Runtime · Rome · 2024
Kernel
The model should not own the loop. Kernel does: named tools, memory, retries on a dropped connection, spend caps, the stop. Operator, Fleet, and Arena share it so a score from last month still means something. A tool that lives only in a notebook is a demo. If an agent runs in production here, it runs on this runtime.
Tools, retries, spend caps, a stop. The model does not own the loop.
Blog: How does tool calling work in an AI agent that edits files, runs tests, and calls APIs?
Next