Skip to main content

Assistant Configuration 

An Assistant is the template every Voilo live instance uses when it joins a call. The settings below decide its vibe, voice, and secret super‑powers.

Prompt Template 🍰​

What’s a Prompt Template?​

A short mission briefing we hand the LLM before show‑time. It sets tone, role, and boundaries - think concise instructions for AI.

Predefined Prompts​

We’ve got a few ready‑made templates for common use cases. You can use them as is or customize them to your needs using the Platform Console.

TemplateElevator Pitch
1‑on‑1 Meeting AssistantKeeps personal check‑ins laser‑focused and agenda‑friendly.
Project Management AssistantHerds tasks, blockers, and next‑steps like a scrum master on caffeine.
Job Interview AssistantScores answers and surfaces red‑flags while HR smiles politely.
Sales AssistantNotes objections, surfaces competitor mentions, and pre‑writes follow‑up emails.
Customer Support AssistantPulls ticket IDs, promises, and sentiment spikes into crisp summaries.
Speaking CoachCounts filler words and compliments your um‑mazing progress.
Education AssistantTurns lectures into flashcards and checks concept mastery on the fly.

JSON Snippet​

{
  "prompt_template": "You are a Sales Assistant embedded in a live sales call. Your mission:
  • Spot and capture objections around price, features, or timing.
  • Identify decision makers versus influencers.
  • Record actionable next steps including owners and deadlines.
  • Watch out for competitor mentions and flag them with ⚠️.
  • Notice positive buying signals and mark them with ✅.
  • Recommend personalised follow‑up resources (case studies, demos, free‑trial links) at wrap‑up.
Return a JSON object using the exact schema in meeting summary:
{
  \"objections\": [],
  \"decision_makers\": [],
  \"next_steps\": [],
  \"competitor_mentions\": [],
  \"buy_signals\": [],
  \"recommended_resources\": []
}"
}

Transcription Engine 🎙️​

Choose who turns raw audio into text.

Supported Providers & Models​

OpenAI:

  • gpt-4o-mini-transcribe – fastest OpenAI ASR; tuned for low‑latency live calls with solid accuracy.
  • gpt-4o-transcribe – flagship model; top‑tier accuracy, multilingual coverage, handles noisy environments.
  • whisper-1 – cost‑efficient classic; great for long recordings and large batch jobs.

Deepgram:

  • nova-3 (English) – flagship Deepgram ASR; top‑tier accuracy for English, excels in noisy & crosstalk‑heavy rooms.
  • nova-3-multilingual – same flagship engine; full multilingual coverage (30+ languages) without sacrificing accuracy.
  • nova-2 – mid‑tier classic; solid accuracy for non‑English calls at a friendlier price.
  • enhanced – accuracy‑boosted upgrade; lower WER than base and supports keyword boosting.
  • base – volume‑optimised workhorse; lowest cost per minute for massive transcription backlogs.

Real‑time vs Delayed​

Trade‑off cheat‑sheet

  • Real‑time - answers instantly; premium price.
  • Delayed - ≈ 25 % cheaper and better word‑error‑rate, but transcript lands ≤ 60 s after words are spoken.

JSON Snippet​

{
  "transcription": {
    "provider": "deepgram",
    "model": "nova-3",
    "mode": "delayed"
  }
}

Diarization 🗣️​

Let the Assistant tag who said what so follow‑ups don’t become detective work.

  • Toggle: diarization: true | false (Deepgram‑only)
  • Output looks like:
[Speaker:0] We should ship on Friday.
[Speaker:1] Friday works for me!

Experimental Voiceprint (Beta)​

  1. Contact support to enable the feature.
  2. Upload ≥ 5 min of clean speech per teammate (.wav or .mp3).
  3. Next call, the Assistant labels real names instead of generic Speaker IDs.

JSON Snippet​

{
  "diarization": {
    "enabled": true,
    "voiceprint": true // contact support first!
  }
}

Key Terms 🏷️​

Feed the Assistant a company‑specific glossary so it never misspells “K8s‑as‑a‑Service”.

Custom vocabulary “boosting” helps the speech‑to‑text engine nail awkward brand spellings and niche jargon on the first pass - dramatically lowering word‑error‑rate.

  • Input: text files (.txt, .md, .csv, .pdf).
  • The extractor highlights matches and synonyms during calls.

Great candidates to upload:

  • Brand & product names: AcmeCloud™, RocketDB
  • Industry jargon & abbreviations: ARR, CAC:LTV, Kubernetes, K8s
  • Employee & stakeholder names: Maria Zhang, J‑P Michaud
  • Compliance & legal phrases: GDPR Article 25, SOC‑2
  • Custom acronyms & frameworks: RACI, PRINCE2

JSON Snippet​

{
  "key_terms": {
    "enabled": true,
    "files": ["glossary.txt", "acronyms.md"]
  }
}

Example Full Config 🚀​

How to use: Copy‑paste this payload into POST /v1/assistants (our public API) or configure the same settings in the Console under Settings → Assistants. Every JSON field maps 1‑for‑1 to a toggle, dropdown, or text box in the UI - no hidden magic.

{
  "prompt_template": "You are a Project Management Assistant that identifies blockers and next steps.",
  "transcription": {
    "provider": "openai",
    "model": "gpt-4o-mini-transcribe",
    "mode": "realtime"
  },
  "diarization": {
    "enabled": false
  },
  "key_terms": {
    "enabled": true,
    "files": ["glossary.txt", "acronyms.md"]
  }
}

Need something wilder? 🚀

Adjust the settings above, hit Save, and your next Voilo Assistant will join the meeting with its updated configuration.