feat(workflows): make the voice-to-voice template real

[Hoang130203]

Summary

config/n8n/05-voice-to-voice.json was a placeholder — manualTrigger +
sticky note + "connections": {} — behind a catalog card advertising "Speak,
get AI response as audio", "featured": true, "setupTime": "2 minutes".

It is now the full loop, on your own hardware:

Webhook POST /webhook/ods-voice  (multipart audio)
  -> Audio Attached? (IF)
       false -> Return 400
       true  -> whisper:8000      /v1/audio/transcriptions
             -> llama-server:8080 /v1/chat/completions
             -> tts:8880          /v1/audio/speech   (responseFormat: file)
             -> Return Audio      (binary)

curl -X POST http://localhost:5678/webhook/ods-voice \
  -F 'file=@question.wav' --output answer.mp3

Optional form fields: voice (default af_heart, matching the tts README),
system (override the assistant prompt), stt_model.

Every hop uses the manifest's in-network port, not the published one, so
neither the audio nor the transcript ever leaves ods-network.

Three details this needed to actually work

• The TTS response is binary. The node sets
  options.response.response.responseFormat = "file" with
  outputPropertyName: "speech", and the responder uses
  respondWith: "binary" / responseDataSource: "set" /
  inputFieldName: "speech". Without the file format n8n would try to parse
  the MP3 as JSON and fail at the last step.
• The system prompt is written for speech, not screen — two or three short
  sentences, no markdown, no lists, no code blocks, because the reply is read
  aloud. A default chat prompt produces bullet points that Kokoro then reads
  out as literal asterisks.
• The upload is referenced as file, not data. n8n's handleFormData()
  names a webhook binary property after the form field key, and setting
  options.binaryPropertyName appends a counter (data0) rather than
  renaming it — see feat(workflows): make the voice transcription template real #2498 for the code path.

AI Assistance

AI assisted with drafting the node graph, the spoken-output prompt, and this
description. I verified all three endpoints and ports against their service
manifests, and confirmed the binary-response parameter shapes against the node
definitions rather than from memory.

Release Lane

• Stable hotfix targeting release/2.6.x
• Mainline change targeting main
• Next-minor work targeting the next feature/minor release
• Not sure; reviewer should help classify

Stable hotfix reason:

n/a

Changed Surface

• Docs only
• Tests only
• Dashboard UI
• Dashboard API / host agent
• Installer / bootstrap / lifecycle
• Docker Compose / service manifests
• Model routing / Hermes / capabilities
• Network exposure / auth / proxy
• Dependencies / runtime wiring

(One JSON file under config/n8n/. An import payload for n8n; no ODS code
executes it. The catalog entry is unchanged.)

Risk And Validation

• Risk level: Low
• Validation run:
  • git diff --check
  • Markdown/link sanity for docs
  • Focused tests listed below
  • Dashboard lint/test/build
  • Extension audit / compose validation
  • Release-grade fleet or scoped hardware validation
  • Stable-lane patch validation, if targeting release/2.6.x

Commands/results:

# Validated against the exact node package ODS ships. compose.yaml pins
# n8nio/n8n:2.6.4; `npm view n8n@2.6.4 dependencies.n8n-nodes-base` -> 2.6.2.

$ node verify.js 05-voice-to-voice.json
n8n-nodes-base version: 2.6.2
node types loaded: 417
  checked 05-voice-to-voice.json: 8 nodes

ALL WORKFLOWS VALID

# Binary-path parameter shapes checked against the node definitions:
#   httpRequest options.response.response values
#     -> fullResponse, neverError, responseFormat[autodetect|file|json|text],
#        outputPropertyName
#   respondToWebhook respondWith options
#     -> allIncomingItems, binary, firstIncomingItem, json, jwt, noData,
#        redirect, text
#   respondToWebhook responseDataSource -> automatically | set
#   respondToWebhook inputFieldName     -> string
#
# Endpoints checked against extensions/services/*/manifest.yaml:
#   whisper      port 8000  (published 9000)
#   llama-server port 8080  (published 11434)
#   tts          port 8880

Caveat, and it is a bigger one for this workflow than the others: Docker is
not running on my dev host, so I could not run the three-hop chain end to end.
Static validation proves the file imports and that every node parameter and
version is real; it does not prove the data actually threads through three
services. This is the longest chain of the set, so it is the one I would most
want a live run on. Say the word and I will get one on a machine with Docker
before you merge.

Operational Change Check

An import payload for n8n. Nothing in the installer, compose stack, ods-cli,
or dashboard-api executes it. No existing install changes until a user imports
it.

• This is not an operational change.
• This is an operational change and validation is recorded above.
• This is an operational change and validation is intentionally deferred for:

Notes For Reviewers

This is the workflow whose catalog dependencies were wrong. It declares
["whisper", "llama-server", "kokoro"], and kokoro matches no service, so
check_workflow_dependencies() silently reports it satisfied — the card says
"ready" even with TTS stopped. #2495 fixes that with a kokoro -> tts alias.
Worth landing that one first, or this workflow will look installable on a box
with no TTS and fail at the last hop.

Voice list. af_heart is the tts README's example voice. If ODS pins a
different default anywhere I did not find it, point me at it.

Part of the series making the 18 stub templates real: #2496 (chat,
code-assistant), #2497 (summarizer), #2498 (transcription). Independent files,
no overlapping lines.