2 min read
Reading the wire — what the runtime actually streams
Every agent turn surfaces as structured SSE events. Here's the event vocabulary, and why the web app reads the exact same stream your client would.
Blog
Weights & Signals
Writing on supervised multi-agent execution, durable memory, and the thinking behind Arbiter — from the people building it.
2 min read
Why we built a runtime for supervised agents
Most agent failures aren't model failures — they're the absence of anything watching the model. Arbiter puts a supervisor in the loop at the runtime layer.