v1.0 — community-drafted, pending legal review

This document was drafted by the Voxa team in plain English and is intended as a good-faith, founder-grade description of how the service works. It has not yet been reviewed by qualified legal counsel. We are publishing it now so that early users can read clearly what the product does with their data, and so that legal review can be performed against a real artefact rather than a blank page. A formally reviewed v2.0 will replace this document; users will be notified by email at least 30 days before any material change.

Legal

Privacy policy

Effective date: May 9, 2026 v1.0 — community-drafted, pending legal review

Voxa is a voice agent for macOS. The single most important fact about how we treat your data is this: your raw audio never leaves your Mac. Speech recognition runs locally using Apple's on-device Speech framework. Only the resulting text transcript and the small context you have explicitly authorized travel to our servers. This document describes everything else: what we do store, who we share it with, how long we keep it, and the rights you have under the Digital Personal Data Protection Act, 2023 (India), the EU and UK General Data Protection Regulations, and the California Consumer Privacy Act.

01

The short version

Voxa is a voice-to-action assistant for macOS. The wake word “Saara” activates an on-device speech recognizer. Your audio is converted to text on your Mac and discarded; only the resulting transcript text leaves the device. That text travels over HTTPS to our backend, which routes it to a large language model so Voxa can decide what to do, and may dispatch the resulting tool calls (such as sending a Gmail message you authored, or creating a calendar event) to integrations you explicitly connected.

We store your account record, your conversation history, the memory facts you have asked Voxa to remember, and an audit log row per request. The audit log records a SHA-256 hash of the transcript — not the transcript itself — together with the list of tools that ran, hashed identifiers of any recipients (so we can investigate abuse without exposing who you spoke to), and basic latency and success metrics. OAuth tokens for connected services are stored encrypted at rest. You can export, correct, or delete any of this at any time.

02

Who we are

For the purposes of the DPDP Act, 2023, the GDPR, and the CCPA, the data fiduciary / controller / business is:

  • Operator: Adwait Keshari, sole proprietor, trading as Voxa.
  • Address: Bhopal, Madhya Pradesh, India.
  • Privacy contact: privacy@voxa.app
  • Grievance Officer (DPDP, India): Adwait Keshari, reachable at the same address.

Voxa is a small, founder-operated product. There is no in-house data protection team. We commit to responding to every verified request within the statutory timelines (30 days under DPDP, one month under GDPR, 45 days under CCPA), and to escalating to qualified counsel where a request requires it.

03

Data we collect

We collect the smallest amount of data that lets Voxa work. The categories below are exhaustive for the current product; if we add a new category we will update this page and notify signed-in users.

Account data

  • Email address. Required. We use it as your account identifier and for magic-link sign-in. We do not see or store passwords.
  • Display name. Optional. Shown in the menu-bar app and in any Voxa-generated message that signs off on your behalf.

Voice transcripts (text only)

  • The text produced by Apple's on-device Speech framework from what you said. This text is sent to our backend over HTTPS so a language model can plan a response.
  • We do not store the raw transcript text in long-term storage. A SHA-256 hash of the transcript is written to the audit log so we can correlate events without retaining what you said.
  • The transcript may be cached transiently in conversation history if it forms part of an ongoing conversation (see below); in that case it is retained until you purge it.

Conversation history & memory facts

  • Conversation history. Auto-titled threads of your past requests and Voxa's replies, so the assistant can refer back. Visible in the History UI of the macOS app and editable / deletable from there.
  • Memory facts. Short preferences you explicitly ask Voxa to remember — for example, “remember my home address is X”. Stored against your account and viewable / deletable from Settings.

Audit log

For every request that Voxa processes, we write a single audit row containing:

  • transcript_hash — SHA-256 of the transcript. The plaintext transcript is not stored.
  • tools_dispatched — the list of tool names the model chose to run (for example, gmail_send, calendar_create).
  • recipient_hash — SHA-256 of recipient email addresses or contact identifiers, so we can investigate abuse complaints without ever storing who you contacted in plaintext.
  • latency_ms, success, model_id — basic operational telemetry.

OAuth tokens for connected integrations

When you connect Gmail, Google Calendar, Google Meet, Notion, Slack, Telegram, Discord, Twitter / X, or any other supported integration, we receive an OAuth access token (and, where the provider returns one, a refresh token) scoped to the permissions you approved. These tokens are encrypted at rest in the database and decrypted only at the moment Voxa needs to call the corresponding API on your behalf.

Telemetry & crash reports

We use Sentry to capture unhandled exceptions and crash reports from both the macOS app and the backend. Personal identifiers are stripped before send; what we receive is the stack trace, software version, OS version, and a generated session id. You can opt out of telemetry from Settings in the macOS app.

Voice-clone samples (only if you opt in)

If you opt into the ElevenLabs voice-clone feature, the audio samples you upload are sent directly to ElevenLabs' API. Voxa stores only the resulting voice_id identifier in our database; we do not retain the raw samples. You can delete the voice from Settings, which deletes both thevoice_id from our database and triggers a deletion request to ElevenLabs.

MCP server connections (only if you opt in)

You may connect your own Model Context Protocol (MCP) servers. Voxa stores only the connection metadata you provide (endpoint URL, authentication header, a label). Any data sent to an MCP server you have configured travels directly between our backend and that endpoint and is governed by the privacy policy of whoever runs that server.

04

Data we do not collect

  • Raw audio. Your microphone stream is consumed locally by Apple's Speech framework and discarded. It is never uploaded.
  • Long-term raw transcripts. Only the SHA-256 hash is retained in audit logs. The plaintext transcript exists in conversation history only if a conversation thread is active and only for as long as you keep that thread.
  • Plaintext OAuth tokens. All third-party tokens are encrypted at rest.
  • Plaintext recipient identifiers. We hash recipient emails and contact handles before writing them to the audit log.
  • Screen contents at large. Voxa reads the title of the focused window and, when you explicitly invoke it, the text you have selected. It does not screenshot, OCR, or scrape the rest of your display.
  • Browsing history, location, contacts.
  • Advertising or marketing identifiers. No third-party advertising trackers run on this site or in the desktop app.
05

How we use your data

We use the data we collect strictly to operate the service and for the lawful bases described below.

  • Agent execution. Forwarding your transcript to a language model so it can plan and dispatch tool calls on your behalf. Lawful basis (GDPR): performance of contract. DPDP basis: necessary for purpose for which consent was given.
  • Conversation continuity. Storing your past requests so a follow-up like “send the same email but to my brother instead” resolves correctly. Lawful basis: performance of contract.
  • Abuse prevention & rate limiting. Counting requests per account, watching for patterns that look like spam or scraping, and acting on third-party abuse complaints. Lawful basis (GDPR): legitimate interest.
  • Product improvement. Looking at aggregated, hashed audit logs to understand which tools are popular and which ones break. We do not read your conversation content for product research; the audit log contains hashes, not transcripts. Lawful basis: legitimate interest.
  • Security & fraud prevention. Investigating account takeovers, payment fraud, and platform abuse. Lawful basis: legitimate interest and legal obligation.

We do not sell or rent your personal data. We do not share it with data brokers. We do not use it to train any model that we publish or monetise.

06

Who we share data with

Voxa relies on a small, named set of third-party processors. Each one receives only the categories of data needed for it to perform its function. The list below is exhaustive as of the effective date of this policy.

Infrastructure

  • Neon — managed Postgres database. Stores account records, conversation history, memory facts, audit logs, encrypted OAuth tokens, and integration metadata. Neon operates in the United States, the European Union, and India; the Voxa production database is hosted in the Mumbai (India) region.
  • Hostinger — virtual private server hosting for the backend application. The Voxa production VPS is located in Singapore.
  • Cloudflare — DNS, TLS termination, and tunnel transport between the backend and the macOS app. Operates a global edge network.

Language-model providers

Your transcript text and the small context you have authorized (such as the focused window's title and any selected text) may be forwarded to one of the following providers so a language model can plan a response. Voxa routes between these providers based on availability, cost, and the requirements of the specific task.

  • OpenRouter — United Kingdom. Aggregator that routes to one of several underlying models.
  • Google Gemini — United States. Used as a primary planning model.
  • Groq — United States. Used for low-latency inference on supported open models.
  • Anthropic — United States. Used as a planning and tool-calling model for complex requests.
  • OpenAI — United States. Used for tool-calling and as a fallback.

We do not opt API traffic into any provider's model-training endpoints, and we rely on each provider's published commitment that API traffic is not used to train their models. We will update this page if any provider changes that commitment.

Operational tooling

  • Sentry — error monitoring and crash reporting. Operated in the United States and the European Union. Receives stack traces and anonymised session metadata; not your conversation content.

Optional, only if you opt in

  • ElevenLabs — United States. Text-to-speech and instant voice cloning. Voice samples you upload are sent directly to ElevenLabs; we store only the resulting voice_id.
  • OAuth providers — Google (Gmail, Calendar, Meet), Notion, Slack, Telegram, Discord, Twitter / X, and any future integration you connect. We send only the data you have authorised the relevant action to send (for example, the contents of an email you asked Voxa to draft).
  • MCP servers you configure — you choose the endpoint, you accept the risk; the privacy policy of that endpoint governs.
07

International transfers

Voxa is operated from India. Personal data routinely crosses borders to make the product work:

  • The backend application runs on a VPS in Singapore.
  • Persistent storage is in the Mumbai (India) region of Neon Postgres.
  • Language-model inference is performed in the United States or United Kingdom depending on the provider Voxa selects.
  • Cloudflare and Sentry operate globally distributed networks.

Where data leaves the European Economic Area, the United Kingdom, or India, we rely on the lawful transfer mechanisms available to each processor: Standard Contractual Clauses (GDPR / UK GDPR), adequacy decisions where they exist, and the recipient's published commitments under DPDP. Each named subprocessor publishes its own data processing addendum and SCCs, which we have accepted as part of using their service.

08

Retention

  • Conversation history. Kept until you delete your account or explicitly purge a conversation from the History UI in the macOS app. There is no automatic expiry; the choice is yours.
  • Memory facts. Kept until you delete them, either individually from Settings or by deleting your account.
  • Audit logs. Kept for twelve (12) months from the date of the request, then automatically purged. We retain audit logs for this period so we can investigate abuse complaints, resolve disputes (including chargebacks), and meet security obligations.
  • Voice clone voice_id. Kept until you delete the voice from Settings, which also triggers deletion at ElevenLabs.
  • OAuth tokens. Kept until you revoke the integration from Settings, or automatically deleted 90 days after last use, whichever is earlier.
  • Account record. Kept until you delete your account, after which it is removed within 30 days subject to any legal obligation to retain it longer (for example, tax records of payments).
09

Security

We apply the security controls a small founder-operated SaaS can reasonably maintain:

  • TLS 1.2+ for all network traffic between the macOS app, the backend, and third-party APIs.
  • OAuth tokens encrypted at rest using authenticated symmetric encryption; the key is held outside the database.
  • Short-lived JWTs for session authentication; secrets live in the macOS Keychain on-device.
  • Token-bucket rate limiting per account to contain abuse and protect against runaway loops.
  • Encrypted-at-rest database backups managed by Neon.
  • Principle of least privilege on infrastructure access; production credentials are not shared.

No system is perfectly secure. If we discover a breach affecting your personal data, we will notify you without undue delay and in any case within 72 hours of becoming aware, in line with GDPR Art. 33 / 34 and the DPDP Act's breach-notification requirements.

10

Your rights

You have rights over your personal data. The catalogue varies by jurisdiction; the table below maps the most important ones. To exercise any right, write to privacy@voxa.app from the email address on your account, or use the in-app controls where available.

DPDP Act, 2023 (India)

  • Right to access a summary of your personal data and how it is being processed.
  • Right to correction of inaccurate or incomplete data.
  • Right to erasure of personal data that is no longer necessary for the purpose it was collected.
  • Right to nominate another individual to exercise your rights in the event of death or incapacity.
  • Right to grievance redressal. Our Grievance Officer is Adwait Keshari, contactable at privacy@voxa.app. We will acknowledge a grievance within 7 days and respond substantively within 30 days. If you are not satisfied with our response, you may approach the Data Protection Board of India.

GDPR / UK GDPR (EU, EEA, UK)

  • Right of access (Art. 15).
  • Right to rectification (Art. 16).
  • Right to erasure / “right to be forgotten” (Art. 17).
  • Right to restriction of processing (Art. 18).
  • Right to data portability, in a structured machine-readable format (Art. 20).
  • Right to object to processing based on legitimate interests (Art. 21).
  • Right to lodge a complaint with your national supervisory authority.

Honest disclosure: Voxa does not currently have an EU representative under GDPR Art. 27 or a UK representative under UK GDPR. If our user base in those jurisdictions grows past the threshold where representation is required, we will appoint one and update this page. EU and UK users may exercise their rights against us directly today.

CCPA / CPRA (California)

  • Right to know what personal information we have collected about you and with whom it has been shared.
  • Right to delete the personal information we have collected about you.
  • Right to correct inaccurate personal information.
  • Right to opt out of sale or sharing of personal information. Voxa does not sell or share personal information for cross-context behavioural advertising, so there is nothing to opt out of, but you have the right to ask anyway.
  • Right to non-discrimination for exercising any of the rights above.
  • Right to limit use of sensitive personal information.

How we verify a request

We verify a request by replying to the email address on your account and asking you to confirm. For deletion requests we additionally require you to be signed in to the macOS app at the time the request is processed, so that someone with transient access to your inbox cannot wipe your data without your knowledge.

11

Children & minors

The minimum age to use Voxa is 13 years. We do not knowingly collect personal data from anyone under 13. If you believe a child under 13 has created an account, write to us and we will delete it.

Under the DPDP Act, 2023, any data subject under 18 years of age in India is treated as a child and processing requires verifiable parental consent. Voxa does not currently target users under 18 in India and does not have age-verification tooling in place; until we do, we ask Indian users under 18 not to create an account.

We also do not engage in tracking, behavioural advertising, or targeted advertising directed at children, regardless of jurisdiction.

12

Cookies & trackers

This marketing website uses only essential session cookies required to remember your theme preference and, if you sign in, to keep you signed in. There are no advertising cookies, no third-party analytics that profile individuals, and no cross-site trackers.

The macOS app does not set cookies. Authentication uses tokens stored in the macOS Keychain, scoped to the Voxa application.

13

Changes to this policy

We will update this policy from time to time. The effective date at the top of the page reflects the most recent version. For material changes — for example, adding a new category of data, a new subprocessor that handles your conversation content, or shortening user rights — we will notify signed-in users by email and via an in-app notification at least 30 days before the change takes effect.

Non-material changes (typographical fixes, clearer phrasing, restructuring) take effect immediately and are noted in the change history we keep alongside the source of this page.

14

Governing law

This Privacy Policy is governed by the laws of India, without regard to its conflict-of-law principles. Any dispute arising out of or related to this Privacy Policy is subject to the exclusive jurisdiction of the competent courts located in Bhopal, Madhya Pradesh, India, subject to the dispute-resolution clause in our Terms of Service for matters covered there.

15

Contact & grievance officer

For privacy questions, data subject requests, or grievances under the DPDP Act, 2023:

  • Email: privacy@voxa.app
  • Grievance Officer (DPDP, India): Adwait Keshari, Bhopal, Madhya Pradesh, India.
  • Postal address: Bhopal, Madhya Pradesh, India. Specific street address available on request to verified data subjects, to keep this page from becoming a doxxing target.

We will acknowledge every privacy-related communication within 7 calendar days and respond substantively within the timeline applicable to your jurisdiction (30 days under DPDP, one month under GDPR, 45 days under CCPA).