Finchvox is an open source observability tool for voice agents built with Pipecat. It replays a development session locally, combining the conversation audio with OpenTelemetry traces, logs, and metrics in one UI, and it highlights voice-specific problems such as interruptions and long user-to-bot latency that general-purpose OpenTelemetry tooling does not show. The current release targets local development; a self-hosted production option is planned.
Integration is three steps in your bot: call finchvox.init(servicename=...), add FinchvoxProcessor to the pipeline after transport.output(), and enable metrics, tracing, and turn tracking on the PipelineTask. The finchvox start command runs an OTLP collector on port 4317 alongside the UI. init accepts endpoint, insecure, capturelogs, and log_modules options; logs from pipecat., finchvox., main, and your project's source files are captured by default. It requires Python 3.11+ and Pipecat 0.0.68+.
Features
- Session replay: conversation audio aligned with traces, logs, and metrics in a single timeline
- Voice-specific signals: surfaces interruptions and user-to-bot latency
- OpenTelemetry-based: ships traces and logs over OTLP gRPC to a local collector
- Pipeline processor: FinchvoxProcessor captures audio from the Pipecat pipeline
- Log capture: application and Pipecat logs sent to the collector, with extra module prefixes configurable
- Local server: finchvox start runs the collector and UI on your machine
- Telemetry control: minimal anonymous usage events, disabled with --telemetry false or FINCHVOX_TELEMETRY=false
