The Decentralized Compute Company retweeted
The TDCC compute network is live. 🟢
To celebrate, we're opening uncensored AI to everyone.
No account.
No login.
No filters.
Free.
Open it and ask anything 👇
chat.decentralizedcompute.co…
The TDCC compute network is live. 🟢
To celebrate, we're opening uncensored AI to everyone.
No account.
No login.
No filters.
Free.
Open it and ask anything 👇
chat.decentralizedcompute.co…
Your GPU should get paid like you do.
Soon on TDCC: rent out your GPU, get paid in USDC on Stellar. Seconds to settle. Under a cent in fees.
Anywhere on Earth.
Data center or gaming rig, same deal.
Oct 27.
Which GPU are you putting to work? 👇
Today: the new TDCC is live.
Tomorrow: desktop apps go GA.
Oct 23: everyday PCs become compute nodes.
Oct 27: compute → USDC → stocks.
Oct 31: the GPU marketplace opens to everyone.
8 drops. 3 weeks. One $COMPUTE.
decentralizedcompute.company…
2/ Oct 31 is the finale: rent a GPU by the hour, or list yours and earn. Instant SSH & Jupyter machines, API rentals over the mesh, payouts to hosts.
Hosts: list now and be live on day one → decentralizedcompute.company…
Only trust the contract address posted by @computecompany. We never DM first. Not financial advice.
We don't just talk. We compute.
Introducing Exact Core.
An open-source AI chip whose every answer anyone can re-check.
Same bits, every time, on any machine. Cheat once and the fingerprint gives you away, down to the exact instruction.
Research ships. Hardware ships. solana:4WtWLfKMMexVN5RkDAh88bs5rUx89GXuoDGnb7QVStar powers it.
decentralizedcompute.company…
4/ Escrow, receipts and penalties settle on Solana. Honest providers get paid. Cheaters lose their stake.
5/ 10 models running bit-exact on the card. 85.8 tok/s on LFM2.5-230M.
Code: github.com/the-decentralized…
Meet Compass. 🧭
TDCC's first in-house decision model.
→ 0.8B params
→ Runs on a plain CPU
→ Open source
Yesterday: quantization algorithm.
Today: our own model.
Friday: our own AI chip.
Code: github.com/the-decentralized…
Weights: huggingface.co/computecompan…
$COMPUTE
We just published CASK-Q, a new quantization algorithm from TDCC.
Low-bit quantization doesn't only break weights. It breaks the query × key interaction that attention runs on. CASK-Q calibrates Q and K together, so that interaction survives.
Results:
→ −25.2% attention-output error at 2-bit
→ −28.1% at 3-bit
→ 60/60 paired seed wins
→ same bytes on disk
→ 12.6× faster calibration loss than dense at long context
Paper 👇
decentralizedcompute.company…
Open-source code 👇
github.com/the-decentralized…
We don't just talk. We compute. $COMPUTE
What the paper claims, stated plainly: these results are from controlled synthetic attention experiments, measured against independent Q/K reconstruction on the same quantizer. The next step is full pretrained LLMs. Everything is in the PDF, including the limitations. Code, scripts and results are open source: github.com/the-decentralized…
CASK-Q is the research layer under the TDCC network. The goal is lower-bit models with less error, so everyday PCs and GPUs can run bigger models and do useful compute. $COMPUTE
Official CA only from @computecompany. We never DM first.
AI doesn’t always need to generate.
Sometimes, it just needs to decide.
Tomorrow, our first Jev-inspired System One decision models are coming to @huggingface 🤗
Built for a new generation of AI agents, automation, and intelligent decision-making.
And we’re just getting started.
New research papers are on the way.
We’re not just building decentralized compute. We’re building the intelligence that runs on it.
solana:4WtWLfKMMexVN5RkDAh88bs5rUx89GXuoDGnb7QVStar






