agentos.guide › blog
Dessa AI: The Quick Start For Block's Speech-Only Voice Model, And Where It Fits Your Agent OS
Dessa AI is a full-duplex speech-to-speech model from Block, and it was trained from scratch on human conversation audio with no text and no pretrained models.
Block, the company behind Square and Cash App, built it in a small research lab that reports directly to Jack Dorsey.
The official 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 is the quick start, plus where a speech-native model fits an Agent OS workflow.
I don't have a Dessa video, so here's my Hermes voice agent video, which shows the kind of voice setup Dessa is trying to beat.
Dessa AI at a glance
This table is the whole product on one screen, taken only from Block's official page.
| Question | What Block's page says |
|---|---|
| What is it? | It is a full-duplex speech-to-speech model, so audio goes in and audio comes out. |
| Who made it? | A very small AI research lab at Block made it, and the lab reports directly to Jack Dorsey. |
| What did it learn from? | It learned from human conversation audio only, with no text and no pretrained models. |
| Where does it run? | The page says it runs locally on a MacBook Pro M4 Max or better. |
| How fast is it? | Block says inference runs faster than real time. |
| How was it trained? | Block says it trained Dessa on only one NVIDIA GB200 NVL72. |
| Can I get it? | The page shows Download and Join the waitlist buttons and says launch links are coming soon. |
| What does it cost? | The page lists no price, no plan and no API. |
I checked that page on 24 October 2026, and I haven't used Dessa myself yet.
So this guide is a quick start for what you can do today, plus how a model like this could fit an Agent OS.
The Dessa AI quick start
This is the shortest path from hearing about Dessa to being ready to test it.
| Step | What you do | Where |
|---|---|---|
| 1 | You open Block's official page and listen to the six headline demos. | block.xyz/dessa |
| 2 | You listen to the 15 extra samples further down the page. | The More samples section of the same page. |
| 3 | You check whether you have a MacBook Pro with an M4 Max chip or better. | Your Mac's About This Mac screen. |
| 4 | You come back for the Download and Join the waitlist links once they go live. | The buttons at the top of the page. |
| 5 | You apply for early access only if you are an AI safety researcher. | The request form, which asks for an institutional email. |
The demos are the only part you can use today, and Block says every one is a live recording that nobody steers.
Each clip plays as two synchronised speaker tracks, so you hear both the person and Dessa at once.
Block also says the subtitles were added afterwards, so they aren't something the model produced.
Don't apply through the safety form unless you genuinely research AI risk, because the form is built for universities, companies and labs.
What Dessa AI is
Dessa AI is a conversation model trained only on speech, and Block's page puts its approach in one phrase, which is no text and no pretrained models.
It learned to hold a real-time conversation simply by listening, the way children learn to talk before they can read.
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.
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 natively hears and speaks emotion, sarcasm, laughter, a frustrated sigh and a voice cracking with sadness.
Jack Dorsey wrote on X that it's "small enough to run on your macbook", and the official page names a MacBook Pro M4 Max or better.
🔥 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.
How Dessa AI differs from a normal voice agent
Most voice agents, including the ones I run, are cascaded, which means several separate pieces pass the conversation along.
| Stage | A cascaded voice agent | Dessa AI |
|---|---|---|
| Listening | Your speech is turned into text first. | Your speech goes straight into the model as audio. |
| Thinking | A text model writes a reply. | The same speech model decides what to say, when and how. |
| Speaking | A separate voice engine reads the reply aloud. | The model speaks in its own voice and personality. |
| Interrupting | Most agents wait for you to finish. | It listens and talks at the same time. |
| Emotion | Tone and sighs are mostly lost in the text step. | Block says it hears and speaks emotion natively. |
Block's argument is that turning your voice into text throws away information that text can't carry.
Its example is a sentence said several times with the stress on a different word each time, where the words stay the same but the meaning changes.
Block also says this flattening is why most voice AI feels stilted, slow and annoying.
Every behaviour in the Dessa AI demos
These are the six headline clips on the page, in the order Block shows them.
| Demo | What Block says it shows |
|---|---|
| Talking about the release | It laughs mid-sentence when something is funny and gets louder when asked to speak up. |
| Remembering a movie | The person forgets a film title and Dessa finishes the sentence. |
| An anniversary surprise | It gasps under its breath and answers with what sounds like teary gratitude. |
| Expressive conversation | It shows a huge range in intonation and speed. |
| Life advice | Block says it sounds genuinely empathetic. |
| Dessa in conversation with GPT-Live-1 | It talks with ChatGPT Live, both voices are AI-generated, and Block notes that the developer of GPT-Live-1 has not endorsed Dessa. |
Block says nobody taught Dessa to get louder when asked, which is the claim I'd test first.
The page also lists 15 more samples, with names such as Refund, Safety lab, Ronaldo, Zelda discussions and Singing nonsense Spanish.
Block says more recordings will be shared on X soon.
The numbers behind Dessa AI
Block says Dessa learned language from hundreds of times fewer words than large language models need.
The page also says a single spoken word takes thousands of times more bytes than the same word written as text.
Speech never repeats exactly, so even one sentence said twice by the same person produces different data.
Despite that, Block says it trained Dessa on only one NVIDIA GB200 NVL72.
The page doesn't publish the model's size, its memory needs or the operating systems it supports.
Who built Dessa AI
Block describes the team as a very small AI research lab, founded by former members of Dessa, the Toronto deep learning startup.
The lab says it has built deep learning models since 2012 and generative speech models for over eight years.
It reports directly to Jack Dorsey, who Block lists as its cofounder and chairman.
Ragavan Thurairatnam signs the page as Dessa lead, and the page names six researchers and several current and former interns.
My separate guide to Block Dessa covers the company side of the story in more depth.
🔥 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.
How Dessa AI could fit an Agent OS workflow
My Agent OS already has a voice copilot called Apollo, so the real question for me is where a speech-native model would slot in.
| Agent OS layer | What it does today | Where Dessa AI could fit later |
|---|---|---|
| Voice | Apollo hears you, turns speech into text and speaks back in a butler voice. | A speech-native model could make the talking feel far more natural. |
| Realtime mode | Apollo's realtime mode lets you talk continuously and interrupt it mid-sentence. | Dessa is full-duplex by design, so interruptions are part of how it learned. |
| Brain | A language model answers quick questions or hands real jobs to an agent. | Block's page does not mention tools, so the agent would still do the work. |
| Hands | Hermes opens apps, opens websites and builds apps while you watch. | Dessa would sit in front of this layer, not replace it. |
| 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 would stay in charge of that. |
Apollo's guide shows the loop I use today, which is wake word, speech to text, the brain answers or runs an agent, and then it speaks back.
That loop is brilliant for getting jobs done, but it doesn't hear tone, which is exactly what Dessa is built for.
The best future setup would let Dessa handle the conversation while Hermes and the rest of the crew handle the work.
Block hasn't published an API or any developer docs, so that's a plan for later rather than a setup guide for today.
If you want to try voice in your own stack now, my Apollo guide and my Hermes Jarvis guide show what works today.
Why Block says it built Dessa AI
Block gives three reasons, which are curiosity, better understanding between humans and AI, and safety.
The page says AI keeps gaining power and independence while staying nearly deaf to the emotion in our voices.
Its example is a person in serious pain while a voice AI has no idea and just keeps going.
Block calls Dessa a step toward fixing that, and it's giving a small group of AI safety researchers early access to look for risks.
The Dessa AI limits to plan around
These limits are the gaps on the official page, not things I've found by testing.
| Open question | What I would do about it |
|---|---|
| There is no release date for the download. | I would plan nothing client-facing until the links go live. |
| There is no price, plan or API. | I would not quote a client or build an integration yet. |
| The page does not say which languages it speaks. | I would test my own language on day one. |
| The page does not mention tools or actions. | I would keep my agents doing the work and treat Dessa as a voice. |
| The safety research is still early. | I would read whatever findings Block publishes before using it with customers. |
If you want help working out where voice and agents fit an SEO or agency business, you can book a free SEO strategy session with my team.
My three takeaways on Dessa AI
First, the conversation itself is the breakthrough, because a model that hears tone can respond like a person rather than a script.
Second, it's a voice rather than an agent, so your agents still do the work.
Third, access is the bottleneck, so listen to the demos now and test it the day the links go live.
Also On Our Network
FAQ
What is Dessa AI?
Dessa AI is a full-duplex speech-to-speech model from a small research lab at Block. Block says it was trained from scratch on human conversation audio, with no text and no pretrained models.
Can I download Dessa AI today?
Not yet. On 24 October 2026 the official page showed Download and Join the waitlist buttons and said launch links are coming soon.
What hardware does Dessa AI need?
Block's page says it runs locally on a MacBook Pro M4 Max or better, and that inference runs faster than real time.
Does Dessa AI have an API?
The official page lists no API, no SDK and no price, so there is no developer route yet.
Can Dessa AI replace my voice agent?
Not as a whole agent. Block presents it as a conversation model and does not mention tools, so an agent such as Hermes would still do the actual work.
Who leads Dessa AI?
Ragavan Thurairatnam signs the official page as Dessa lead, and the lab reports directly to Jack Dorsey.
Final word on Dessa AI
Listen to the demos, check your Mac against the M4 Max requirement 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 get ready for Dessa AI.
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.
Related guides on agentos.guide
→ Apollo: The Voice Copilot In My Agent OS
→ Hermes Apollo: Run Your Mac By Voice
→ Hermes Jarvis: The Voice Agent
Get the whole Agent OS
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 👉