Skip to main content
When it breaks, you can see where - exactly. That is the whole job of Observability, the part of the Control layer that exposes execution. A communication journey crosses more systems than almost anything else you run: a carrier delivers audio, the speech pipeline turns it into text, a model produces a turn, a tool call fires, an outcome gets written. When a customer says “your agent hung up on me”, the answer has to be findable in one place - not reconstructed from five dashboards. The target is one correlated trace per journey: carrier to speech to model turn to tool call to outcome, on a single trace ID. Today that correlation exists at the call level - call logs, recordings, transcripts, and speech-pipeline observability, all production - and the journey-wide trace is in build on top of it.

​
The primitives

​
On the spine

Observability sits in the Control layer. It writes nothing to the spine; it reads all of it. Every interaction leaves a trail - logs, a recording, a transcript, per-turn metrics - and every event the spine emits is a point on a trace. When the correlated journey trace ships, one trace ID will follow an entity’s journey across channels, so a failed tool call and the carrier leg it happened on appear in the same view.

​
What stays hidden

The telemetry pipeline internals stay hidden: how spans are collected, buffered, and shipped, which storage backend holds traces, how sampling and retention are implemented. You query logs, traces, and metrics; how they got there is not your problem, and keeping it hidden means the pipeline can change without breaking your queries.

​
Status

Call-level observability: production. Call logs, recordings, and transcripts carry production traffic today, alongside speech-pipeline observability for live calls. Correlated journey trace: in build. One trace ID from carrier to outcome, across channels, shipping in the open.

​
Go deeper

Speech observability

Live-call and pipeline observability inside the Speech Stack.

Call logs

Every call’s timeline, status, and hangup cause in the Studio.

Recordings and transcripts

Retrieve the audio and turn-by-turn text of any call.

Analytics

Aggregated metrics across calls and agents.