Show HN: Endo Familiar, an O-cap based JavaScript agent sandbox

cowbertvonmoo

I’m Kris Kowal and I’m the tech lead on the Familiar project and happy to answer questions about building this new kind of user agent for safely managing and extending the capabilities of bots and apps.

hierophantos

I wanted to ask perhaps an impossible question on this notion that I've been chewing on — particularly around the composition of "capabilities" (LLM sense of the word) and what they could acquire through some transitive closure over their "capability set(s)" in a multi-agent scenario. I stumbled into two recent papers, thinking that it was about "capabilities" in the OCap sense and datalog (a related technology to what I'm working on): • https://arxiv.org/html/2603.26725 Capability Safety as Datalog: A Foundational Equivalence; Cosimo Spera; March 2026 • https://arxiv.org/html/2603.15973 Safety is Non-Compositional: A Formal Framework for Capability-Based AI Systems; Cosimo Spera; March 2026 Since capabilities in the OCap sense are just transitive closure of references on some reference graphs, it turns out that these two concerns with conflated terminology are nonetheless the same. Curious if you or anyone has thoughts on managing capability escalations in multi-agent systems. Some thoughts be a-brewing...