agentos.guide › blog
Block Dessa is the speech-to-speech AI model from Block, the company behind Square and Cash App, and it was trained only on human conversation audio.
A very small research lab at Block built it, and the lab reports directly to Jack Dorsey.
Block's page says it runs locally on a MacBook Pro M4 Max or better, and on 24 October 2026 its download and waitlist links were marked as coming soon.
This page explains the news, the team and what it means for voice inside an Agent OS.
I don't have a Dessa video, so here's my Hermes Jarvis video, which shows the voice agent I'd compare it with.
This table is the whole announcement on one screen.
| Question | What Block and its team have said |
|---|---|
| Who announced it? | Block announced it on its own page, and Jack Dorsey and Dessa lead Ragavan Thurairatnam posted about it on X. |
| Who is Block? | Block is the company behind Square, Cash App, Afterpay and TIDAL, which its privacy notice names. |
| Who built it? | A very small AI research lab at Block built it, and the lab reports directly to Jack Dorsey. |
| What is it? | It is a full-duplex speech-to-speech model trained from scratch on human conversation audio. |
| What hardware does it need? | The page says it runs locally on a MacBook Pro M4 Max or better. |
| What was it trained on? | Block says it trained the model on only one NVIDIA GB200 NVL72. |
| What can I use today? | You can listen to the demos, while the download and waitlist links were marked as coming soon on 24 October 2026. |
| What is missing? | The page gives no price, no API and no release date for the download. |
I read Block's page on 24 October 2026, and I haven't used Dessa myself yet.
So this guide is the news explained, plus what it means for anyone running voice inside an Agent OS.
Block announced Dessa as a conversation model trained only on speech, and its page sums up the method as "No text, no pretrained models".
It learned to hold a real-time conversation just by listening to human conversation audio.
It's full-duplex, so it listens and speaks at the same time and knows when to cut in or make small sounds that show it's with you.
Block says it hears and speaks emotion natively, including sarcasm, laughter, a frustrated sigh and a voice cracking with sadness.
Block's claim is that, as far as it knows, Dessa is the first AI you can have a realistic live conversation with that was trained only from human speech.
My separate guide to Dessa AI walks through every demo and the product side in more detail.
The team is the most interesting part of the news, because it explains why a payments company is shipping speech research.
| Team fact | What the page says |
|---|---|
| Size | Block calls it a very small AI research lab. |
| Origin | It was founded by former members of Dessa, the Toronto deep learning startup. |
| Experience | The lab says it has built deep learning models since 2012 and generative speech models for over eight years. |
| Reporting line | It reports directly to Jack Dorsey, Block's cofounder and chairman. |
| Lead | Ragavan Thurairatnam signs the page as Dessa lead. |
| People | The page lists six researchers, three interns and two former interns. |
The model shares its name with the Toronto startup the founders came from.
The lab also says it chases frontier problems that others ignore.
That reads to me like a long-standing speech research team that now sits inside Block, rather than a brand-new experiment.
🔥 Want the exact voice AI setup I run inside my Agent OS?
Inside the AI Profit Boardroom you get the installable Agent OS, step-by-step video tutorials, four coaching calls a week and 3,400+ members building real automations.
Jack Dorsey framed the launch around one belief, which is that voice is the most natural way to interact with intelligence.
He also said on X that Dessa is small enough to run on a MacBook, which matches the MacBook Pro M4 Max requirement on the page.
Ragavan Thurairatnam described it on X as a full-duplex speech-to-speech model that runs on consumer GPUs.
He also compared the launch to the GPT-3 moment for text, which is a big claim that only outside testing can confirm.
Block gives three reasons on its page, which are curiosity, better understanding between humans and AI, and safety.
The curiosity reason is whether a model could learn to talk the way babies do, without ever seeing a written word.
The understanding reason is that today's voice AI flattens speech into text and can't keep up with human timing.
The safety reason is that AI keeps gaining power while staying nearly deaf to the emotion in our voices.
Block's example is a person in serious pain while a voice AI has no idea and just keeps going.
The page doesn't say Dessa will go into Square, Cash App or any other Block product.
This is what was actually open when I checked the page.
| Door | Status on 24 October 2026 | What is behind it |
|---|---|---|
| Demo recordings | It is open now. | There are six headline clips and 15 more samples, each a live, unsteered conversation. |
| Download | The button is shown, but launch links were marked as coming soon. | It is the route the page shows for getting the model. |
| Join the waitlist | The button is shown, but launch links were marked as coming soon. | It is the other route the page shows. |
| Safety research access | It is open to a small group of researchers. | The form asks for a name, an institutional email and an optional Google Scholar or blog link. |
| More recordings | Block has promised them. | The page says more recordings will be shared on X soon. |
The headline demos show Dessa laughing mid-sentence, getting louder when asked, finishing a forgotten film title and gasping at a surprise.
One clip shows Dessa talking with GPT-Live-1, which Block identifies as ChatGPT Live, and both voices in it are AI-generated.
Block notes that the developer of GPT-Live-1 isn't affiliated with Block and hasn't reviewed or endorsed Dessa.
Block also says nobody steers Dessa in any clip, and the subtitles were added afterwards.
Block says Dessa learned language from hundreds of times fewer words than large language models need.
The page says a single spoken word takes thousands of times more bytes than the same word in text.
It also says inference runs faster than real time, so there's no awkward wait.
The page doesn't publish the model's size, its training hours or the languages it speaks.
🔥 Want the exact voice AI setup I run inside my Agent OS?
Inside the AI Profit Boardroom you get the installable Agent OS, step-by-step video tutorials, four coaching calls a week and 3,400+ members building real automations.
My Agent OS already has a voice copilot called Apollo, and Hermes Jarvis is the voice agent I show most often in my videos.
Apollo's loop is a wake word, speech to text, a brain that answers or hands real jobs to an agent, and a butler voice that replies.
Here's how that compares with what Block has described for Dessa.
| Agent OS layer | How Apollo handles it today | What Block Dessa changes |
|---|---|---|
| Hearing | Apollo turns your speech into text before the brain sees it. | Dessa takes the audio itself, so tone is not thrown away. |
| Turn-taking | Apollo's realtime mode lets you talk continuously and interrupt it. | Dessa is full-duplex by design and learned turn-taking from real conversations. |
| Emotion | The text step loses most of the tone in your voice. | Block says Dessa hears and speaks emotion natively. |
| Doing the work | Hermes opens apps, opens websites and builds apps while you watch. | Block's page does not mention tools, so the agents would still do this. |
| Memory | Apollo writes what you tell it to your Obsidian vault and reads you a briefing. | The page does not mention memory, so the vault stays in charge. |
The short version is that Dessa improves the conversation, while the Agent OS still owns the work.
The ideal future setup would put a speech-native voice like Dessa in front of Hermes and the rest of the crew.
Block hasn't published an API, so that's a plan to watch rather than a setup to build today.
If you want voice working in your stack now, my Apollo guide and Hermes Apollo guide show what runs today.
I'd keep Apollo and Hermes Jarvis running for every voice job that needs an action.
I'd listen to all of Block's demos and note the behaviours my current voice stack can't match, such as laughing at the right moment.
I'd check whether I have a MacBook Pro with an M4 Max chip or better, because that's the hardware Block names.
I'd bookmark block.xyz/dessa and check back for the launch links.
If you want help working out where voice AI fits an SEO or agency business, you can book a free SEO strategy session with my team.
These are the signals that will decide whether Dessa ever slots into an Agent OS.
| Signal to watch | Why it matters for an Agent OS |
|---|---|
| The launch links going live | You cannot test anything until the download or waitlist opens. |
| An API or SDK | Without one, there is no way to plug Dessa into Apollo or Hermes. |
| Price and licence terms | These decide whether you can use it for client work. |
| Language support | The page lists learning new languages as an open research question. |
| Safety findings | Block invited researchers to find risks it had not anticipated. |
If you join the waitlist, the privacy notice on the page says Block may share information with group companies such as Square and Cash App, and that it doesn't sell your information.
First, the team behind it has years of speech research, so this launch deserves attention.
Second, it changes how a voice agent should sound, but it doesn't replace the agent that does the work.
Third, access is the bottleneck, so use the voice tools you have now and test Dessa when Block opens the doors.
Block Dessa is a full-duplex speech-to-speech AI model from a small research lab at Block, the company behind Square and Cash App. Block says it was trained from scratch on human conversation audio.
A very small lab at Block built it. Block says the lab was founded by former members of Dessa, the Toronto deep learning startup, and that it reports directly to Jack Dorsey.
Ragavan Thurairatnam signs the official page as Dessa lead.
The page does not say Dessa is part of any Block product. It presents Dessa as a research model.
Not yet. The page lists no API or developer docs, and on 24 October 2026 the download and waitlist links were marked as coming soon.
Block's page says it runs locally on a MacBook Pro M4 Max or better.
Keep your current voice agents doing the work, listen to Block's demos and come back for the launch links.
If you want a second opinion on where voice AI fits your business, you can book a free SEO strategy session with my team.
If you want the whole Agent OS with Apollo, Hermes and the training already set up, it lives inside the AI Profit Boardroom, with 3,400+ members and four coaching calls a week.
That's how I'd stay ahead of Block Dessa.
About Julian
I'm Julian Goldie, an AI entrepreneur, SEO expert and the founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation and SEO.
My YouTube channel has 400,000+ subscribers, and I run Goldie Agency, a seven-figure SEO agency.
→ Apollo: The Voice Copilot In My Agent OS
→ Hermes Apollo: Run Your Mac By Voice
→ Hermes Jarvis: The Voice Agent
The installable system behind every guide on this site — dashboard, agents, memory and pipelines — updated as new tools land, with 3,400+ members and four coaching calls a week.
Join the AI Profit Boardroom →📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉