Trust, Governance & Security · 09.05
Observability: Live Reasoning, Token Pills & OpenTelemetry
Every run streams reasoning, a live activity tail back-fills a surface opened mid-run, and a realtime token pill shows spend as it accrues.
Nothing has to be inferred after the fact
A run's reasoning streams live over the same realtime rooms every other cross-cutting system uses — open a run's detail view mid-flight and an activity tail immediately back-fills what already happened, rather than only showing events from the moment you opened the tab forward. A token pill accrues spend in realtime as a run actually consumes it, so cost is visible while a run is happening, not just totaled up afterward in a report.
OpenTelemetry is opt-in, and genuinely absent when unset
Setting AGENTIS_OTEL_ENDPOINT turns on real OTEL export; leaving it unset falls back to a no-op tracer rather than silently buffering spans nobody's collecting. That matters for the "zero external services to start" promise — observability tooling is something you opt into when you need it, not baseline overhead the slim install pays for whether you use it or not.
Continue
Ask, Plan, and Auto, plus deployment + per-workspace autonomy toggles that decide how much a run can do unsupervised.
The Experiment primitive: define, assign, record, and read results — A/B-style assignment and outcome tracking, feeding the same trust + learning loop the Brain uses.