AI as a new TTS Provider. The Voices are clean and expecially the european voices are very good. https://cdn.discordapp.com/attachments/1531641937709367490/1531641937956700221/image.png?ex=6a69f436&is=6a68a2b6&hm=754b1ed998fb5da1ecbf4921ee1347b2e66915fc4c70adb2ad057fa3fc458c64&

The Gemini Llms are Crazy good for pur use cases. Also the gpt Luna Price Updates by openai does Affect your prices as well? Thanks in advance for your answer and work to integrale all the good Models for our Voice Agents.

We'd like to use Linq infra for calls, but mostly SMS. They have iMessage features. Could you integrate with them? We don't have much options in regards to SMS with Twilio and its costly.

We’d like to request a feature that allows the agent to automatically deliver a closing spiel after 15 seconds of voice inactivity and then end the call immediately after the closing spiel is completed. Currently, after the closing spiel is triggered due to inactivity, we have to wait for an additional 20 seconds before the call automatically ends. Desired behavior: 15 seconds of voice inactivity → Agent delivers closing spiel → Call ends immediately after the closing spiel This would help avoid the additional waiting period and allow calls with inactive callers to end more naturally and efficiently.

Always on the Hunt of the newest, best Models Please add this new Sonic Model TTS.

We have a system when you call, our agent says 'Welcome to COMPANY_NAME, please press 1 for X, press 2 for Y or press 3 for Z'. It would be nice if that voice was different to the agent's voice because it gives the impression options and the agent you speak to are AI if you get me? It would just be nice to be able to say 'This node will use this voice' .

could it be possible, to create, delete and manage Workspaces via API? based from an existing user and maybe his IDs or credentials. it would be also super helpful if we could add billing info via API into the workspaces.

You guys at retell are Always fast in integrating the newest and best Models for Building Voice agents. So the new gpt-live-1 Model is Some type of a new possible Architekturen for Building with a frontend and backend Model? It would be super important to Test it in your harness in some production environment cause its billing is based on scale with 0,05$ per Minute which is super Interesting and attractive for us and our customers. So please make it possible to integrate gpt live 1 in sort of a say to Build Voice agents on top of this Model in retell ai ❤️

Guys, the pricing of this new llms from the side of openai is criminal cheap. Massive step forward for our customers, please be fast with the Integration of this new Models.

TLDR; Add JS Event Listeners, more styling customisation and focusing inputs, attach to window in JS and add bouncing chat bubble for better UX I've been changing our Voice Agent to a live chat so we can add live chat to our website. I've come across a few things that I think would be good nice to haves 1. When you type a message, submit it, it dissables the input but as soon as the AI has replied, it un-focuses the input, it's a little annoying. I could maybe understand un-focusing on Mobile due to the keyboard taking up majority of the screen's real-estate. 2. Javascript Events. I.e: When a message is sent (from user or AI), you could trigger an event. When the conversation has ended, you could trigger an event (I know you have webhooks for this but it's just an example), when the live chat window has been opened etc. 3. Give us more styling options. I get it you want to keep it simple but it's really annoying to try and change the size of the window. I want it to be more full screen of which I have to make a listener, wait for something to be added to the DOM and then change the shadow-roots CSS. It's just very long winded and I feel like this could be so easily changed just via variables or just allowing us to insert a style block to override the defaults (whether thats via a data attribute or a javascript event listener). 4. Attach the live chat to the window in JS so we could have window.retellAiLiveChat Or something and then we can just call``` window.retellAiLiveChat.open()``` just to give us more options and functions that we could call 5. Bouncing Eclipse when AI is thinking just to let the end user know something is happening, better UX

Small changes seem much bigger and harder to comb through when 95% of them are just x, and y changes because of the auto layout button.

Title says it all (again) - Sorry for spam. But these new (mini) models are getting better with every new release and they are affordable in real prod. environments. Would be great to have the possibility to use this model inside of retell.