Simular is launching Sai, a computer-use agent that runs on the operating system you already work on and drives your real applications through mouse, keyboard, and screen reading rather than APIs. Sai operates macOS, Windows, the browser, and Android to finish multi-step tasks on its own , and it lives in a floating window so your cursor stays yours while the agent works alongside you. The pitch is aimed at operators tired of babysitting demo-grade agents that stall and hand tasks back to the human, which the launch tweet and video call out directly.
The launch also introduces SaiFleet, a way to run many agents in parallel on the same class of work. Sai can run many virtual machines in parallel on the same class of task, the same form filled across fifty portals, the same report pulled from twenty dashboards, so throughput scales with machines rather than with the length of one queue . Sai also runs on Windows on your own machine, not only in a remote VM , which matters for teams whose work sits inside legacy desktop software that was never built to be automated.
Simular was founded by ex-DeepMind scientists and is led by co-founder and CEO Ang Li, previously at Google DeepMind, Baidu Apollo, and FAIR. The company's open-source Agent S framework has ranked at the top of the OSWorld benchmark for computer-use agents, and the Sai release lands shortly after a $21.5 million Series A led by Felicis, with NVentures (Nvidia's venture arm), existing seed investor South Park Commons, and others joining in , giving the launch weight for founders and operators evaluating where desktop automation is actually usable today.
Comments (11)
Sign in to join the discussion.
P
Priya Vatsan22d ago
Finally an agent that admits it trips instead of quietly deleting my downloads folder. The bar is subterranean but I'll take it.
K
Kwesi B.22d ago
The floating window framing is smart. Half these agent companies are still shipping a chat sidebar and calling it computer use.
A
Ara Celi Mora22d ago
That opening line in the tweet does the whole job of positioning against the last 18 months of fake demos. Whoever wrote the copy earned their coffee.
H
Hallie N.22d ago
Been watching this team since the early screen agent stuff. The pivot from lab demo to actual desktop shipping is the part nobody else on my list has cracked.
D
Dmitri Volkov22d ago
Cool robosecretary but who pays for the wrong click that emails my whole contact list. Curious how you handle the undo layer at the OS level.
C
Chesa Okonkwo22d ago
Local-first computer use on your own machine is interesting because the inference cost story is completely different from the cloud agent crowd. Waiting to see the pricing page before I get excited.
Y
yuki22d ago
Building in adjacent territory and honestly the SaiFleet naming is annoyingly good. We spent three weeks in a Notion doc and came up with 'Runner.'
F
Farrukh Z.22d ago
Editing on the launch video is tight but the voiceover sounds like it's apologizing for existing. Push the mids up next time.
B
Benoît Marchetti22d ago
We prototyped something like this internally in 2019 and killed it because the eval loop was brutal. Genuinely curious what changed to make it viable now.
N
Nikhil K.22d ago
So the agent asks me for help and I ask the agent for help. Congratulations, we invented pair programming with extra steps.
L
Linnéa Bergstrom22d ago
Is any of this open? Would love to see how the action space is defined before I recommend it in our community server.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
Simular AI
The Autonomous Computer Company
The Autonomous Computer Company
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
Simular is launching Sai, a computer-use agent that runs on the operating system you already work on and drives your real applications through mouse, keyboard, and screen reading rather than APIs. Sai operates macOS, Windows, the browser, and Android to finish multi-step tasks on its own , and it lives in a floating window so your cursor stays yours while the agent works alongside you. The pitch is aimed at operators tired of babysitting demo-grade agents that stall and hand tasks back to the human, which the launch tweet and video call out directly. The launch also introduces SaiFleet, a way to run many agents in parallel on the same class of work. Sai can run many virtual machines in parallel on the same class of task, the same form filled across fifty portals, the same report pulled from twenty dashboards, so throughput scales with machines rather than with the length of one queue . Sai also runs on Windows on your own machine, not only in a remote VM , which matters for teams whose work sits inside legacy desktop software that was never built to be automated. Simular was founded by ex-DeepMind scientists and is led by co-founder and CEO Ang Li, previously at Google DeepMind, Baidu Apollo, and FAIR. The company's open-source Agent S framework has ranked at the top of the OSWorld benchmark for computer-use agents, and the Sai release lands shortly after a $21.5 million Series A led by Felicis, with NVentures (Nvidia's venture arm), existing seed investor South Park Commons, and others joining in , giving the launch weight for founders and operators evaluating where desktop automation is actually usable today.
Comments (11)
Finally an agent that admits it trips instead of quietly deleting my downloads folder. The bar is subterranean but I'll take it.
The floating window framing is smart. Half these agent companies are still shipping a chat sidebar and calling it computer use.
That opening line in the tweet does the whole job of positioning against the last 18 months of fake demos. Whoever wrote the copy earned their coffee.
Been watching this team since the early screen agent stuff. The pivot from lab demo to actual desktop shipping is the part nobody else on my list has cracked.
Cool robosecretary but who pays for the wrong click that emails my whole contact list. Curious how you handle the undo layer at the OS level.
Local-first computer use on your own machine is interesting because the inference cost story is completely different from the cloud agent crowd. Waiting to see the pricing page before I get excited.
Building in adjacent territory and honestly the SaiFleet naming is annoyingly good. We spent three weeks in a Notion doc and came up with 'Runner.'
Editing on the launch video is tight but the voiceover sounds like it's apologizing for existing. Push the mids up next time.
We prototyped something like this internally in 2019 and killed it because the eval loop was brutal. Genuinely curious what changed to make it viable now.
So the agent asks me for help and I ask the agent for help. Congratulations, we invented pair programming with extra steps.
Is any of this open? Would love to see how the action space is defined before I recommend it in our community server.