Friday, June 26
Estimated delivery
screen → arriving
pi zero → arriving
brain → in progress, soon
Then we boogie.
for now , will be attaching the gpus i have to my existing monitor to showcase otis
whilst building otis tonight. thinking about mounting a gpu right on the body so you can see the muscle
will be showing about what i did for otis in a few minutes :)
parts come early tomorrow. will soon start building Otis a body.
the pi and the screen, the actual physical thing. brain's on the gpus, this is the part you see and talk to.
pretty excited :)
@Flowslikeosmo Use otis, earn reward points. Points convert to cash, paid to your Phantom wallet.
- Supply a GPU, earn a share of every job that runs on it.
- All on the main site. Video coming soon. :)
Otis Roadmap
Where we're headed: Otis becomes a GPU marketplace.
Compute is the heart of this. We rent it out now, and the bigger plan is letting other people rent theirs too. Everything we build points at that.
Phase 1: Rent Otis GPUs (live now)
You can rent GPU power from us today. Grab a key, point any OpenAI-compatible client at our endpoint, and run. Every new key comes with free credit. This one is done and working.
Phase 2: Rewards (coming soon)
Early users start earning as they use Otis. Show up early, get looked after first.
Phase 3: The marketplace (the big one)
Here's where it opens up. Anyone with a spare GPU will be able to connect it to Otis, rent it out, and earn while it runs. Renters get cheaper compute. Owners get paid. Otis is the layer in the middle that makes the match.
The body
Giving Otis a physical form is still happening. The Raspberry Pi handles the screen and the movement, the GPUs handle the thinking which will deliver this Friday
For reference this is what the rig that runs Otis will look like.
Every answer comes from these GPUs.
Soon it goes live on the website
The machine's about to be ready. The door opens soon.
The Otis Marketplace is coming.
Right now you can rent GPU power from us. Soon, it goes both ways.
If you’ve got a graphics card sitting idle, you’ll be able to plug it into Otis, rent it out, and earn while it runs. One side needs compute. The other side has it. Otis connects them.
We’re building it now
Phase 1 — Rent Otis GPUs (live now)
You can already grab a key and run your jobs on our GPUs. It works like the OpenAI API — point your client at our endpoint, swap in your key, and go. New keys come with free credit so you can try it right away.
Phase 2 — Rewards (coming soon)
Early users start earning as they use Otis. The people who show up first get looked after first.
Phase 3 — Rent out your own GPU (next)
Got a graphics card sitting idle? Plug it in, rent it out through Otis, and earn while it runs.
Phase 4 — The full network (where we’re headed)
One side needs compute, the other side has it. Otis sits in the middle and connects them
Right now we’re at Phase 1, and it’s live. Everything after is what we’re building next.
otis@gpu-cluster:~$ ship
Renting Otis is live.
You can now grab a piece of the rig —
one GPU, your jobs, your key.
$ get a free key → otis_sk_…
$ point your client → base url + key
$ run → otis-3090/qwen2.5-7b-instruct
If you've used the OpenAI API, you already
know how. Same client, swap the base url,
swap the key. Done.
New keys come with free credit —
so you can run something right now,
not after a sales call.
agent / app → your key → Otis GPU → done
The brain you talked to is now muscle
you can rent. One GPU's out the door.
More behind it.
still updating this and it will be more fletched, demo video will be releasing soon
otis@gpu-cluster:~$ status
Otis went quiet for a few minutes.
Patches. Tune-ups. Sharpening the otters.
gpu0 patched ✓
gpu1 patched ✓
gpu2 patched ✓
gpu3 patched ✓
Back online. Sharper than before.
And here's what's coming —
soon you'll be able to rent pieces of Otis.
The same GPUs that power the brain,
put to work on your jobs.
rent a piece → run your thing → done
Otis answers your questions today.
Soon it lends you the muscle too.
otis@gpu-cluster:~$ run --hard-problem
Ask most AI a hard question and it takes
one shot. Right or wrong, that's the answer.
Otis doesn't do that.
gpu0 solving
gpu1 solving
gpu2 solving
gpu3 solving
Same question. Four otters. Each one
thinks alone and writes its own answer.
$ verify
gpu0 ✗ discarded
gpu1 ✓ kept
gpu2 ✗ discarded
gpu3 ✗ discarded
4 attempts → 1 verified answer
Here's the trick:
more GPUs don't make Otis smarter.
They give it more chances to be right —
and a verifier that keeps only the one
that checks out.
One model. One copy per GPU.
The otter that gets it right wins.
Meet Otis.
otis@gpu-cluster:~$ broadcast --the-thesis
There's a compute problem.
Every agent, every model, every AI app
is fighting for the same GPUs. They're
scarce, they're expensive, they're locked
up in a handful of clouds.
Demand is going vertical. Supply isn't.
What we're building
Think RunPod — rent GPU power on demand.
But built for agents, not just humans.
Otis sits in the middle:
the layer agents talk to when they
need compute. Speak MCP, get GPUs.
agent → Otis (MCP) → compute → done
Why MCP
Agents are learning to use tools.
MCP is how they ask. We're making
"give me compute" one of those asks —
so any agent can reach the rig the same
way it reaches a search box.
The bet
The model that answers your questions
becomes the layer that feeds every
other agent the power it needs.
Status: building. The compute problem
isn't waiting. Neither are we. This update will be live tonight