Dessa is Block's full-duplex speech model for live conversation. Kompozy generates and publishes on-brand content across 9 platforms. The honest comparison.
If you searched "Dessa alternative," start by naming what you actually want to replace. Dessa is a full-duplex speech-to-speech model from Block — the payments company co-founded by Jack Dorsey — built by an in-house research lab drawn partly from the original Dessa deep-learning startup. It is audio in, audio out: you talk to it and it talks back in real time, listening and speaking at once, with emotion, laughter, and natural interruptions. Block says it was trained from scratch on human conversation alone and runs locally on a high-end Mac. It is a genuinely ambitious piece of engineering. If your need is a live two-way voice — a conversational agent, a companion, a voice interface — Kompozy is not the tool you're looking for, and I'll say so plainly below.
I run Kompozy, and I only want the readers this page actually fits. Kompozy is a content generation and publishing engine, not a speech model you converse with. People land on "Dessa alternative" from two directions. Some are comparing conversational voice models — Dessa against GPT-Live-1, a voice-agent SDK, or another full-duplex system — on naturalness, latency, and access; for them, the honest answer is to benchmark those directly (and note Dessa is waitlist-only today), and Kompozy isn't in that race. Others heard about Dessa while trying to "make more content," got excited about lifelike AI voice, and then realized their real bottleneck was never the voice. It was producing enough finished, on-brand posts, in enough formats, and getting them published everywhere.
That second reader is who this page is for. The choice that matters isn't "which speech model" — it's "do I need to hold a conversation, or do I need a content operation?" A conversation model hands you spoken audio and stops. It writes no script, makes no video or image, captions nothing, and publishes nothing. If you keep drowning trying to turn one idea into a week of posts across nine platforms, a model you talk to is the wrong shape for that problem, however natural it sounds and however impressively it runs on your laptop.
Everything below reflects both products as of 2026-10-23. Dessa's capabilities, training, and availability are drawn from Block's own announcement and demo page — the performance figures are vendor-reported and there are no independent benchmarks yet, so verify them on the official pages before you build. No invented weaknesses: Dessa's expressiveness, its full-duplex design, and its on-device claim are real, and I frame them as such.
Dessa is a full-duplex, speech-to-speech conversation model unveiled by Block in October 2026. Because it is full-duplex, it listens and speaks at the same time and learned from real conversation when to cut in and when to backchannel with short acknowledgments like "yeah" or "got it." Block says it was trained from scratch on human conversation alone — no text, no pretrained models — on a single NVIDIA GB200 NVL72, learning language from hundreds of times fewer words than large language models, and that it natively handles emotion, sarcasm, laughter, a frustrated sigh, and a voice cracking with sadness. Inference is said to run faster than real time, and the model reportedly runs locally on a MacBook Pro M4 Max or better, with lead researcher Ragavan Thurairatnam saying it can also run on consumer GPUs. What Dessa produces is a spoken conversation. It has no app, no dashboard, no scheduler. It does not write copy, generate images or short-form video, build a carousel or a blog, hold a written brand voice across a content week, or publish anything. It is also early: at announcement it is waitlist-only, with access limited to a small group of AI safety researchers, and Block has named no pricing, license, or general release date. It is a model and, for now, a demo — the entire pipeline around it is yours to assemble, if and when you can access it.
People look past Dessa for a content-creation alternative for one honest reason: it solves live conversation, and live conversation was never the whole problem. If your goal is a steady stream of finished posts, a voice you can talk to is not even an ingredient most of the time — you need something to write the on-brand copy, generate the images and short-form video, keep everything consistent, and publish it across platforms. Dessa does none of that, because that is not what it is. There is also a practical reality that the impressive demo hides. First, you can't use it yet: it's waitlist-only with researcher early access, so planning a content workflow around it today isn't possible. Second, even once it ships, a conversation model and a content engine are different categories — one makes talk sound human, the other turns ideas into published, on-brand content. And for social talking-head video specifically, a tool that generates avatar video with its own built-in voice already removes the need to wire up a separate speech system at all. None of this is a knock on Dessa's engineering; it's a scope-and-shape mismatch. If you need a content engine, a model you converse with is the wrong thing to build on.
| Feature | Dessa | Kompozy | Note |
|---|---|---|---|
| Real-time full-duplex conversation | Yes | No | Dessa is purpose-built for live two-way speech. Kompozy is a content engine and does not hold conversations. |
| Expressive, emotion-native speech | Yes | Partial | Dessa speaks emotion, laughter, and sighs natively. Kompozy uses HeyGen native TTS inside its avatar video only, not a standalone expressive voice engine. |
| Runs locally / on-device | Yes | No | Dessa reportedly runs on a MacBook Pro M4 Max or better and on consumer GPUs. Kompozy is a cloud content engine. |
| Generally available today | No | Yes | Dessa is waitlist-only with researcher early access. Kompozy is a live product you can sign up for now. |
| AI text generation (posts, scripts, blogs) | No | Yes | Dessa speaks; it does not write. Kompozy generates copy governed by a Persona Brief. |
| AI image generation (carousels, quote cards, photos) | No | Yes | Dessa is audio-only. Kompozy generates brand-exact visual formats. |
| AI short-form / avatar video generation | No | Yes | Kompozy produces Persona Shorts, Clipped Shorts, and avatar video; Dessa only converses. |
| Blog + newsletter generation | No | Yes | Kompozy ships long-form text formats from one source; Dessa can only talk. |
| Persona Brief / brand-voice governance | No | Yes | Kompozy enforces tone and banned phrases per brand across written output. A conversational voice is not a written brand voice. |
| Cross-platform scheduling & publishing | No | Yes | Dessa has no scheduler or social connections. Kompozy publishes to eight social platforms plus blog and email. |
| Repurpose recorded conversation into content | No | Yes | Dessa generates the talk; Kompozy turns a recording or transcript into clips, quote graphics, a blog, and a newsletter. |
| Ready-to-use without engineering | No | Yes | Dessa is a model you'd deploy and build around. Kompozy is a finished workflow you operate from a dashboard. |
| Tier | Dessa plan | Dessa price | Kompozy plan | Kompozy price |
|---|---|---|---|---|
| Entry | Dessa (waitlist) | No pricing announced — waitlist / researcher early access | Kompozy Starter | $199/mo (5,500 credits) |
| Mid | Dessa (local / on-device) | Cost of a MacBook Pro M4 Max or better + engineering | Kompozy Pro | $499/mo (18,000 credits) |
| Top | Dessa (your voice product) | Your own infrastructure + development time | Kompozy Enterprise | Custom (sales-led) |
Here's the honest pitch, because the categories don't overlap the way "alternative" implies. Dessa is a full-duplex speech model you converse with. Kompozy is a content operation. If what you need is a lifelike, real-time, emotionally expressive voice — for a conversational agent, a companion, a voice interface, or a voice product — then watch Dessa and don't let this page talk you out of it; the full-duplex design and the on-device claim are the reasons it's worth following (just note it's waitlist-only today).
Kompozy is the alternative for the reader who got excited about lifelike AI voice while trying to fix a content-volume problem. If you keep struggling to turn one idea into a full week of on-brand posts across every platform, the voice was never your constraint — the expensive part was always writing the copy, making the visuals, and shipping it all. Kompozy writes the copy under a Persona Brief, generates the short-form video, carousels, quote cards, photo posts, blog, and newsletter, and schedules and publishes the whole set across eight social platforms plus blog and email — with Autopilot and a per-post review pipeline. For social talking-head video it brings its own built-in TTS, so a separate conversational voice model becomes optional. And when you do record a conversation, Kompozy turns the audio or transcript into captioned clips, quote graphics, a blog, and a newsletter — the publishable half a speech model can't produce.
The sensible read for most creators: Dessa (once it ships) is a voice to talk to; Kompozy is the engine that turns what gets said into a published presence. Start on Kompozy Starter at $199/mo (5,500 credits) and let each tool do the half it's built for.
No. Kompozy is a content generation and publishing engine, not a conversational speech model. It generates copy, images, carousels, short-form and avatar video, blogs, and newsletters, and publishes them across nine platforms. For social talking-head video it uses HeyGen's built-in TTS, but it is not a full-duplex voice you converse with the way Dessa is.
Only if what you actually needed was a content operation, not a live voice. If you need real-time conversation or a voice agent, Dessa (or a comparable model you can access today) is the right tool and Kompozy doesn't replace it. If you were drawn to Dessa hoping it would help you produce more finished posts, Kompozy replaces that broader workflow.
Because they do different jobs. Dessa (if it ships on-device) does one narrow thing: hold a spoken conversation. Kompozy writes the on-brand copy, generates the visuals and video, and schedules and publishes across nine platforms — the production and distribution work Dessa doesn't touch. If your bottleneck is content volume, that's what you're paying for.
In time, yes. Once Dessa is available, you could record a conversation with it and use Kompozy to turn that audio into captioned clips, quote graphics, a blog, and a newsletter, then publish them across platforms. Dessa makes the talk; Kompozy makes it publishable.
Block has announced no pricing, license, or general release date for Dessa — it is waitlist-only with researcher early access, and reportedly runs locally on a high-end Mac. Kompozy is a content engine priced by generation volume — Starter $199/mo (5,500 credits) and Pro $499/mo (18,000 credits). They price two different things: a conversation model versus a full content operation.