Dograh Integration Overview

Connect your Dograh voice agents to Noveum: every completed call exports automatically as a full trace with transcripts, latency, tokens, and audio

Dograh is an open-source platform for building AI voice agents with a visual, no-code workflow editor. The Noveum integration ships built into Dograh as a workflow node: drop it into your agent's workflow, paste your Noveum API key, and every completed call is exported to Noveum as a structured trace, with no SDK setup and no code.

Each call arrives as a conversation trace with per-turn spans for STT, LLM, and TTS: transcripts, prompts and completions, token usage, per-service latency breakdowns, interruptions, and (optionally) per-turn audio segments plus a full stereo call recording for audio-level evaluation.

How it works

Dograh's voice engine is built on Pipecat, and the integration uses Noveum's Pipecat observer under the hood, so Dograh traces have the same span structure as the Pipecat integration, enriched with Dograh call metadata.

The export is post-call and bring-your-own-key:

During the live call, Dograh records spans and audio locally; no Noveum network I/O touches the call path, so tracing adds zero latency to the conversation.
When the call finishes, Dograh uploads the finished trace to Noveum using the API key configured on the node, then uploads the recorded audio segments.
The trace appears in your Noveum project within moments of the call ending.

Prerequisites

A Noveum account and API key (create one under Settings → API Keys in the dashboard)
A running Dograh instance (cloud or self-hosted) with a voice agent workflow

Setup

1. Add the Noveum node in Dograh

In the Dograh workflow editor, click Add node, scroll to Integrations, and select Noveum. The node is standalone; it doesn't connect to other nodes in the conversation graph.

Dograh workflow editor: adding the Noveum node from the Integrations panel

2. Configure the node

Click the node and fill in:

Field	Value
Noveum API Key	Your API key from the Noveum dashboard
Noveum Project	The project traces should land in (created on first export if it doesn't exist)
Noveum Environment	Environment label stamped on traces: `production` (default), `staging`, etc.
Record audio	Capture per-turn STT/TTS audio and the full-conversation recording (on by default)
Enabled	Toggle on to activate the export

Dograh Noveum node configuration modal

Click Save, then Publish the workflow.

3. Make a call and view the trace

Run a test call through your agent. When it completes, open your project in the Noveum dashboard; the call appears as a new trace. Open it to walk the turn-by-turn timeline, inspect each STT/LLM/TTS span, and play back the captured audio.

A Dograh call trace in the Noveum dashboard with turn, STT, LLM, and TTS spans

What gets traced

Each call produces one conversation trace. The span tree matches the Pipecat integration, with Dograh-specific metadata on the trace:

Trace: pipecat.conversation
│   dograh.workflow_run_id, dograh.call_disposition
│   dograh.agent_version, dograh.mode, dograh.recording_url
│   stt.model_label, llm.model_label, tts.model_label
│   noveum.project, noveum.environment
│   conversation.total_input_tokens, conversation.total_output_tokens
│   conversation.turn_count, conversation.barge_in_rate
│
├── Span: pipecat.turn  (one per user→bot exchange)
│   ├── turn.user_input, turn.duration_seconds, turn.was_interrupted
│   ├── turn.user_bot_latency_seconds
│   ├── turn.latency.ttfb.<service>_ms  (per-service TTFB: stt, llm, tts)
│   │
│   ├── Span: pipecat.stt   - transcript, interim results, confidence, latency, audio
│   ├── Span: pipecat.llm   - message history, system prompt, output, tokens, tool calls
│   └── Span: pipecat.tts   - synthesized text, voice, TTFB, audio
│
└── Span: pipecat.full_conversation  (when Record audio is on)
    └── full-call stereo WAV: left channel = user, right channel = bot

With Record audio enabled, per-turn STT/TTS segments and the full recording are uploaded and attached to their spans, so audio scorers and playback work exactly as they do for SDK-based integrations.

Privacy & payload capture. Exported traces contain transcripts, prompts, completions, and optionally call audio. Review what your agent handles before enabling the export in production, and turn off Record audio if you don't want audio stored.

From traces to evaluation

Once Dograh calls are flowing into Noveum you can:

Debug calls: replay any conversation turn-by-turn with synchronized audio and spans
Build datasets: curate production calls into datasets for regression testing
Score quality: run automated scorers over transcripts and audio to track agent quality across versions (the dograh.agent_version attribute is stamped on every trace)

Troubleshooting

No traces appearing: confirm the workflow is published (not just saved), the node's Enabled toggle is on, and the API key/project have no stray whitespace.
Trace has no audio: confirm Record audio is on in the node configuration.
Calls that never connected show no trace: this is expected, since runs that fail before the pipeline starts (e.g. telephony errors) produce nothing to export.

Learn more

Dograh's Noveum integration guide: the same setup from the Dograh side
Pipecat Integration Overview: full span/attribute reference shared with this integration

Dograh Integration Overview

On this page