
What would you back if a small AI team could pool GPUs with an entire community? We’re building Citrate’s Orchard so people can share compute and run models together. What would you want the first cohort to build? Join the pre-release waitlist.

We built a network for training Sovereign AI, on sovereign soil. 🇺🇸

What would you back if a small AI team could pool GPUs with an entire community? We’re building Citrate’s Orchard so people can share compute and run models together. What would you want the first cohort to build? Join the pre-release waitlist.

Join Citrate and our partners @OrbitalBeam for an interactive workshop… and learn the tricks we used to build a blockchain with @claudeai
Catch us in Miami hosting another Claude workshop with @OrbitalBeam We are excited to learn with you guys. If you’re gonna be in Miami for art Basel then sign up. Attendants get a free month of @claudeai Max on arrival. Seats are limited. luma.com/2hcoiasi


The open orchard keeps fruiting: • Tencent Hy4 Flash: 224B params, 12B active, open weights, from $0.070/M input tokens • SenseNova-Loop-9B: Apache 2.0, 4.9GB at 4-bit, runs on consumer GPUs The best models are getting smaller, cheaper, and yours to run. Waitlist Now 👇


Anthropic's CEO says open-source AI removes safeguards like usage monitoring and the ability to revoke access. Bill Gurley called it a lobbying push to limit public access and keep control. Local AI is an inevitable reality; not "if, but "when" Join the Waitlist Today 🍊

Sen. Hawley wants to pause new data centers and make them bring their own power. Meanwhile a researcher turned old 8GB mining cards into 80GB AI cards that beat an A100 in FP16. The future isn't one giant plantation. It's a million groves.🇺🇸

Qwen3.8-27B, quantized by Unsloth, now targets setups with just 12GB RAM. Apache 2.0. An open-source router reports 97.5% of Opus 5.5 quality at ~1/10th the cost. The harvest keeps getting cheaper, and it grows locally. Join the Waitlist today! 🇺🇸🍊


Big article from our founder Larry, featured in Newsweek! Bring Federated learning models to the most foundational layers of our civilization: Hospitals & Schools. Keeping your data safe and accessible at scale; primed & ready for any AI model.
The Editor at @Newsweek changed a couple of my words, but it would highly advantage schools and hospitals to take part in federated learning. They have the data, they have the machines, and they have the SME’s. newsweek.com/us-should-make-…

Microsoft has 35 US data centers under construction Now it's lobbying the White House because China supplies 67% of the optical transceivers that wire them together. That's centralized AI: one giant supply chain; one giant point of failure. We're building the solution.

Google just announced two new TPUs. Its TPU supply is reportedly already contracted to some of the biggest names in tech. When compute is spoken for before it ships, everyone else waits in line. We'd rather grow our own. Sovereign nodes, on sovereign soil.

GM orchard 🍊 OpenCode just launched Browse: an on-device browser agent that keeps tabs and logins local, with human handoffs for 2FA. More control over where your AI works. That's a direction worth growing and is something we support wholesale.

GM 🍊 Liquid Energy's NanoDC pairs solar, batteries and Starlink for AI compute without grid power or water. Local intelligence doesn't have to wait for a mega-campus. Grow it where you need it. Plant a node. 🇺🇸

Fresh fruit on the open-weight tree: 🍋 Ripple-65B Preview: Apache 2.0, $32M budget, 80.5 on Terminal-Bench 2.1 🍊 Kimi K3.1 Pro: full weights now on Hugging Face 🍈 TwIL-8B: ~54 tok/s on a Mac mini Weights you can hold. Compute you can own. Plant a node. 🇺🇸

Pruna just open-sourced P-H3-Image. 143 GB down to 20 GB. 8 steps instead of 50. 4MP images in 2.6 seconds, with testers running it on a single RTX 5090. Open weights keep getting lighter. The fruit is hanging lower every season. Pick it yourself. Sovereign AI is the Future