AI Transparency

Last updated: April 9, 2026

Tonos is built on AI. We believe you deserve a clear, honest explanation of what the AI does, how your data moves through it, and what that means for you. This page explains the technology in plain language.

1. What Tonos Does

Tonos is a voice cloning service — not voice audio cloning, but writing style cloning. You give Tonos examples of how you write: Slack messages, emails, text conversations, professional outreach. Tonos uses AI to analyze those samples and build a structured model of your writing style — your "voice profile." When you or an application asks Tonos to draft a message, it uses that voice profile to generate text that sounds like you wrote it.

The goal is to make AI-generated outreach indistinguishable from messages you wrote yourself. Tonos does not generate audio, video, or any media other than text.

2. How It Works — Data Flow

Here is what happens to your data at each step:

1. You submit writing samples — via the web app or API. These are the raw texts you provide to train your voice profile.
2. Extraction — Your samples are sent to Anthropic's Claude API along with a structured extraction prompt. Claude analyzes the samples and returns a structured voice profile: formality, warmth, directness, punctuation habits, vocabulary patterns, and other style dimensions.
3. Samples are redacted or kept (your choice) — By default, raw sample text is redacted after extraction: samples with cached analysis data have their text replaced with "[redacted]", and samples without analysis data are deleted entirely. You can opt to keep your samples by checking "Keep samples after generation" before extracting — this lets you add more samples later and regenerate your profile.
4. Voice profile stored — The structured profile (a JSON document describing your writing style) is stored in your account. This profile contains derived style dimensions, not copies of your original messages.
5. Context sub-profiles — If you have enough samples across different contexts (email, DM, professional), Tonos automatically extracts context-specific sub-profiles that capture how your voice shifts per context. During this process, the top 3 highest-quality samples per context are retained as exemplars to improve future generation accuracy (see Section 4).
6. Draft generation — When you request a draft, your voice profile (or the matching context sub-profile) and the draft context (recipient, platform, instructions) are sent to Claude. Claude generates a message in your style. The generated draft, along with its context and metadata, is stored server-side to enable your draft history, feedback, and quality measurement.

3. The AI Model We Use

Tonos uses Claude by Anthropic for all AI operations, including voice profile extraction and message generation. We use Claude through Anthropic's API.

Key facts about how Anthropic handles API data:

• Anthropic does not use API inputs or outputs to train its models by default
• API data is not retained by Anthropic beyond what is needed to process the request
• We do not fine-tune, train, or customize the Claude model using your data
• We use structured output features (Claude's built-in JSON schema enforcement) to ensure reliable, deterministic voice profile extraction — not prompt tricks

For full details on Anthropic's data handling practices, see Anthropic's Privacy Policy.

4. What We Store

• Voice profiles — Structured JSON describing your writing style dimensions, detected patterns, and context-specific variations. Retained while your account is active.
• Generated drafts — Every draft you generate is stored server-side, including the generated text, recipient, context, source text (for rewrites), and platform. This enables your draft history, the feedback system, and internal quality measurement. Drafts are retained while your account is active.
• Retained exemplar samples — When context sub-profiles are extracted, the top 3 highest-quality writing samples per context type (email, DM, professional, casual) are automatically retained as exemplars. These are used as few-shot examples during generation to improve accuracy. Up to 20 exemplars are retained per profile, with oldest evicted when the cap is reached. If embedding support is enabled, retained exemplars also store a vector embedding for similarity-based retrieval. Retained exemplars are stored until you reset the corresponding context profile or delete your account.
• Writing samples (if you opt to keep them) — If you check "Keep samples after generation," your raw writing samples remain in your account. You can delete individual samples, clear all samples, or let them be redacted on the next extraction. Samples you do not opt to keep are redacted or deleted after extraction as described in Section 2.
• Feedback data — When you accept, edit, or reject a generated draft, that feedback is stored and used to measure and improve generation quality for your profile.
• Account and billing data — Your email, subscription status, and usage metrics. See our Privacy Policy for details.

5. What We Don't Store

• Raw writing samples (by default) — Unless you opt to keep them, raw sample text is redacted after voice profile extraction. Samples with cached analysis have their text replaced; samples without analysis are deleted entirely.
• Anthropic API request logs — We do not log the full prompts or responses exchanged with Claude's API. Only structured outputs (voice profiles, generated text) are stored as described above.
• Reversible data — The voice profile is a derived abstraction. It describes patterns (e.g., "casual formality, frequent use of lowercase") but cannot be reversed to reconstruct your original writing samples.

6. AI Limitations

Tonos does not guarantee that generated content will perfectly match your writing style in all contexts. Voice profiles are statistical approximations. Generated text may contain factual errors, awkward phrasing, tone miscalibrations, or content that does not reflect your actual views. You must review all generated content before use.

7. Your Responsibility

AI-generated content is a tool. You are responsible for how you use it.

• Review before sending. Always read generated content before sending it. AI can make factual errors, miscalibrate tone, or produce content that is inappropriate for your context.
• Own the output. When you send a message generated by Tonos, you are sending it — legally, ethically, and reputationally. The AI generates suggestions; you make the decision to use them.
• Accuracy is not guaranteed. Tonos does not verify facts, claims, or information in generated drafts. Do not use Tonos to generate content you haven't personally verified for accuracy.

Disclosure obligations

Under guidance from the Federal Trade Commission (FTC), businesses must not make deceptive claims about AI capabilities. We do not claim that Tonos output is human-written. If you send AI-generated content in contexts where recipients would reasonably expect a personally written message (such as one-on-one emails or direct messages), you may have an ethical or legal obligation to disclose that AI assisted in drafting. Tonos does not automatically add disclosure labels to generated content — the decision to disclose is yours and may depend on your jurisdiction, industry, and relationship with the recipient.

8. Ethical Boundaries

Tonos is designed for one core use case: helping you communicate as yourself, more efficiently. We have built guardrails into the product and our policies to prevent misuse:

• No impersonation without consent. Using Tonos to generate content in someone else's voice requires their documented written consent. This is required by our Terms of Service and Acceptable Use Policy.
• No fraud or deception. Tonos may not be used to deceive recipients about who authored a message, to commit fraud, or to misrepresent the identity of the sender.
• No harmful content. The Service may not be used to generate threatening, harassing, or illegal content. See our Acceptable Use Policy for the full list of prohibited uses.

We monitor for abuse signals and will investigate and act on credible reports. Accounts found violating these rules will be suspended or terminated. See our DMCA & Abuse Reporting page to report misuse.

9. Voice Cloning and the Law

Writing style is increasingly recognized as a protectable aspect of personal identity. Several U.S. states have enacted or are considering laws addressing AI-generated likeness and voice:

• The Tennessee ELVIS Act (Ensuring Likeness, Voice, and Image Security Act, 2024) establishes that a person's voice — including distinctive writing patterns that constitute a recognizable "voice" — is a property right. Using AI to replicate someone's voice without authorization may violate this law.
• Multiple states have enacted or proposed deepfake legislation that may apply to text-based impersonation when used for fraud, harassment, or deception.
• Federal legislation on AI-generated content disclosure continues to evolve.

Tonos is designed for you to clone your own writing voice. If you use Tonos to generate content in another person's style, you must: (a) obtain their documented written consent before creating a profile from their writing samples; (b) retain proof of that consent; (c) comply with all applicable voice, likeness, and identity protection laws in your jurisdiction and the subject's jurisdiction.

10. Data Rights and Deletion

You have full control over your data:

• You can view your voice profile and context sub-profiles from the Profile and Settings sections
• You can export your voice profile as JSON from the Profile page
• You can delete your voice profile at any time from Settings
• You can delete individual writing samples or clear all samples from the Profile page
• You can reset context sub-profiles individually from Settings
• Deleting your account permanently removes your voice profile, drafts, retained exemplars, feedback data, and all associated data within 30 days

For full data handling details — including data retention periods, your rights under GDPR and CCPA, and how to request data export — see our Privacy Policy.

Contact

Questions about how Tonos uses AI, or concerns about your data? Reach us at max@tonos.fyi.