Hello, We are incredibly excited to officially open the waitlist for the alpha version of AI Ally! :backhand_index_pointing_right: Sign up for the Alpha Waitlist Here :backhand_index_pointing_right: Watch the Video Demo Here [AI Ally Teaser] A quick disclaimer before we dive in: This is a very …

Sorry I’m a bit late to the party. I didn’t fully understand: does Bobrik rely on cloud or does it run entirely locally?

[image] raph: does Bobrik rely on cloud or does it run entirely locally? These two quotes from Kaye answer your question: [image] kaye: At the moment, Bobrik in this alpha test is powered by Claude Sonnet on our infra and through Anthropic’s API—this is communicated to all testers. …

Thank you for highlighting those! I’ll also wait for a beta that runs locally :saluting_face:

Theoretically, we have the functionality today to already use local models with Bobrik (and we have tried it ourselves)—it’s simply not optimised for it yet. The main issue is that local models and device power are too limiting today for stuff like mini apps and massive bulk-editing—especially on mo…

[image] kaye: Claude Sonnet on our infra and through Anthropic’s API If you are running on a cloud provider and I were to use the feature, I’d much rather have my data going to something along the lines of Amazon Bedrock running on a cloud and country close to your hearts than a specific prov…

[image] niemeyer: This is simply not true, Kaye. Many of us will have devices that are more than enough for working with notes in our own machines, and in some cases even our phones. I agree that many people have enough, no doubt. But there are plenty of users (from what we can see) who are d…

To be clear, and I said that before, I’m not against having a cloud-based offering, as today unless you do that the vast majority of users will simply not use the feature at all. I’m simply suggesting that having the ability to set the endpoint towards a local standard completions API should be supp…

I’ve been experimenting with AnyType’s AI agent a bit more intensively over the weekend. At some point on Sunday I had one of those moments where you stop and think “oh, this is actually different.” My use case was building a geopolitical and systemic risk dashboard. Group signals, hypotheses, anal…

[image] betree: Reusable prompt templates or skills: I can define once and invoke by name would make recurring workflows much faster. I actually did something like that (or rather, the agent did it). In the 1-on-1 chat, it learned how to create Feedback objects on request (with properties to…

I have three questions for @Kaye : Am I correct in understanding that a Native Agent is a local model? How large will that be approximately in GB? Will I need my own server to host that model? If it isn’t located locally on my machine, what path will the requests I send to a Native Agent take? …

AI Agent on Anytype — Alpha Demo

News

kaye September 23, 2026, 11:14am 71

Cross-posting this great writeup from @niemeyer

It explores the new architecture we’ve created for Anytwo and getting agents to work privately on the tech stack.