Deepgram Release Notes

Follow

174 release notes curated from 249 sources by the Releasebot Team. Last updated: Nov 18, 2026

Get this feed:
  • Nov 17, 2026
    • Date parsed from source:
      Nov 17, 2026
    • First seen by Releasebot:
      Nov 18, 2026
    • Modified by Releasebot:
      Nov 18, 2026
    Deepgram logo

    Deepgram

    November 17, 2026

    Deepgram improves Nova-3 speech-to-text models for German and Hindi.

    Improved Nova-3 Models for German and Hindi

    Speech-to-Text

    Original source
  • Nov 16, 2026
    • Date parsed from source:
      Nov 16, 2026
    • First seen by Releasebot:
      Nov 17, 2026
    • Modified by Releasebot:
      Nov 17, 2026
    Deepgram logo

    Deepgram

    November 16, 2026

    Deepgram expands Nova-3 speech-to-text to support 36 additional languages.

    Nova-3 Adds 36 Languages: Abkhazian, Adyghe, Akan, Albanian, Amharic, Azerbaijani, Basque, Bodo, Central Kurdish, Divehi, Eastern Mari, Galician, Hausa, Icelandic, Kabardian, Kabyle, Kalenjin, Kashmiri, Kinyarwanda, Konkani, Kyrgyz, Luganda, Luo, Malayalam, Northern Kurdish, Sindhi, Sinhala, Swahili, Taita, Tswana, Uyghur, Uzbek, Venda, Welsh, Yoruba, and Zulu

    Speech-to-Text

    Original source
  • All of your release notes in one feed

    Join Releasebot and get updates from Deepgram and hundreds of other software products.

    Create account
  • Nov 13, 2026
    • Date parsed from source:
      Nov 13, 2026
    • First seen by Releasebot:
      Nov 18, 2026
    Deepgram logo

    Deepgram

    November 13, 2026

    Deepgram updates its CLI with audio file output for dg speak, non-interactive setup confirmation, and clearer dg listen output.

    Deepgram CLI 0.5.0: dg speak gets --out-file, dg init needs --yes outside a terminal Changes how dg speak saves audio, how dg init confirms setup outside a terminal, and what dg listen -o prints. Developer Tools

    Original source
  • Nov 12, 2026
    • Date parsed from source:
      Nov 12, 2026
    • First seen by Releasebot:
      Nov 12, 2026
    • Modified by Releasebot:
      Nov 18, 2026
    Deepgram logo

    Deepgram

    November 12, 2026

    Deepgram CLI 0.4.1 adds npm installation, removes the dg mcp SSE transport, and improves Nova-3 speech-to-text models across several languages.

    Deepgram CLI 0.4.1: npm install, dg mcp drops SSE Adds an npm install path, drops the dg mcp SSE transport, and narrows what a project config file can set. Developer Tools Improved Nova-3 Models for Catalan and Japanese Speech-to-Text Improved Nova-3 Models for Arabic, Armenian, Bengali, Bulgarian, Danish, English (Australia, New Zealand), Greek, Gujarati, Hebrew, Nepali, Slovak, Tagalog, and Urdu Speech-to-Text

    Original source
  • Nov 10, 2026
    • Date parsed from source:
      Nov 10, 2026
    • First seen by Releasebot:
      Nov 11, 2026
    Deepgram logo

    Deepgram

    November 10, 2026

    Deepgram introduces a self-hosted offering, with no further release details provided.

    November 10, 2026

    Self Hosted

    Original source
  • Similar to Deepgram with recent updates:

  • Nov 10, 2026
    • Date parsed from source:
      Nov 10, 2026
    • First seen by Releasebot:
      Nov 11, 2026
    Deepgram logo

    Deepgram

    Get your Spanish contact center ready for Ley 10/2025 with Nova-3

    Deepgram adds Basque and Galician transcription to Nova-3 alongside Catalan, with EU endpoint and self-hosted availability for live and recorded calls. Catalan support also arrives in early access for Flux speech models, helping customer service teams serve Spain’s co-official languages.

    Nearly 20 million people in Spain live in a region with a second official language: Catalan, Valencian, Galician, Basque, or Aranese. When they call customer service, many of them can still only get help in Spanish.

    That is about to change.

    Under Ley 10/2025, a customer in one of those regions can ask to be served in its co-official language, on every channel, and companies have until December 28, 2026, to make it work inside the same three-minute answer targets as every other call. The law applies whether customer service is in-house or outsourced, so if you run, build, or power customer service lines for essential services or large consumer brands in Spain, you will be responsible for meeting that mark.

    Deepgram Nova-3 now has dedicated Basque and Galician models next to Catalan, on our EU endpoint and in self-hosted deployments, so you can understand those callers through a single API.

    What you need to know

    • Ley 10/2025 covers every utility, telecom, passenger transport, postal, and financial services company in Spain, whatever its size, plus large consumer-facing companies, and they have until December 28, 2026, to adapt.
    • The law applies whether customer service is run in-house or by a third party, so providers of contact center services, platforms, and voice agents to these companies will be expected to serve callers in their co-official language and reply in the language they used.
    • Nova-3 now transcribes Basque (eu), Galician (gl), Catalan (ca), and Valencian (via the Catalan model) for both recorded and live calls on api.deepgram.com, on the EU endpoint api.eu.deepgram.com, and as self-hosted model files.
    • You switch languages with one query parameter, and your existing API keys keep working.

    This post covers what the law asks of your automated channels, what Deepgram covers now, and what is still on the way. It is not legal advice, so check your own obligations with your legal team.

    Understand what Ley 10/2025 requires

    Ley 10/2025 applies regardless of the communication channel. For anyone running voice, that puts every call in scope, whether a person, a phone menu, or a voice agent answers. The full text of Ley 10/2025 has the details, and these are the parts that matter most to a contact center:

    • Language. A customer in a region with a co-official language can ask to be served in it, as long as the company operates there, and the reply has to come back in the language they used. The law applies on every channel, so plan for voice agents and phone menus to work in these languages too. Co-official languages are set by each region’s statute of autonomy and include Catalan, Valencian, Galician, Basque, and Aranese. (Aragonese and Asturian are protected regional languages, but not co-official, so they are not covered under Ley 10/2025.)
    • Answer times. Phone wait times are capped at three minutes on average. Callers over 65 and people with disabilities get priority, and companies can’t cut off a call because the wait is long.
    • A human on request. A customer can always ask to speak to a person, so your voice agent needs a clean handover path, and 95% of requests to be transferred to a person must take place under three minutes. If an unhappy customer asks to speak with a supervisor and the transfer can’t happen within three minutes, the company can call back the same working day.
    • Resolution deadlines. Queries, complaints, and incidents must be resolved within 15 working days, unless sector rules set a different deadline. Billing disputes must get a response within five days, and questions about outages of essential services such as energy, water, or telecoms within two hours.
    • Records and audits. When a customer consents, calls must be recorded and the company can keep either the recording or its transcript until the customer is informed of the outcome. Every year (or every two years for smaller companies), companies have to hire an auditor accredited by ENAC, Spain’s national accreditation body, to check that the customer service measurements they publish are within 5% of real values.

    Meet each requirement with Nova-3

    The transcript Nova-3 produces on every call helps you meet these obligations and show that you met them.

    Here is how it helps with each requirement:

    • Language. Transcribe Catalan (Valencian included), Basque, and Galician calls in real time, so your voice agent understands the caller and any person who takes over sees what was already said.
    • Answer times. Route callers using these languages straight to the right queue, which helps calls stay inside the three-minute target and makes sure those who say they’re over 65 or have a disability get to the priority queue.
    • A human on request. Catch the request for a person in any of these languages the moment it’s said, and pass the live transcript to the agent so the customer doesn’t have to repeat themselves.
    • Resolution deadlines. Use the transcript to tag every call by topic and flag outages, billing disputes, and complaints as they happen, so each one starts on the right clock.
    • Records and audits. Give your QA and compliance teams a transcript of every call that shows how each request was handled and in which language, and use it as evidence for the annual audit.

    Keep call audio in the EU or on your own servers

    Though Ley 10/2025 doesn’t say where call audio has to be processed, most Spanish enterprises want call audio to stay in the EU. The Deepgram EU endpoint at api.eu.deepgram.com processes requests inside the EU and never falls back to another region, and the Basque and Galician models are available there for both batch and streaming. You keep the same API key and change two things: the host and the language parameter.

    This request sends a recorded Basque call to Nova-3 on the EU endpoint. Grab a key from a free Deepgram account first.

    curl -X POST "https://api.eu.deepgram.com/v1/listen?model=nova-3&language=eu" \
    -H "Authorization: Token $DEEPGRAM_API_KEY" \
    -H "Content-Type: audio/wav" \
    --data-binary @basque_call.wav
    

    For a live call, open a WebSocket to the same host and pass language=gl for Galician or language=eu for Basque. A typical streaming request looks like this:

    wss://api.eu.deepgram.com/v1/listen?model=nova-3&language=gl&encoding=mulaw&sample_rate=8000
    

    Call Deepgram from your own server and send the key in the Authorization header, as in the examples above. Never put it in browser code or in a URL, where it ends up in logs and history.

    The regional endpoints guide can also show you the SDK setup for Python and JavaScript if you would rather use an SDK client (instead of cURL requests) for the EU host.

    Additional questions

    Can I run the Basque and Galician models on my own infrastructure (self-hosted)?

    • Yes. The Nova-3 Basque and Galician models ship as model files for self-hosted deployments, in batch and streaming versions. Ask your Deepgram account team to add them to your license, and your existing self-hosted setup can serve language=eu and language=gl alongside the Nova-3 models you already run.

    How do I route a call to the right language?

    • Route on what you already know: the region of the phone number, the language the customer picked in the IVR, or their saved preference in your CRM. Pass that as the language parameter when the call starts. A caller in Bilbao who chose Basque gets language=eu, and a caller in Santiago who chose Galician gets language=gl.

    What’s the best way to test Nova-3’s capabilities?

    • Pick the region where most of your co-official language calls come from, point a test batch of recordings at Nova-3 on the EU endpoint with the right language code, and compare the transcripts with what your agents heard. Create a free Deepgram account to get an API key.

    Help shape our roadmap for Spain’s languages

    At the time of writing, Nova-3 covers Catalan (ca), Basque (eu), and Galician (gl), and Valencian is served by the Catalan model. For callers who switch languages mid-call, Nova-3’s multilingual mode (language=multi) listens for 55 languages at once, including Spanish, Catalan, and Galician, and tags each word with its language, so you don’t have to pick the language up front. Multilingual mode runs on our cloud API today, with self-hosted support on the way. Basque isn’t one of its languages yet, so Basque calls run on their own with language=eu.

    For voice agents, both Deepgram Flux STT, our conversational speech model, and Deepgram Flux TTS, our text-to-speech model, now support Catalan in early access. For Basque and Galician, anything your system says back will still need recorded prompts, a person, or your current voice provider. Tell your account team which languages and channels matter most to you, because that feedback sets our priorities.

    Original source
  • Nov 4, 2026
    • Date parsed from source:
      Nov 4, 2026
    • First seen by Releasebot:
      Nov 6, 2026
    Deepgram logo

    Deepgram

    Deepgram Marketplace: Voice AI’s open ecosystem

    Deepgram launches an open marketplace where voice AI builders can discover vetted partners across models, platforms, testing, audio, telephony, security, avatars, and health signals. It aims to make building a best-of-breed voice stack easier while preserving provider choice.

    Voice AI belongs on an open foundation. Teams building voice agents need the freedom to choose the best providers for each layer of their stack, and the flexibility to swap them out when something better ships. That openness breeds innovation, but juggling multiple vendors or even finding the right ones in the first place, can quickly undermine it.

    A team putting a support agent on the phone might need one company to test how it handles interruptions, another to clean up calls from noisy kitchens, another to connect to the phone network, and another to keep card numbers safe. Each one starts a new search with a vendor to vet, a contract to sign, and a bill to manage. Multiplied across every part of an agent, this kind of overhead has pushed developers toward closed systems for years, even when it meant locking into someone else’s choices.

    Deepgram Marketplace, aunching today, offers another path. It provides the flexibility of an open ecosystem without the friction of sourcing every piece yourself. Builders can find the capabilities a production voice agent needs beyond Deepgram in a single catalog and later, through shared credits, eligible customers will be able use their Deepgram commitment on partner products with one contract and monthly statement.

    Why a marketplace belongs in an open strategy

    We believe the future of voice AI has to be open source, and we're building toward it one step at a time. The first was bringing the Daily team and Pipecat into Deepgram and promising to keep the framework open source and vendor neutral. The marketplace is the next. It carries the same openness from the code to the providers developers buy from, as an open voice AI marketplace built to make choice easy and accessible.

    “Voice is quickly becoming the primary way people interact with software, and the builders creating that future need a place to find each other. Joining the Deepgram Marketplace puts Coval alongside the companies shaping how voice agents are built and deployed. Together, we're making it faster for developers and enterprises to go from idea to production-ready voice experiences, whether that's a customer service agent, a clinical assistant, or something no one has imagined yet. We're proud to help grow this community and can't wait to see what it builds."

    Brooke Hopkins
    Founder & CEO | Coval

    The catalog covers a range of voice capabilities and partner types, from products and integrations to service partners, so builders can look across the ecosystem and find what fits their needs. We review every listing before it goes live, which makes the marketplace a trusted source of options for anyone building a voice AI stack. Partners set their own prices and support their own products.

    “Giving developers access to specialized voice AI models helps them build voice experiences that solve real-world challenges. We’re excited to join the Deepgram Marketplace and help more teams bring natural, real-time voice experiences to their customers.”

    Robert Schoenfield
    EVP Licensing and Partnerships | Krisp

    This is what a level playing field looks like. You can see which partners integrate with Deepgram, and you're free to choose ones that don't. When options are easily available, nobody gets pushed into a choice they don’t want to make, and the best products earn their place on merit. That puts a best-of-breed voice AI stack within reach of every builder.

    Who's in it at launch

    The Deepgram Marketplace is launching with 19 partners across 8 categories, including:

    • Cartesia and Inworld in voice models
    • Vapi and LiveKit in voice agent platforms
    • Bluejay, Cekura, Coval, and Hamming in testing and evaluation
    • ai-coustics and Krisp in audio enhancement
    • Twilio, jambonz, Vonage, and UniSpeech in telephony and voice platforms
    • Fortanix and SottoAI in security and trust
    • LemonSlice and Tavus in avatars and video
    • thymia in health signals

    If you build for voice AI developers and would like to be included, you can submit your product for review and listing through the Deepgram Marketplace partner program.

    Keeping voice AI open

    I’ve spent my time at Deepgram working with partners because voice AI advances fastest when we all push the frontier together. Our commitment to the voice AI community is to keep moving the field forward with our own products and to keep the marketplace open, so innovation and healthy competition happen on equal terms. That makes voice AI better for everyone building it, Deepgram included.

    Browse Deepgram Marketplace.

    Original source
  • Oct 29, 2026
    • Date parsed from source:
      Oct 29, 2026
    • First seen by Releasebot:
      Oct 30, 2026
    Deepgram logo

    Deepgram

    Deepgram is acquiring Daily, makers of Pipecat, to offer the first complete, production-ready, open stack for conversational voice AI

    Deepgram introduces the open Real-Time Context Standard with Pipecat, helping voice AI agents retain audio cues like tone and speaker identity across models so they can respond more naturally and ask for clarification when needed.

    At today’s inaugural Deepgram Speak event, the company will introduce the Real-Time Context Standard, an open protocol that carries context across an entire conversation, developed in partnership with Pipecat

    SAN FRANCISCO — October 29, 2026 — Deepgram, the voice AI infrastructure company built for the demands of real-world conversation, today announced that it is acquiring Daily, creators of Pipecat, the world's most popular open-source voice agent framework. Pipecat’s open, vendor-neutral ecosystem puts open source at the heart of Deepgram’s strategy to help developers build production-scale voice AI. Daily’s global real-time network infrastructure connects end users to voice agents with ultra-low latency and high reliability. With Pipecat and Daily underpinning Deepgram’s new Voice Agent Platform, alongside Deepgram’s speech models, Deepgram will offer the first complete, open, production-grade platform for conversational voice AI, so developers can build and scale reliable agents while keeping control over their architecture and model choice.

    The Pipecat framework will remain fully open source and free to all developers under its existing BSD V2 license. Together with many of the organizations that have contributed to and use Pipecat, the companies are establishing an independent Pipecat Foundation. Deepgram will invest in Pipecat’s continued development and community while keeping all of Pipecat open source and vendor neutral, letting developers choose and combine models, tools, network transport, and APIs from different providers. Following the acquisition, the Daily team will join Deepgram.

    “Keeping Pipecat open was fundamental to our decision to join Deepgram,” said Kwindla Hultman Kramer, co-founder of Daily, who joins Deepgram as the new Head of Open Source. “Deepgram has always been the voice AI company that takes production seriously, and their commitment to keeping Pipecat open is rooted in a belief, which I share, that the future of voice depends on developers being free to combine the best ideas from across the industry and build on each other’s work.”

    “Kwin and the Daily team deeply understand what it takes to keep a live conversation working reliably, and Pipecat gives developers the freedom to build with the models they choose,” said Scott Stephenson, CEO and co-founder of Deepgram. “Acquiring Daily allows us to bring that expertise together to tackle the whole voice system and give developers a faster path from prototype to production.”

    Connecting all of the pieces of the voice AI stack is the first part of the job. Making sure those pieces share context and the information needed to carry a conversation forward is a separate challenge.

    Introducing the Real-Time Context Standard: an open protocol that keeps the conversation flowing

    At Deepgram Speak today in San Francisco, Stephenson will also introduce the Real-Time Context Standard (RCS), an open protocol co-developed with the Pipecat team. RCS addresses a gap in today’s voice AI systems: the words get passed from one model to the next, but the rich audio context around them, like tone, who’s speaking and what’s happening in the background, gets lost along the way.

    RCS defines three new types of metadata: context from the audio, such as a whisper or an unclear word; instructions for how a language model should use those signals; and guidance for how the response should sound. That lets an agent ask for clarification when it didn’t hear clearly or respond quietly to someone whispering. Adapters translate those signals and delivery instructions for each model, so developers can use the same approach with models from different providers.

    “Too many teams treat voice AI as a feature to add late in the roadmap. In reality, it’s a distinct infrastructure discipline, with its own demands around latency, accuracy, and scale,” said Stephenson. “We’ve all seen a voice AI demo that makes the future feel close, only to fail the moment it’s needed. The real challenge is making that experience reliable when people actually depend on it. Our Real-Time Context Standard addresses what gets lost along the way.”

    The Real-Time Context Standard is open and extensible. Deepgram invites developers, model providers, and teams building voice applications to help shape RCS by testing it in their own systems, sharing feedback and contributing to the specification.

    Deepgram’s acquisition of Daily and the introduction of RCS set the stage for Deepgram’s broader platform vision: a complete set of tools that, together, make it easy for developers to build state of the art, human-like, enterprise-ready voice agents and conversational AI applications.

    Later today at Speak, Chief Product Officer Ed Anuff will preview the forthcoming Deepgram Voice Agent Platform, which brings Deepgram’s models and Pipecat’s open infrastructure together to give developers an integrated path to production while preserving model choice. Anuff will also introduce two new models built for live conversation, along with expanded language support and a Global API designed to improve connection reliability and reduce latency.

    Learn how to build with Real-Time Context Standard in Pipecat here and try the demo here. For updates on the Deepgram Voice Agent Platform, please contact us. Full technical details, demos and availability for the remainder of today’s product announcements will be published on Deepgram’s blog following Anuff’s keynote.

    About Deepgram

    Deepgram is bringing the world’s most natural interface—voice—to the technology we use every day. Its speech models and real-time infrastructure enable AI systems that keep up with the way people actually speak, powering conversational voice AI in production for more than 5,000 organizations globally. The Deepgram Voice Agent Platform brings together speech-to-text, text-to-speech and speech-to-speech capabilities, with voice-native foundation models available through cloud APIs or self-hosted and on-premises deployments. Built for accuracy, low latency and cost efficiency, Deepgram makes high-quality voice experiences practical to build and operate at scale. More than 650,000 developers have used Deepgram to process more than 200,000 years of audio. Learn more at deepgram.com, explore developer documentation, or follow Deepgram on X and LinkedIn.

    About Daily / Pipecat

    Daily built the real-time infrastructure layer that enterprise voice and video AI requires — a global network of 75 points of presence engineered for the latency and reliability demands of live AI. Pipecat, Daily's open-source voice agent framework, is the most widely-used orchestration layer for production voice AI, with integrations across every major model provider, telephony system, and deployment environment. Pipecat is maintained at pipecat.ai and on GitHub under BSD V2 license.

    Media Contact

    US: LaunchSquad for Deepgram
    [email protected]

    EMEA: Tyto PR for Deepgram
    [email protected]

    Original source
  • Oct 29, 2026
    • Date parsed from source:
      Oct 29, 2026
    • First seen by Releasebot:
      Oct 30, 2026
    Deepgram logo

    Deepgram

    Voice agents that keep the thread

    Deepgram introduces the open Real-Time Context Standard and voice-focused Think-Live and Think-Fast LLMs, while expanding multilingual speech support and launching a new U.S. endpoint. Think-Live is generally available, Think-Fast is in early access, and the Voice Agent Platform is still in development.

    Deepgram's CPO on the Real-Time Context Standard, the Think-Live and Think-Fast LLMs, and why we're building the Voice Agent Platform on Pipecat.

    I've spent most of my career on new interaction layers, and on the plumbing underneath them that lets software talk to software. The web needed HTTP and HTML before it could change how people found information. Mobile needed APIs before apps could reach people wherever they were. Each time, the interface everyone noticed was only as good as the connections nobody saw.

    Voice is next, and it's hitting the same wall. I've been saying for a while that voice is the new keyboard, because talking is how people actually explain what they need. We add context as we go, we back up, we correct ourselves. Most voice agents still can't keep up with that. They talk over you, go quiet while they look something up, and if you change your mind halfway through, they charge ahead with the first version of what you said. Using one is a lot like filling out a form out loud.

    The models themselves have gotten remarkably good over the last few years. What holds voice back now is mostly plumbing, which happens to be the part of the stack I find most interesting. This morning at Deepgram Speak, I walked through what we're doing about it, and I want to go a little deeper here.

    We announced the Real-Time Context Standard (RCS) for voice, an open standard for carrying meaning between the parts of a conversational agent, which we developed with Pipecat. Pipecat is now part of Deepgram, and I'll come back to why that matters. We also introduced two LLMs built for voice: the Think-Live LLM, for the conversation itself, and the Think-Fast LLM, for the work behind it.

    Losing everything but the words

    Most voice agents I see today are three models in a row. Speech-to-text feeds an LLM, the LLM feeds text-to-speech, and at each handoff almost everything except the words gets thrown away. The audio carries a caller whispering, a sigh, a word that barely came through, and the LLM receives a plain transcript. The LLM may decide the reply should be gentle and brief, and the text-to-speech model receives a plain string with no idea how to say it.

    We had a version of this problem with APIs, too. Two systems could connect perfectly well and still misread each other, and what fixed it was agreeing on a contract. RCS is a contract for everything beyond the words, which is often a big part of what's being communicated. It adds three fields to the chat completions format most LLM APIs already use. One carries how the person sounded, one is where you tell the agent what to do about it, and one tells the text-to-speech model how the reply should be spoken.

    If you’ve built with tool calling, the shape will look familiar, and that’s deliberate. Tool calling gave developers a standard way for a model to ask an application to do something, like look up an order, and MCP turned that into a shared standard any app or model can use. I’ve written before that MCP is starting to look like the new API. RCS applies the same idea to how people sound, and to how agents should sound back.

    The standard itself doesn't depend on any vendor, so any framework can implement RCS with any model. It's open for comment, too. Context only helps if every component in the pipeline can read and write it, whoever made that component. If you build speech models, LLMs or frameworks, we’d like you implementing RCS and telling us how to improve it.

    Giving each LLM one job

    It took us a while to say this plainly, but a voice agent's LLM has two jobs, and they pull in different directions. The first is managing the conversation. In an instant, it has to decide whether the caller has finished or is pausing to think, and whether "yeah" means yes or "go on." The second is the work behind the conversation: following the business's rules, calling the right tools and writing a reply that sounds right out loud.

    In my experience, when you ask one general-purpose model to do both, it ends up doing neither particularly well. So we split them. Our new Think-Live LLM is the model for the moment. It decides when to listen, wait or respond, so the agent stops talking over people and stops leaving dead air, and it's generally available in the Deepgram Voice Agent API. The Think-Fast LLM is the model for the task. It follows your rules, calls tools in the middle of a conversation and writes replies built to be spoken, and it's available in early access on the Deepgram LLM API. You can pair either one with the model you already trust, or use both and let each do the job it was built for.

    Watching it work on a real call

    Here's an example. A customer calls to move a delivery, and while the agent is checking available slots, they add, "actually, Thursday's better than Friday." In a typical pipeline that correction either gets talked over or arrives after the agent has already offered Friday. With the Think-Live LLM managing the conversation, the agent acknowledges the change right away and keeps the customer engaged while the Think-Fast LLM updates the request and finishes the lookup. Because RCS can carry the hesitation in the customer's voice, the reply slows down and confirms Thursday clearly.

    The conversation and the work no longer have to wait on each other, so the agent can listen, adjust and act at the same time, the way a good human agent does.

    Building on Pipecat, in the open

    Developers building voice agents have mostly had two options. They could assemble an open stack themselves and own every integration, or start on a closed platform and rewrite once they outgrew it. We hear the same thing from some of Deepgram's longest-standing customers: they want to keep their options open. We don't think they should have to give that up to get to production.

    That's why we're building the Deepgram Voice Agent Platform on Pipecat, the open framework so many voice agents already run on. I've been inside an acquisition twice, at Apigee and at DataStax, so I know the first question an open-source community asks when its project changes hands. Pipecat stays open source and keeps supporting the open ecosystem, so you can use any model from any provider, including ones that aren't ours. We're running toward open source, not away from it.

    Pipecat is where you build an agent, and the platform is where you'll run it in production, whether that's hosted by Deepgram or in your own cloud. The platform brings Pipecat, RCS and our models together, with one team accountable for how they work together, so you can start on Pipecat today and move to production without rewriting your agent. The platform itself is still ahead. We'll share timing soon, along with the evaluation, observability and agent design tools we're building to work across every model provider.

    Taking voice agents beyond English

    Most of the people a voice agent will ever talk to don't speak English as their first language. Teams expanding into new markets have tended to end up with speech recognition from one vendor and a voice from another, and customers can hear the seams. We've been closing that gap from both sides of the conversation.

    On the listening side, Flux STT Multilingual now has 10 languages available with 31 more in early access, and teams already using it get the new languages without a migration. It also uses earlier turns to make sense of what comes next, which is its own way of keeping the thread. For transcription beyond voice agents, Nova-3 Multilingual now covers 55 languages, also up from 10.

    On the speaking side, Flux TTS is now generally available in Spanish, Portuguese, French, German, Italian, Dutch, Russian, Hindi and Japanese, with 32 more languages in early access, including Mandarin. Wherever the two sides overlap, a team can listen and reply in its customers' languages with one provider.

    Where requests run matters just as much, and it's one of the first questions enterprise teams ask me. Today we're launching a new U.S. endpoint as part of a broader plan to give developers a choice between global reach and regional control. Coming soon, our Global API will provide more reliable, lower-latency connections from anywhere through the endpoint developers already call. Regional endpoints in the U.S., EU, India and Australia keep Deepgram’s processing and storage in-region, so you can choose reach or control workload by workload.

    Where voice agents go next

    Every interaction layer I've worked on has followed a similar arc. The first wave proves the interface can work at all, and the next wave makes it work the way people expect. We strongly believe voice will become the primary way people interact with AI within the next two years. When that happens, nobody will be impressed that an agent got the words right. The bar will be whether it noticed the pause, caught the correction and finished the job without making anyone repeat themselves.

    Every technology wave I've been part of has come down to software talking to software. This time, the software also has to listen. We're building that in the open, with the developers who are already building on it, and the voice agents that win will be the ones that keep the thread.

    Get started

    • Interested in learning more about the Deepgram Voice Agent Platform or evaluating the Think-Fast LLM in early access on the Deepgram LLM API? Contact us or get in touch with your account owner directly.
    • The Think-Live LLM can be accessed in the Deepgram Voice Agent API.
    • Learn how to build with Real-Time Context Standard in Pipecat and try the demo here.
    Original source
  • Oct 27, 2026
    • Date parsed from source:
      Oct 27, 2026
    • First seen by Releasebot:
      Nov 18, 2026
    • Modified by Releasebot:
      Nov 18, 2026
    Deepgram logo

    Deepgram

    October 27, 2026

    Deepgram adds options to verify callback requests using Basic Auth or extra metadata.

    Verify Callback Requests with Basic Auth or Extra Metadata

    Speech-to-Text

    Text-to-Speech

    Intelligence

    Original source
  • Oct 27, 2026
    • Date parsed from source:
      Oct 27, 2026
    • First seen by Releasebot:
      Nov 2, 2026
    Deepgram logo

    Deepgram

    OCT 27, 2026

    Deepgram adds Basic Auth and extra metadata options for verifying callback requests.

    Verify Callback Requests with Basic Auth or Extra Metadata

    Speech-to-Text

    Text-to-Speech

    Intelligence

    Original source
  • Oct 23, 2026
    • Date parsed from source:
      Oct 23, 2026
    • First seen by Releasebot:
      Oct 24, 2026
    Deepgram logo

    Deepgram

    October 23, 2026

    Deepgram expands Nova-3 to 55 languages with language hinting and Japanese numerals, while Flux TTS adds 41 languages and 230 voices.

    Nova-3 Multilingual Expands to 55 Languages and Adds Language Hinting

    Speech-to-Text

    Japanese Numerals on Multilingual Models

    Speech-to-Text

    Flux TTS Adds 41 Languages and 230 Voices

    Text-to-Speech

    Original source
  • Oct 20, 2026
    • Date parsed from source:
      Oct 20, 2026
    • First seen by Releasebot:
      Nov 18, 2026
    • Modified by Releasebot:
      Nov 18, 2026
    Deepgram logo

    Deepgram

    October 20, 2026

    Deepgram CLI 0.4.0 now stores an expiring credential for browser logins, introducing a breaking change for existing signed-in users.

    Deepgram CLI 0.4.0: Browser Login Stores an Expiring Credential

    Browser login now stores a time-limited credential, a breaking change for anyone who signed in before this release.

    Developer Tools

    October 20, 2026

    Self Hosted

    Original source
  • Oct 20, 2026
    • Date parsed from source:
      Oct 20, 2026
    • First seen by Releasebot:
      Oct 5, 2026
    • Modified by Releasebot:
      Nov 17, 2026
    Deepgram logo

    Deepgram

    October 20, 2026

    Deepgram lists an October 20, 2026 update under Self Hosted.

    October 20, 2026

    Self Hosted

    Original source
  • Oct 19, 2026
    • Date parsed from source:
      Oct 19, 2026
    • First seen by Releasebot:
      Oct 21, 2026
    • Modified by Releasebot:
      Oct 24, 2026
    Deepgram logo

    Deepgram

    October 19, 2026

    Deepgram now supports configuring Nova-3 keyterms mid-stream across every regional endpoint.

    Confirm Nova-3 Configure Updates

    Speech-to-Text

    Update Nova-3 Keyterms Mid-Stream on Every Regional Endpoint

    Speech-to-Text

    Original source
Releasebot

Curated by the Releasebot team

Releasebot is an aggregator of official release notes from hundreds of software vendors and thousands of sources.

Our editorial process involves the manual review and audit of release notes procured with the help of automated systems.