Privacy

Who this covers

Draki is operated by Darijan Ducic, an independent developer in San Francisco, California, United States. This policy covers the public Draki iOS app, the cloud services that support it and optional MCP connections you configure with compatible AI clients. Our initial service is for adults in the United States. The website section below separately explains information collected when you visit our site or join the waitlist.

Information Draki handles

• Account details. Email, name, language preference and sign-in information. Google sign-in shares basic identity, not your Gmail, Calendar, Drive or password.
• Your content. Recordings, transcripts, summaries, imported chats, questions and answers, and saved people and LinkedIn links. With your permission, new recordings are uploaded to Draki and processed by OpenAI to create transcripts. Transcripts and related context are stored in your Draki account. Content can include information about other participants.
• People you identify. Names, participant labels and details you add to conversations. The public app does not enroll your voice, create voice profiles or automatically recognize speakers by their voices.
• Time and optional location. Recording times and locations, including precise coordinates when permitted. With Remember my places and Always access enabled, visits and significant location changes are collected in the background, outside recordings. This does not record audio.
• Service information. Device identifiers, IP addresses, connection records, usage and error information, and messages you send to support.

Recording and cloud transcription

New recordings in the current public iOS app use OpenAI for transcription. Before you start, Draki asks permission to send recordings through its server to OpenAI. This choice applies to subsequent recordings for that account on that installation. You can decline; recording in this mode will not start. Microphone permission alone is not permission for cloud processing.

Once authorized, recording can work offline. Audio waits on your iPhone for an eligible Wi-Fi connection, then uploads in parts to Draki's server. OpenAI processes parts containing speech to return text. Transcription needs an internet connection and does not run locally for new recordings. Pending recordings made with an earlier local version keep their original on-device processing path; their text still syncs to your account.

Transcript text, recording-part timings and related context are stored in your Draki account and made available in the app. Context includes recording times, a device identifier and location when permitted. This mode does not generate word-level timestamps or automatically identify speakers. Our servers store and process text to organize your conversations, create summaries and support search and answers. Relevant text and context are processed by OpenAI for these features. This cloud processing is part of Draki's features and does not depend on connecting ChatGPT, Claude or another assistant.

Connecting your own assistant is optional. Draki currently provides a manually configured MCP (Model Context Protocol) connection for compatible clients. It shares authorized text and context from your Draki account, never raw recordings or voice profiles. You can use the app without connecting an assistant.

Account, sign-in, support and website services separately use the information described in this policy to operate and protect the service, resolve problems and respond to your requests. Information you provide for troubleshooting may be used to evaluate and improve those features.

We do not sell your personal data or use your conversation content for advertising.

Who receives information

Our service providers include Google Cloud for hosting and storage, Supabase for account services, Resend for sign-in emails and OpenAI for audio transcription, text analysis, summaries and search. OpenAI receives recorded audio for transcription and relevant transcript or imported-chat text, questions and context for text features. This can include personal information about you and other participants. This processing is separate from any assistant you choose to connect.

OpenAI states that API data is not used to train its models by default unless the API customer opts in. Provider retention differs by service: its audio transcription endpoint lists no content retention, while text services may retain abuse-monitoring content for up to 30 days, with exceptions described in its policy. These are separate from Draki's storage and backups; they are not a promise that every copy is deleted immediately. See OpenAI's API data controls.

Authorized operators may access cloud-stored account information, retained audio, conversation text and service records for support, maintenance and security. They cannot remotely access content stored only on your phone.

We may also disclose information to meet legal obligations, protect people or the service, or as part of a business transfer subject to appropriate safeguards and notice where required.

When you connect your AI

You set up MCP manually in a compatible client's connection settings, then sign in to Draki and authorize access. Draki does not currently offer an officially listed plugin or connector for ChatGPT or Claude. Compatibility and setup depend on the client, your plan and any workspace restrictions; a provider name here does not imply approval or partnership.

When your client makes an MCP request, Draki checks its authorization, retrieves content from your synced account and returns the requested results. These may include transcript passages or full conversation text, summaries, people and saved LinkedIn links, imported chats, timestamps, and available conversation locations or place history. Requests are processed by Draki's cloud service and may use the text-processing services described above for search.

Authorization grants ongoing read access within your account until revoked or expired; it is not limited to one conversation or one question. Your client controls when it makes requests and whether it asks you to approve each one. MCP access is currently read-only: it cannot edit or delete your Draki data, retrieve another account's content or export raw recordings, voice samples, voice profiles or content kept only on your phone.

The client you connect receives the returned text and context. If it uses a cloud AI provider, that provider may also receive and process those results under its own policies and your settings. This includes manually configured connections to ChatGPT or Claude where supported. These optional connections are separate from the OpenAI processing used for Draki's own transcription and text features.

Disconnect in Profile → Connect with your AI to stop future access. Information already received by an assistant is governed by that provider's policies and your settings, including its retention and model-improvement settings. Disconnecting Draki does not erase those copies. See OpenAI's privacy policy, Anthropic's privacy policy, or your chosen provider's policy.

Recordings and earlier test versions

The public app does not offer voice enrollment or automatic speaker recognition. Earlier or private test versions may have uploaded recordings, voice samples or derived voice profiles for cloud processing. Switching to the public app does not delete information already stored from those versions. Contact us to request access to or deletion of that information. Raw audio and voice profiles are not exposed through MCP.

Recording permission and permission to recognize a person's voice are different. Obtain any required permissions before recording, importing or identifying someone. Contact us about a voice profile or recording involving you, even if you do not have a Draki account.

Where information stays

Audio is saved on your iPhone while awaiting upload and processing, and uploaded audio is also held on Draki's server. Transcripts and summaries are stored in your account, with downloaded content cached in the app. Our primary cloud storage is in the United States. Providers, support and recovery copies of cloud information may involve other countries.

• Recordings. For new cloud recordings, the app removes each audio part from the phone after the server reports processing complete. Server cleanup targets completed originals after 24 hours; cleanup is periodic, not an exact deletion deadline. Pending or failed audio may remain on the phone and server for recovery until resolved or deleted. Earlier local recordings remove audio after saving their local transcript. Removing audio does not delete its transcript.
• Account and synced content. Account details, synced transcripts, summaries, imported chats and related context remain while needed for the service, until removed or a deletion request is completed. A connected assistant may retain its own copies under its policies.
• Earlier test data. Previously uploaded recordings and voice information may remain in cloud storage or recovery copies under the retention settings of those features. Contact us to identify and remove retained data from your account.
• Logs and recovery. Operational records are retained as needed for support, security and legal obligations. Limited identifiers and transcription usage records remain after a conversation is deleted to prevent its recreation and enforce usage limits; these do not retain its transcript. Daily backups can include uploaded audio and text. Cloud backup expiry is seven days, with a further seven-day soft-delete recovery period. Failed local backup copies may need separate removal.

Cloud storage is encrypted at rest, but this is not end-to-end encryption: authorized server processing can read uploaded content. Deletion requests cover relevant live records and retained copies, subject to legal exceptions and backup expiry. We explain any applicable limits when handling a request and can provide instructions for managing information stored on your phone. Signing out or removing the app does not delete your cloud account or copies already shared with an AI.

You're in control

You can cancel the cloud-processing prompt before recording. To stop a capture, use Stop recording; this does not withdraw parts already sent or cancel queued uploads. To remove a stopped conversation, swipe it in Transcripts and choose Delete. Local deletion is recorded immediately; server deletion waits for a connection. Saved contacts remain. Contact us to withdraw cloud-processing permission or request account deletion. There is currently no in-app switch to revoke that permission. Deletion cannot recall content already received by OpenAI or a connected assistant.

Manage microphone and location permissions in iOS Settings. In Profile → Permissions, turn off Remember my places to stop new collection or clear its saved history from your phone. Turning it off keeps existing history; signing out also stops collection. Locations attached to conversations may sync with their text. Clearing local history does not necessarily remove synced cloud copies; contact us for help deleting those. If an assistant has received location context, also use its data controls for those copies.

Email darijanduci@gmail.com to request access, corrections, a copy, deletion or help with voice information. Full account deletion is currently handled by request. We may verify your identity and clarify the request without exposing someone else's information. Depending on applicable law, you may have additional rights, including an appeal or complaint to a regulator. We do not penalize you for exercising applicable privacy rights. Instructions are on our support page.

When you visit the website

We receive information you submit to forms, including the waitlist, and technical request information. We use cookies or similar storage for website functions and, where enabled, Google Analytics for site usage. Those providers may receive browser, device, IP and usage information. You can control cookies in your browser. We do not currently change website behavior in response to browser Do Not Track signals. Service messages and requested updates are separate from your recorded conversations.

Age and updates

Draki is for people 18 and older and is not directed to children. Contact us if you believe a child has provided personal information so we can investigate and address it. We update this page as the service changes and notify you of material changes in the app or by email where required. The date above identifies this version.