Privacy Policy
Draft for the public release. Not yet in force. Clauses marked as public-release requirements depend on features and safeguards being completed before launch.
Who is responsible
Orage is published by Axel Rock, acting as an individual until a company is formed. The data controller is [Publisher legal name and address]. This field must identify Axel Rock and his legal business address before publication. Orage is not OpenAI, Anthropic or Apple.
This policy covers orage.ai, its waitlist, app.orage.ai, and the Orage apps for iPhone, iPad and Mac. Contact privacy@orage.ai for privacy matters and support@orage.ai for help.
For personal data in repositories or connected business services that you control, we process that content on your instructions to run Orage. You remain responsible for having authority to use it. Business use involving other people's data requires a data processing agreement with us. This policy does not replace that agreement.
What we collect and why
We receive data from you, your device, services you connect, and agents working for you. You choose your projects and connections. The data they contain can include other people's information. Do not supply sensitive information unless the feature and the provider are suitable for it and you have a lawful basis.
In this table, contract means we need the data to provide the service you request. Consent means an optional use you can turn off. Legitimate interests means our need to secure, maintain and improve Orage, balanced against your rights. Legal duty means a duty under applicable law. Providers are recipients for the stated purpose; their exact role depends on the service and contract.
| Data category and source | Purpose | Legal basis | Providers that receive it |
|---|---|---|---|
| GitHub account ID, login, avatar, connection scopes; model account email, identity and plan | Sign in, show your account, connect models and repositories | Contract | Railway, GitHub, OpenAI; Anthropic if connected; Boat for cloud account setup |
| Session identifiers, OAuth access and refresh tokens, API keys and connection records | Keep sessions and connections working; authenticate agent requests | Contract; legitimate interests in secure access | Railway, the relevant connected provider, Boat where an agent needs credentials; Orage's Keys service when configured |
| Prompts, chats, code, repository history, file contents, uploads, screenshots, tool results and agent output | Run tasks, keep history and let you retrieve results | Contract; separate permission before sharing with third-party AI | Railway, Boat, Railway's Tigris file store, GitHub when used, OpenAI or Anthropic and any selected AI gateway or tool provider |
| Workspace and computer records, schedules, model choices, approval settings, saved agent memories and work summaries | Resume work, remember preferences, run schedules and organize tasks | Contract; consent for optional AI summaries | Railway, Boat, Tigris for saved files and memory copies; Vercel AI Gateway and its selected model provider for AI summaries |
| Data from optional connected services, such as chat messages, names, emails, calendar items, files, issues and product errors | Answer requests and show connected work; rank or summarize it if enabled | Contract for requested access; consent for optional background AI processing | Railway, relevant connected services, Keys when configured; Vercel AI Gateway, TypeSafe AI, OpenAI or Anthropic when AI processing is enabled |
| Music service connection, library and playlist items, likes, track IDs, titles, artists, queue, playback times and listening history | Play your music, keep your queue and show listening stats | Contract; consent for library permission and optional AI discovery | Railway, your Jellyfin server; Apple for permitted music features and artwork; Vercel AI Gateway and OpenAI for AI discovery if enabled |
| Permitted music audio, offline downloads, song analysis and separated audio parts | Playback, offline access and transitions where the source permits them | Contract; consent for optional remote analysis | Railway, Tigris; Modal if remote analysis is enabled; your Jellyfin server |
| Host scripts, task summaries, episode scripts, generated speech, captions and timing data | Speak agent updates and play spoken episodes | Consent for optional AI writing and speech | Railway, Tigris, Vercel AI Gateway, Anthropic or TypeSafe AI as needed; ElevenLabs for speech and alignment |
| Microphone sound, speech transcripts and voice conversation context | Local dictation, live transcription and voice commands | Consent for microphone access and optional remote voice processing; contract for a task you send | Local dictation audio stays on your device. Remote voice audio goes to ElevenLabs. Transcripts and context go to Railway and, where needed, Vercel AI Gateway, OpenAI or Anthropic |
| Push tokens, device names, notification settings, thread identifiers and notification text | Deliver alerts and Live Activities | Contract for alerts you request; consent for system notification permission | Railway and Apple |
| Device presence, screen state, idle time, audio output type and output name | Keep devices and playback in sync and provide context for your agent | Contract; consent for optional use in AI context | Railway; Vercel AI Gateway and relevant model provider if AI context is enabled |
| Usage and error events, account or installation ID, thread and tool IDs, device class, app and OS version, timing, CPU, memory, token counts and usage limits | Diagnose failures, measure performance and manage capacity | Legitimate interests in reliable, efficient service; consent where local law requires it for optional measurement | Railway and PostHog EU; Apple supplies native performance reports |
| IP address, request metadata, user agent, page path and referrer | Deliver pages, prevent abuse, secure requests and count visits | Legitimate interests in security and site measurement; consent where required for measurement | Railway, Vercel and PostHog EU as applicable; file hosts and download providers receive network metadata |
| Waitlist email, submission time and originating page | Keep your place and send the access updates you requested | Consent | Railway; [Waitlist email delivery provider] when messages are sent. PostHog receives the page and repeat-signup flag, not the email |
| Support messages, attachments and privacy request verification | Answer you, investigate problems and honor your rights | Contract for support; legal duty for rights requests; legitimate interests in handling disputes | [Support mailbox provider]; hosting providers if records are stored there |
| Future purchase records, subscription status, invoices and refund records | Deliver a paid plan and meet accounting duties | Contract and legal duty | Apple for App Store purchases; [Web payment provider] if web payments are introduced |
Telemetry is linked to a GitHub numeric ID or a persistent installation ID. It is not anonymous account telemetry. The event code avoids sending prompts, assistant text, raw commands, tool arguments, file contents, passwords or tokens to PostHog. Native reports include crash and hang categories, not raw call stacks. We do not record sessions or capture everything you click.
The public site sends one page-view event per page load, when measurement is configured. It uses a new random ID each time and stores no analytics cookie. It sends the page origin and path, without that page's query or fragment, and the referrer supplied by your browser. That referrer can contain a full URL. PostHog still receives the network request. See our Cookies and browser storage page.
AI processing and your choice
AI features send personal data to third parties. Connecting your own ChatGPT or Claude plan does not make all AI processing local, and it does not cover every background AI feature.
| AI recipient | What it receives | Feature |
|---|---|---|
| OpenAI | Prompts, selected files, code, images, relevant chat history and tool results | Codex tasks using your connected ChatGPT account, or API access you choose |
| Anthropic | Prompts, code, relevant history and tool results | Optional Claude coding through your connected account or the selected API route |
| Vercel AI Gateway, then OpenAI | Work context and summaries; voice transcripts and routing context; music likes and most-played artists | Work cards, voice routing and optional music discovery |
| Vercel AI Gateway, then Anthropic | Work summaries, host context, voice transcripts, recent conversation and agent answers | Host writing and spoken answers, even if you have not connected a Claude plan |
| Vercel AI Gateway, then TypeSafe AI | Connected message text, thread excerpts, task and device context, or song metadata | Inbox ranking, work priorities and host decisions |
| OpenRouter, then the selected Anthropic service | Coding prompts and context | Optional alternate Claude API route |
| ElevenLabs | Microphone audio and vocabulary hints for live transcription; text for synthetic speech; audio and matching text for alignment | Remote voice, spoken answers, host clips and episodes |
| Modal | Permitted song audio, which may contain personal recordings | Optional remote song analysis |
Public-release requirement: before the first transfer for an AI feature, Orage will name each recipient, explain which data it receives and why, and ask for your explicit permission. Permission for a model account, the microphone or a connected service is not a substitute for this disclosure. Background AI features must stay off until you allow them. New recipients or purposes need a new choice. You can withdraw permission in Settings or through privacy@orage.ai. Withdrawal stops future transfers for that feature. It does not undo lawful processing already done. Declining an optional feature leaves unrelated features available. A cloud AI task cannot run without sending its context to the chosen provider.
We do not use your code, chats, music library, listening history or voice recordings to train our own general AI models or to advertise to you. A provider's own retention and model-improvement rules depend on its product, your plan, settings and our contract. A personal subscription is not the same as an API agreement. We do not promise that every provider keeps no data or never trains on it. Review the provider notices linked on our provider page and your account controls.
AI may infer priorities and interests to help you organize work or discover music. These suggestions do not decide your legal rights or make decisions with a similar significant effect without human involvement.
Music
Orage is a player for content you already have permission to access. It does not supply a music catalog or sell music.
The current private build connects Jellyfin and keeps playback history. The unofficial YouTube route is for the owner's private use and must not ship in the public product.
Public-release requirement for Apple Music: MusicKit must ask for access first. If you allow it, Orage may read the authorized library, playlists, song metadata and available listening history to show and play your music. Access must stop when you revoke it. Apple Music library and listening data will never be sold, used for ads or sent to AI providers through the general music-discovery route. MusicKit audio must stay within Apple's permitted playback system. Orage will not upload it to a song analyzer, copy it to its bucket or turn it into downloadable files. Apple Music is not integrated in the current build.
For Jellyfin, Orage reads the connected library, searches and favorites. It streams audio and artwork through our server so the connection token is not sent to the browser. Choosing a favorite can write that choice back to Jellyfin. Song title and artist queries also go to Apple's artwork search service. Optional discovery sends recent likes and top artists to AI providers. Optional song analysis sends permitted audio to Modal. These uses need the choices described above.
Microphone and voice
Local dictation and remote voice are different features. Local dictation uses a downloaded model on your Apple device. It holds up to 60 seconds of microphone audio in memory, clears it after capture and does not upload or save it. You review the resulting draft and press Send to share it with an agent. Model downloads contact Hugging Face, which receives download and network metadata, not your dictation audio.
Remote voice streams sound from your microphone to ElevenLabs. While voice mode is listening, this can include nearby speech or background sound. Transcripts reach Orage and may reach the AI services above. Stop voice mode to stop sending sound. Device microphone permission can be revoked in system settings. Spoken commands can start tasks without a separate Send press.
We use a supplied synthetic host voice. We do not use the microphone to identify you or create a voiceprint. Host clips and episodes are saved for replay. Ordinary streamed spoken answers are not saved as audio by the speech-stream route. A transcript or spoken answer can still become part of a chat, music session or task record.
How long data stays
The following gives current storage behavior and the limits that must be set before public launch. Removing an app, signing out or disconnecting a provider does not delete the whole account.
| Data | Retention and deletion |
|---|---|
| Account, connections and preferences | While the account is active, then removed through account deletion. Disconnect removes the active connection credentials. Some disconnected Claude identity records remain to match a later reconnection until account deletion |
| Code, workspaces and live agent history | While the workspace or computer is kept. Archived task computers are scheduled for deletion after 24 hours, subject to use and successful history copying. Saved chats and captured files can outlive the computer |
| Saved chats, uploads, captured files, agent memories and memory copies | Kept until you delete them or the account. The private build has no general automatic expiry. Deletion must also cover saved histories, nightly memory copies and provider snapshots |
| Music queue, history, analysis, permitted cached audio, host clips and episodes | Kept for replay and stats until deleted or the account closes. The private build has no time-based purge for long-term music history and files |
| Local dictation audio | Transient device memory, at most 60 seconds per recording; not persisted |
| Local models and offline music | Until removed through device controls, replaced by the offline-library limit, or the app's local data is deleted. Turning dictation off leaves its downloaded model |
| Remote voice context | The voice service keeps at most 12 recent exchanges within a 15-minute window. Pending task answers have a 45-minute relevance window. Text awaiting speech has a two-minute cleanup threshold on later writes. These are memory limits, not guaranteed provider erasure times; saved chats and session records follow their own rules |
| Device presence | Live reports expire or are removed on disconnect. The private build also appends context changes to a log with no automatic expiry. Public log limit: [Presence log retention period] |
| Push records | Until a device is unregistered or the account is deleted; rejected tokens are cleared |
| Local diagnostic journal | Seven-day cleanup window, applied when new events are written; at most 5 MiB per day's file |
| PostHog events and provider operational logs | [PostHog event retention period] and [Infrastructure log retention period], respectively |
| Waitlist | Until you withdraw or [Waitlist retention period], whichever comes first |
| Support and privacy requests | [Support and privacy request retention period], with longer restricted storage only when needed for a legal duty or a specific dispute |
| Future accounting records | Required invoices and accounting evidence: ten years from the end of the relevant financial year under French rules; other purchase records only as needed for the contract or applicable law |
| Backups and snapshots | Deleted data may remain in restricted backups for [Maximum backup and snapshot retention period]. It must not return to active use except for recovery with deletion reapplied |
AI providers and services you connect can keep their own records under their agreements and policies. Deleting Orage cannot delete a GitHub repository, Apple purchase, or provider account. We pass required deletion instructions to our processors. We explain any lawful exception and its period when answering your request.
Who receives data and where
Current infrastructure includes Railway, Vercel, Tigris and Boat. AI and voice recipients include OpenAI, Anthropic, Vercel AI Gateway, TypeSafe AI, ElevenLabs, and optional OpenRouter and Modal. Measurement uses PostHog's EU service. GitHub and Apple receive data for the features described here. Optional connected sources include Slack, Google, Linear, Sentry and Conductor. Hugging Face serves local model downloads. Your Jellyfin host receives its own library requests. Support, email and future payment providers must be identified before they are used.
The Subprocessors and connected providers page lists their country, function, data and privacy notice. Sentry is currently a source of errors from products you connect, not Orage's telemetry collector. We found no Neon database or direct Cloudflare R2 file store in the current Orage setup. Railway's file bucket uses Tigris. Some providers use further suppliers under their own contracts.
Railway's documented production server region is Amsterdam, Netherlands. PostHog is configured for EU ingestion. This does not mean all data stays in Europe. AI processing, support, control systems, downloads and global delivery can involve the United States and other countries. The provider page flags locations that still need verification.
Public-release transfer requirement: before transferring data outside the EEA or UK, we must have a valid transfer mechanism for the actual recipient and service. This can be an applicable adequacy decision, including a valid US Data Privacy Framework certification within its scope, or EU Standard Contractual Clauses with a transfer assessment and added safeguards where needed. UK transfers need the applicable UK mechanism, such as the UK Addendum or IDTA. A provider's US address or an EU server is not proof of safeguards. We must record the mechanism for each route before launch. Ask privacy@orage.ai for a copy or explanation, with confidential terms redacted where needed.
We limit access to people and suppliers who need it to run or support Orage. We may disclose data when law requires it or to handle a specific security incident or legal claim. If the publisher changes, any transfer must remain lawful, and we will notify you of a change affecting your data.
Security
We use authenticated access, HTTPS for public connections and restricted files. Many connection records are sealed with encryption. Not every token or file is encrypted by Orage itself: Codex login files and the current GitHub connection are stored on computer disks. Agents can have full access to their own computer, including files and credentials there. Protect secrets in repositories and choose approval settings with care. No system is perfectly secure.
Your rights and how to use them
Under EU and UK data protection law, you can ask for access, correction, deletion, restriction, and a portable copy where that right applies. You can object to processing based on legitimate interests, including measurement. You can withdraw consent at any time. Some rights have legal limits. We will explain any limit and how to challenge our answer.
Email privacy@orage.ai. Tell us your account identifier and what you want. We may ask for a proportionate way to verify ownership. Do not send a password, provider token or full identity document unless a secure, necessary verification step is agreed. Requests are normally free. We respond within one month under EU and UK rules. If a lawful extension is needed, we tell you within that first month, give the reason and state the new deadline.
Public-release account deletion: open Settings, Account, Delete account in the app and confirm. You can also request deletion at privacy@orage.ai or support@orage.ai. This must delete the account, stop schedules and running tasks, revoke or remove connections, unregister devices, and remove personal content from computers, file storage and processors, subject to the limited legal and backup exceptions above. This control is not present in the current private build. Until it is ready, use email for a manual request. Export wanted files first. Deletion does not itself cancel an Apple subscription; cancel it in Apple's subscription settings too.
You may complain to the French authority, the CNIL, or your local supervisory authority. UK users may contact the Information Commissioner's Office. You do not have to contact us first.
California notice
Where the CCPA and CPRA apply, this section also serves as our notice at collection. The categories in the table above cover identifiers; account and commercial records; internet and app activity; audio and other user content; work-related information in connected content; and inferences such as music suggestions and work priorities. Sensitive personal information may include account credentials, private communications and sensitive details you choose to supply. We do not request precise location or use voice for biometric identification.
Sources, purposes, recipients and retention are stated above. The preceding-12-month inventory must be confirmed before publication: [Verified categories collected and disclosed in the preceding 12 months]. Planned MusicKit, public subscriptions and public account features are not claims of past collection. The technical inventory above describes what each reviewed feature handles, rather than proving that every category was collected from every person.
We do not sell personal information or share it for cross-context behavioral advertising. We do not track you across other companies' apps or websites, use advertising identifiers, or sell or advertise against music or voice data. We do not knowingly sell or share minors' information. Global Privacy Control requires no sale or sharing to be switched off because we do neither; it must not be repurposed for tracking.
You may request the categories and specific pieces we hold, their sources, purposes and recipients; correction; deletion; and a portable copy, subject to applicable limits. You may opt out of sale or sharing and limit uses of sensitive information where those rights apply. We use sensitive information only as needed to supply requested services, secure them or meet law, not to infer sensitive traits for unrelated purposes. There are no data-for-discount financial incentives.
Send requests to privacy@orage.ai or use the in-app deletion route once available. An authorized agent can act for you with proof of authority; we may also verify your identity as allowed by law. We acknowledge qualifying California requests within ten business days and normally answer within 45 days. We will explain any permitted extension or denial and how to seek review. Exercising rights will not lead to unfair treatment. A feature may stop if its necessary data is deleted or its permission is withdrawn.
Age and changes
Orage is not for anyone under 16. If local law or a connected provider sets a higher minimum age, that higher age applies. People aged 16 or 17 need permission from a parent or guardian where required to enter the contract. If we learn that we hold a child's data contrary to these rules, we will remove it. Contact privacy@orage.ai to report this.
We update the date on this page when the policy changes. For a material change, we give notice in the app or by email before it takes effect where required. New consent-based uses need a new choice; a policy edit alone is not consent.
Last updated 2026-10-09