Overview
cueloop is a terminal-first review surface for coding agents. When your agent has a plan or a diff, you review it in the terminal, annotate the exact lines, and send a verdict back - then share it over SSH.
Coding agents move fast. cueloop puts a deliberate human gate between what the agent proposes and what it does. Instead of skimming a wall of text and hitting “yes”, you open the plan or the diff in a real terminal UI, mark up the parts that matter, and hand back a structured verdict the agent can act on.
The shape of it
- One primitive - the review session. A review session pairs an artifact (a plan, a working-tree diff, or a pull request) with annotations anchored to the exact text they discuss. Everything else is built on it.
- Two directions of feedback. You review your agent’s work and send a verdict back to the agent. Separately, you can share a plan with a teammate over SSH and merge their annotations back into yours.
- No browser, no account. A local daemon speaks over a unix socket; the UI is an OpenTUI React app. Sharing runs over plain SSH.
Where to go next
- New here? Start with Install & quickstart.
- Want the model? Read The review session and Annotations.
- Ready to collaborate? Sharing over SSH walks the whole loop.
Terminal-first
cueloop never opens a browser. If it happens in cueloop, it happens in a
terminal - locally, or over SSH on someone else’s.