DialNexa Troubleshooting Reference

Troubleshoot DialNexa issues with agents, providers, published versions, phone numbers, batch calls, workflows, integrations, webhooks, transcripts, transfers, and exports.

DialNexa troubleshooting starts by finding the object that owns the problem, then checking evidence. Most issues belong to one of these areas: agent version, provider stack, phone number routing, launch data, workflow graph, integration credentials, or monitoring evidence.

Debugging gets faster when every theory has a call id, version number, or workflow lead id attached.

Symptom To First Check

Symptom	First place to check
Agent mishears caller	Recording, realtime transcript, transcriber, language, and Denoising Mode.
Agent heard correctly but answered badly	Prompt, dynamic variables, model family, temperature, functions, and knowledge context.
Agent interrupts caller	Response Eagerness, Deepgram or Soniox choice, transcript boundary, and greeting length.
Agent sounds wrong	Voice provider, voice model, speed, volume, language fit, and pronunciation.
Test call blocked	Published version and recipient phone validation.
Campaign fails many rows	Recipient file, number format, outbound assignment, provider route, and status export.
Workflow lead stuck	Lead history current node, last transition, node outputs, and Application node result.
Integration action fails	Connection validation, missing fields, provider error, and invocation evidence.
Webhook not received	Webhook URL, secret, delivery logs, retry count, and timeout.

Provider-Specific Debugging

Area	What to check
Deepgram	Model selection, English versus multilingual behavior, Flux English-only usage, and transcript boundaries.
Soniox	Hindi-English language hints, Response Eagerness, Denoising Mode, and code-switched transcript quality.
ElevenLabs	Voice model compatibility, speed, temperature, pronunciation rules, and repeated phrase cache behavior.
Cartesia	Sonic model choice, language fit, volume handling, speed, and phone playback quality.
OpenAI or Google	Prompt following, function arguments, structured outputs, and temperature.
Groq	Fallback delay, response length, low-latency behavior, and whether the faster answer still follows instructions.
Plivo or SIP	Active number, caller ID, route assignment, audio quality, compliance readiness, and recording.
Wati or Resend	Credential validation, required fields, template or sender rules, and invocation errors.

General Debugging Routine

1. Find the exact record

   Use call id, batch id, workflow lead id, agent version, or webhook delivery id.

2. Read the evidence

   Open detail views before changing settings.

3. Trace back one layer

   Move from call to route, route to agent version, workflow node to integration, or transcript to transcriber.

4. Change one thing

   Make a focused draft edit.

5. Retest and compare

   Use the same scenario and compare recording, transcript, status, invocation, and extracted fields.

Common Root Causes

Wrong published version

Phone numbers, batches, and workflows can still point at an old version.

Missing variable value

Prompt placeholders need launch values or defaults. Check CSV columns, API payload, workflow lead variables, and fallback values.

No outbound assignment

Batch and workflow calls require an outbound number with a published agent version.

Provider mismatch

A Hinglish agent on an English-only transcriber, a voice that does not match the language, or a slow action inside a live call can all look like prompt bugs.

Integration credential expired

Validate or reconnect the integration before editing workflows or agent functions.

Receiver rejects webhook

Check secret rotation, HTTPS receiver behavior, response codes, and retry handling.

Related Reading

Call Detail Page

Start with evidence.

Voice AI Overview

Understand provider layers.

Agent Versions

Check live version.

Workflow Lead History

Debug workflow paths.