Notes d'ingénierie de l'équipe CloudSignal.
RSSAnthropic shipped agent identity with Claude Tag. Here's what it gets right, the three layers it leaves open on the wire, and how an identity-bound broker closes them.
How ACL v2's identity-bound topic patterns kill per-user rule explosion, what we found about wildcard SUBACK along the way, and what v2.2 added on top.
Why we chose a CLI over an SDK, how browser-based login works, and how policy and identity claims now flow from your terminal.
How we built a drop-in for the Vercel AI SDK that gives AI chat offline recovery and multi-device sync, in one config change.
The little signals that make AI chat feel responsive: who is here, what the model is doing, and how to render that without races.
TLS-only listeners, deny-by-default ACLs, and per-org mountpoints. What we picked and why.
Presence, ordered message delivery, and per-document access control. Everything else is product polish.
We added VAPID-based web push so apps can reach users when the tab is closed, without taking a dependency on FCM.
How short-lived MQTT credentials remove the rotation, revocation, and refresh problems baked into long-lived JWTs.
Learn how to connect AI agents using MQTT pub/sub messaging. We'll build a three-agent pipeline that handles research, summarization, and notifications.
43% of product teams report inter-agent communication as their biggest latency bottleneck. Here's why multi-agent systems fail, and the architectural pattern that solves it.
How we replaced polling and SSE plumbing with a topic-per-user pattern that scales without code.
Why we ship a REST publish endpoint alongside MQTT, and when each one is the right tool.
Why naive presence drifts within minutes, and how MQTT's last-will primitive fixes it for free.
What it took to make MQTT pub/sub feel like a managed primitive, not a broker you babysit.
Aucun article ne correspond à votre recherche.