Community
Hackers in the Loop.
The Lab exists for the people around it. Its experiments become the Learning Hub's books and lessons, its projects meet their users in our Discord, and when someone's open work could use a GPU, we offer time on one.
The Hackers in the Loop Lab
A small lab for open research.
GPUs old and new, three DGX Sparks and a few benches of small devices, in one room. It’s where our core projects are built, run every day and measured, and we share what we learn.
What the Lab drives
Where our core projects get built.
The Lab isn’t a showroom. Our core projects are built, run and measured on it, and it exists for the community around them.
Autoresearch
Thermite.
The Lab is Thermite's proving ground. Its rigs and dashboard run on the Lab's work cluster, and its campaigns run for days on the GPU server's cards. Every result it keeps was measured on Lab hardware by a judge the agents can't touch.
Your own bots
fxbot.
fxbot runs every day in the Lab, on one of its virtual machines, and updates itself as each release lands. Daily use is how we find out what it needs next.
Affordable inference
tensorless.
PN3, tensorless's 3-bit format for the P100, was made on a DGX Spark. Thermite campaigns on the GPU server's P100s then put it into the engine, checked its quality and raised its speed to 30 tokens per second. Every release is measured on a real P100 before it ships.
Honest benchmarks
Small Model Arena.
Small Model Arena runs its harness on a DGX Spark and serves models on the GPU server's four V100s, while other Lab machines run its code tasks and grade its research answers. Each run records exactly what it ran on, and how.
How it’s built
Built in layers. Changed through code.
Everything fits in one room, and it’s run like a careful data centre. These are the habits that hold it together.
Boxes inside boxes.
Machines host virtual machines, virtual machines host containers, and agents work in sandboxes inside those. Each layer can be rebuilt without disturbing the one around it.
One work cluster across the fleet.
Kubernetes spans the GPU server, two of the DGX Sparks and the bench machines, without changing their own systems. A workload asks for a GPU by model and gets one. Thermite's rigs, tensorless's release tests and Small Model Arena's V100s all arrive this way.
A private network with walls.
Machines meet over an encrypted private network, and each service is open only to the machines and people that need it. Agents that read the web work in a separate enclave, away from everything else.
Change through code, and a person says yes.
Every machine is set up from one repository. Changes are planned, checked before they're applied, and written up with their evidence afterwards. Agents help run the Lab, but a person approves what changes.
The machines, by job.
- core projects built and tested here
- 5
- GPU memory in one 8-GPU server
- 192 GB
- unified memory across three DGX Sparks
- 384 GB
- shared storage for models and results
- ~24 TB
- Model servers
- 3 × NVIDIA DGX Spark128 GB of unified memory each. They serve models to the Lab, make quantizations like tensorless's PN3, and run Small Model Arena.
- GPU server
- 8 data-centre GPUsFour V100s and four P100s in one 4U server. The P100s are Thermite's research lanes for tensorless; the V100s serve Small Model Arena.
- VM host
- Dell PowerEdge R730512 GB of memory for the Lab's virtual machines.
- Platform
- A mini PCRuns the shared services: sign-in, dashboards and the model gateway.
- GPU benches
- Swappable cardsRadeon VII, BC-160, RTX and spare P100 cards on external docks, one GPU at a time.
- Storage
- 4-bay NASAbout 24 TB usable: model files, logs and results, shared over the network.
- Small devices
- FPGA, edge AI and ESP32FPGA boards, edge-AI modules, ESP32 devices and an electronics bench.
- Rack
- 12U, 10-inch mini rackSwitches, KVM, status screens and battery backup.
How the research runs / Thermite
Run many experiments. Keep the best. Check everything.
Open by default. Most of the code is public, and so are the projects that didn’t work out.
Measured, not claimed. Agents propose changes, but a judge they can’t touch times every run and decides what counts.
Real hardware. Every number comes from the Lab’s own machines, and we say which ones.
An interactive drawing of Thermite, the research harness the Lab built and runs. Pick a scene, or let it play. The drawing shows Thermite’s design; usethermite.com ↗ has the real thing.
Working with the Lab
We reach out.
When we come across open work the Lab could help, like a model to test, a kernel to speed up, or a question that needs GPUs, we get in touch and offer time on it.
The best way to be found is to share what you’re building with the community.
Support the Lab
Have hardware or hosting space to offer?
Hardware and hosting support can make more room for community projects. Tell us what you can offer and any constraints.
Start with a conversation.
Tell us what you can offer and any constraints. Equipment, delivery, hosting, and operating arrangements can be discussed before anything changes hands.
September 6–8, 2026 / The build journey
Built together.
Learned along the way.
This mini cluster was a project Shannon built with his daughter.
The first evening meant laying out parts, assembling the rack, and figuring out how everything would fit. His daughter picked the colorful zip ties for the exhaust fan and suggested putting a mini PC on the back to make room.
After coffee and donuts, they returned to network cabling, cooling, and conversations about AC, DC, and why current matters. By September 8, the first two Sparks and a mini PC were online, brought up one system at a time using a Mac and HDMI capture as the console.
