← Mission Control · architecture ↗

LangChain × TypeSafe AI · live trace

Agent Loop, Traced Live

A real mission, replayed step by step. Watch control pass between the LangChain agent and its Jev supervisor — every judgment, decision, and value below was captured from an actual run.

MISSION  —

agent loop · jev taps at the hook points
tool result → next step no tool call → answer before_model wrap_tool_call MISSIONuser input MODELgpt-5-mini TOOLidle FINALanswer → user TrajectoryWatch before_model · score+noul progress — loop — · on-task — idle AutonomyGate wrap_tool_call · choice+noul risk — aligned — idle CommsSentry tool result · noul injection — screens fetched content idle
mission vitals
Progress0%
Loop risk—
Comms integrity100%
Jev judgments
0
Avg latency
—
Agent
gpt-5-mini
Supervision cost
$0
event stream
0/0
final answer · produced by the real agent
LangChain agent / packet Jev judgment · GO / clear consequential · human poll quarantine / block