See what your agent did, without reading the log.
Ben draws each turn as cards: what it planned, what it changed, what it checked, what it found and what it needs you to decide.
01Cards, not a transcript
Plan, change, verification, finding, verdict, metric, decision and caveat each have a card of their own, drawn the same way every time. A before and after pair outlines what changed; a metric shows the number it moved from and to.
02Failures say so
A check that failed shows as failed, never rounded up to passed. A turn that broke off offers Continue; a send that stalled offers Retry and gives the composer back.
03The evidence stays with the turn
Screenshots, before and after shots, numbers and the files touched stay attached to the turn that produced them, so a claim can be checked where it was made.
- Does it replace Claude Code?
- No. Claude Code (or another agent) does the work; Ben runs it in a sandbox and shows what it did.
- Which models does it run?
- Ben, at the level you pick per channel. Your own Claude or ChatGPT login works too.