Thanks — this is about whether the timeline/Agent Observability view (full transcript, pipeline traces, token counts, etc.) that you saw in a LiveKit demo can be used in a self-hosted environment. Quick answer and options below. Short answer • No — the Agent Observability timeline (with synchronized audio playback, full transcripts, turn-by-turn pipeline traces, token counts, time-to-first-token and related dashboard features) is only available as part of LiveKit Cloud and is not provided out‑of‑the‑box for self‑hosted deployments. • Yes — you can recreate similar functionality yourself by exporting the agent/session telemetry and building a custom observability dashboard with third‑party tools. How you can recreate it when self‑hosting • Export metrics & events from your agent: ◦ The agent SDKs emit observability events (for example

MetricsCollected

) that include LLM token usage, latencies, turn detection, and other metrics. ◦ Log pipeline traces, transcripts, and timestamps from your agent process where you have access to the audio/transcription pipeline. • Send telemetry to monitoring/visualization tools: ◦ Push metrics to Prometheus/Grafana for charts and latency/token metrics. ◦ Push traces/logs to an OpenTelemetry-compatible backend, Jaeger, or Elastic for trace views. ◦ Store transcripts and aligned timestamps in a DB and build a UI to show a synchronized timeline and audio playback. • Build a timeline UI: ◦ Use the stored transcripts + timestamps to render a searchable timeline. ◦ Overlay metrics (token counts, time-to-first-token) and pipeline trace entries at the corresponding timestamps. ◦ Add audio playback synced to the timeline if you record session audio (session recording must be enabled). • Practical pieces to capture: ◦ Transcripts with start/end timestamps ◦ Token counts per response or per turn ◦ Latency metrics (e.g., time-to-first-token) ◦ Pipeline traces / stage logs for each agent turn ◦ Session audio or recording URLs for synchronized playback If you’d like, I can help with next steps — tell me which parts you want to build first: • Capture metrics from the agent (where to hook into events and examples) • Storing transcripts and audio (format and schema suggestions) • Visualization options (Grafana panels vs a custom web UI) • Example data model or a reference architecture for self-hosted observability Which of those would you like to start with? Sources: Streamline troubleshooting with Agent Observability | Streamline troubleshooting with Agent Observability | Streamline troubleshooting with Agent Observability | OpenAI and LiveKit partner to turn Advanced Voice into an API | Pricing | LiveKit