Sovereign AI for enterprises. Build AI you own on infrastructure you control. Private, specialized models that run where your data lives.

Joined May 2022
136 Photos and videos
Pinned Tweet
Mar 18
A New Model for Intelligence. The next era of AI won't come from a single system in the cloud, but from a civilization of models owned by the people closest to the problem.
3
5
26
194,920
AI policy should not be defined only by what frontier labs, cloud platforms, and centralized model providers believe is best. webAI has a different view: ownership, local control, distributed intelligence, practical deployment, and a focus on the people and institutions AI is meant to serve. Here’s what we think AI policy should stand for: – Empowerment over extraction – Control belongs with the user – The strongest intelligence is distributed and specialized – AI should be a utility for everyday work and life – AI policy should reward practical, broadly shared value Read the full piece: webai.com/blog/what-webai-st…
1
4
9
227
AI data centers bring environmental & security concerns. Jason Rathje, president of public sector, @thewebAI and former director, DoD Office of Strategic Capital: Anytime you centralize power, data, and now artificial intelligence, you create an acute threat for our adversaries."
3
3
813
May 28
Intelligence is getting smaller. Our CEO @Davidstout was featured in @theinformation on why the data center boom is a mistake and where AI is actually headed. "Data centers won't disappear, but the majority of work will happen at the edge. Apple made the bet correctly there." Read the full article: theinformation.com/articles/…
1
2
123
May 28
"Any time you centralize power, data, and now artificial intelligence, you create an acute threat for our adversaries." Our President of Public Sector Jason Rathje was on @CNN This Morning talking about AI data centers, the strain they're putting on communities, and what a real alternative looks like.
3
1
8
272
May 27
The winners of our first YOLO26 MLX Build Challenge are in. 7 days. 4 tracks. 4 wins. Here's what builders shipped 🛠
1
2
5
308
May 27
Full demos and write-ups for each winner coming over the next week. Stay tuned. Big shoutout to our co-hosts HackAI, AITX, and Antler for hosting our kickoff events in Austin. Couldn't have pulled it off without them.
1
2
57
May 27
HUGE thanks to every builder who shipped. Already planning the next one. Join our community to get in on the next challenge: community.webai.com/ #YOLOMLX
1
1
47
May 20
Our YOLO26 MLX Build Challenge kicked off this week. Thanks to our co-hosts HackAI, AITX, and Antler for hosting us in Austin. 7 days. 4 tracks. $4,000 in prizes. Open to anyone in the US. Submissions due Sunday May 24. Full brief and how to enter: community.webai.com/t/the-yo…
1
6
160
webAI retweeted
Side Quest: building/testing a Swift MLX iOS implementation for the @Webai YOLO26 MLX Challenge iPhone 16 Pro Max live test: - Live camera tracking overlays✅ - Still-image object detection✅ - On-device inference✅ - Early physical-device validation✅ Write-up coming soon
1
2
4
254
May 15
Our first open source build challenge starts Monday. 7 days. 4 tracks. $4K in prizes. Build something with YOLO26 MLX, our open source object detection model built for Apple Silicon. Two kickoff events in Austin: 🗓 Mon 5/18 with Hack AI (Capital Factory): luma.com/io3nuky7 🗓 Tue 5/19 with @aitxcommunity @AntlerGlobal: luma.com/webai-yolo Full brief and how to join: webai.discourse.group/t/the-…
1
2
8
263
May 14
Our President of Public Sector, Jason Rathje, was on CNBC's Morning Call this morning talking about the US/China AI arms race and what it's going to take for America to stay ahead. He made the point that distributed AI and reducing dependence on centralized cloud infrastructure is not just a technical decision. It is a national security one. Watch the full segment on CNBC: cnbc.com/video/2026/05/14/ja…
1
106
webAI retweeted
Replying to @thewebAI
2
2
12
480
Apr 23
8/ The loss function is proprietary. But the principle: force the model to create cleaner separation between the correct page and every competing page in the batch. Better embedding geometry. Cleaner ranking. Especially important when pages look semantically similar.
1
1
113
Apr 23
9/ That approach earned two of the top three spots on ViDoRe V3: heterogeneous data, targeted rebalancing, efficient adaptation, and an objective built for ranking quality. We're open sourcing both models so anyone can build with them. Blog: webai.com/blog/webai-colvec1… Try both models on Hugging Face: ColVec1-9B: huggingface.co/webAI-Officia… ColVec1-4B: huggingface.co/webAI-Officia…
1
2
134
Apr 23
6/ For adaptation, we used LoRA (rank 32, alpha 32, dropout 0.1) on the Qwen 3.5 backbone plus a retrieval projection layer on top. This gives us a lightweight way to specialize the model for retrieval without full fine-tuning. Trained on 8 A100s with an effective batch size of 512.
2
1
111
Apr 23
7/ Why batch size matters here: we use in-batch negatives. For every query, its matched page is the positive. Every other page in the batch is a negative. With an effective batch size of 512, each query is learning against 511 competing document pages per step. Larger batch = more negatives = stronger contrastive signal.
1
1
86