Observed arrival · 2026-09-04
Milk: switch models, not context
Milk is a developer tool that routes prompts between fast and deep AI agents while carrying the conversation across.
Field notes
Milk presents model routing as a terminal workflow rather than a separate chat interface. Each prompt can remain with a fast primary agent or escalate to a deeper agent whose context is reformatted from the existing conversation. The documentation also describes a Percept memory store that persists facts across sessions, loop detection for repetitive output, and an evaluation harness comparing quality, token use, cache efficiency, and latency. Installation is exposed through a shell command, with source linked on GitHub.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue