Agent loopqueue 80 · done 0
3 workers · prompt → generate → parseidle
System Onequeue 80 · done 0
3 workers · state + questions → answersidle
Scoreboardshared scale per row
agent loopsystem one
Wall time
0 s
0 s
Cost
$0
$0
Malformed
0
0
ConsolePOST /v1/systemone · jev-latest
Latest typed answer appears here
State
{ }
Questions → answers
Both lanes are modeled, not measured. The agent lane uses illustrative frontier-LLM numbers so the shape is honest: a text generator has to write the answer out, then code has to parse it. System One returns the typed answer directly, and adding questions barely moves its response time.
Same call from a terminal: curl -X POST https://api.typesafe.ai/v1/systemone -H "Authorization: Bearer $TYPESAFE_API_KEY" -d '{"model":"jev-latest","state":{…},"questions":{…}}'