Tone Matching

By analyzing emails, messages, and written notes through the Fastino Personalization API, assistants can learn how the user naturally writes — then produce output that sounds indistinguishable from their authentic voice.

Overview

Voice mirroring enables agents to:

• Write or reply in the user’s tone — formal, casual, concise, or warm.

• Adjust vocabulary and punctuation habits automatically.

• Maintain stylistic consistency across tools (email, Slack, CRM, etc.).

• Preserve authenticity while ensuring alignment with professional or contextual tone.

Fastino provides this capability through tone embeddings, phrase-level style memory, and deterministic persona summaries accessible via the API.

Key Components

Example: Ingesting Email and Message Samples

To teach Fastino a user’s writing style, agents can ingest historical communication data:

POST /ingest
{
  "user_id": "usr_42af7c",
  "source": "gmail",
  "documents": [
    {
      "doc_id": "doc_voice_1",
      "kind": "email",
      "title": "Follow-up Email Example",
      "content": "Hey team — great job today. Let’s tighten the copy and ship tomorrow. Thanks!"
    },
    {
      "doc_id": "doc_voice_2",
      "kind": "email",
      "title": "External Client Update",
      "content": "Hi Alex, thanks for the update! We’ll review and confirm by EOD. Best, Ash."
    }
  ]
}

These samples help Fastino learn distinctions between internal tone (“casual, fast”) and external tone (“polite, structured”).

Example: Querying for Voice Style

POST /query
{
  "user_id": "usr_42af7c",
  "question": "How does Ash typically write in emails?"
}

Response

{
  "answer": "Ash writes with a concise, upbeat tone. Uses short sentences, minimal punctuation, and polite closing lines like 'Thanks!' or 'Appreciate it.'"
}

Agents can use this answer directly to condition prompt generation or decoding behavior.

Example: Retrieving a Deterministic Voice Summary

For faster, repeatable mirroring, agents can fetch a tone profile via summary:

Response

{
  "summary": "Ash’s tone is concise, positive, and slightly informal. Frequent use of active voice, contractions, and friendly sign-offs. Avoid long intros or passive phrasing."
}

This summary provides a ready-to-use voice blueprint for email or text generation.

Example: Voice Mirroring in Action

An email-writing assistant might use this data like so:

Prompt Template

Generated Output

The response reflects Ash’s natural rhythm and personality while maintaining precision.

Example Implementation (Python)

import requests

BASE_URL = "https://api.fastino.ai"
HEADERS = {"Authorization": "x-api-key: sk_live_456", "Content-Type": "application/json"}

def get_voice_summary(user_id):
    r = requests.get(f"{BASE_URL}/summary?user_id={user_id}&purpose=voice-style", headers=HEADERS)
    return r.json()["summary"]

def mirror_voice(user_id, message):
    tone = get_voice_summary(user_id)
    return f"User voice style: {tone}\n\nCompose reply in this style:\n{message}"

# Example
print(mirror_voice("usr_42af7c", "Can we reschedule our sync?"))

This workflow can be integrated into an LLM generation pipeline, ensuring consistent tone replication.

Combining with Other Use Cases

Together, these create an end-to-end adaptive communication system.

Multi-context Voice Control

Fastino can differentiate tone by communication context (internal, external, formal, casual):

POST /query
{
  "user_id": "usr_42af7c",
  "question": "How should I adjust tone when Ash writes to clients vs teammates?"
}

Response

{
  "answer": "Ash uses a friendly, relaxed tone internally but adopts concise professionalism in client emails — avoids emojis, includes sign-offs, and maintains clear structure."
}

This allows nuanced voice-shifting across contexts.

Feedback Loop: Improving Voice Accuracy

When users edit or reject generated text, log it as a correction event to refine the tone model.

POST /ingest
{
  "user_id": "usr_42af7c",
  "source": "email_agent",
  "events": [
    {
      "event_id": "evt_voice_feedback_1",
      "type": "correction",
      "timestamp": "2025-10-27T17:00:00Z",
      "content": "User removed greeting — prefers messages that start directly with the action item."
    }
  ]
}

Future summaries and queries will incorporate this preference automatically.

Use Cases

Best Practices

• Use purpose=voice-style summaries for consistent tone grounding.

• Ingest at least 5–10 representative writing samples per context (work, social, formal).

• Avoid overfitting — retain minor stylistic variance for natural feel.

• Capture ongoing corrections via /ingest to improve accuracy.

• Pair with persona summaries for full stylistic fidelity.

• Use separate models for tone inference and content reasoning to prevent drift.

Example: Deterministic Voice Summary

Response

{
  "summary": "Ash’s writing voice is direct, confident, and human. Uses short sentences, light warmth, and avoids excessive qualifiers. Ends messages with friendly affirmations like 'Sounds good!' or 'Thanks!'"
}

This summary can be applied across writing agents to maintain a cohesive and authentic communication identity.

Summary

The Voice Mirroring use case allows agents to communicate in a way that feels authentically human and personally familiar.
By learning from user-written data and applying deterministic tone summaries, Fastino enables your assistants to speak in the user’s own voice — across every tool, message, and context.

Next, continue to Personalization Use Cases → Personalized Retrieval to learn how to fetch the most relevant user-specific information for grounding and reasoning.