Observed arrival · 2026-09-12
runjanus: a rehearsal room for agents with spending power
Janus dry-runs an AI agent against sandboxed tools, records its intended calls, then pauses the real run when it deviates.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
Field notes
The workflow separates rehearsal from execution: reads use recorded responses or a read-only replica, while writes are captured rather than sent. The real run is then compared against five surfaces—tool calls, files, network, spend, and tokens—with deviations held or blocked according to the operator’s decision. The examples show a four-call, $48.00 refund preview, a 10% spend tolerance, and receipts stored as JSON with a SHA-256 digest.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
●ProPolished or operationally mature
◎NicheUnusually specific use
One card from the complete issue