Web search still varies by harness. The same agent can get different results depending on where it runs. Declare @nimble_search on Omnigent’s web_search builtin, and every harness underneath uses the same Nimble Search API. On @databricks, governed internal data plus live web signals sit in one agent. What the company knows, plus what is happening right now. 🔗 Learn more: #Omnigent# #Nimble# #AIAgents#

Web infrastructure just levelled up. Today @SpaceX added around 1% of global internet bandwidth – roughly 2,200 Tbps with 26 @Starlink V3 satellites. Many things are now more accessible than ever, including streaming. Go live from anywhere under the skies 🌌

Web research: on 1,266 BrowseComp tasks, Computer reaches 66.7% accuracy vs 50.2% for Pi and 43.9% for Hermes on their respective search providers. Portable uses the least wall time and the fewest tokens. Inference and private documents stay local; only search touches the web.

Web agent economics are set by inference volume and each step is one inference call. A simple task (extract a field, fill a form, navigate a site) is 10 to 20 calls. A multi-site workflow runs into the hundreds. From @RaiseSummit last month: @abhshkdz, @yutori_ai co-founder/CEO, on how they ship Navigator at 2-5x cheaper and faster than closed models on web tasks: -> ~80% prefix cache hit rate -> Speculative decoding on smaller footprint models -> Post-training Qwen models with SFT and RL They run on Together, where customizable endpoints and auto-scaling let them A/B test new model variants in minutes.