Engineering notes from the team building 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.
No articles match your search.