Privacy Policy
This Privacy Policy explains what personal data Coburg collects when you use coburg.ai — on the web, in the Coburg app for iPhone or in the app for Mac, which both show coburg.ai — why, how long we keep it, and what rights you have. We keep it as short as we can. The main idea has not changed: your chats and files live in your own storage, not on our servers, and our servers pass your messages to the AI provider you choose without storing them.
Coburg is currently in an invitation-only early-access phase. Only people whose email address we have added to our list of invited users can sign in. This policy describes the service as it works today and will be updated before new features that use other data are introduced.
Who we are
The controller of your personal data is:
Hardcarbon Iberia S.L.Calle Padre Joaquín Belón, núm. 1, esc. 1, piso 1, puerta 1
29602 Marbella (Málaga), Spain
NIF B22571335 · EU VAT ES B22571335
For any question about this policy or your personal data, write to support@coburg.ai. To report a security issue, write to security@coburg.ai.
What we collect
When you sign in
You sign in with your Google account or your Microsoft account. You interact with Google or Microsoft directly and confirm there that Coburg may receive your basic profile. We then receive:
- your account identifier at Google or Microsoft (a value that does not change),
- your email address and, where the provider tells us, whether it has been verified,
- your name as shown in that account.
We never see your password. Signing in does not give Coburg access to your email, contacts, calendar or any other data in these accounts. (If you later connect Gmail, Outlook or another service in Connections, that is a separate, explicit step described below.)
From your sign-in we create a small account record: the provider (Google or Microsoft), the identifier, email address, verification status, name, and the times the record was created and you last signed in.
The Coburg app for iPhone signs you in through a system window of your phone. During that hand-over, our server keeps only random one-time codes for two minutes; nothing else about you is added.
Your chats and files in your own storage
Coburg keeps your chats and files in a storage account of your own, not on Coburg’s servers. You choose the storage, and you can switch at any time; the account you sign in with and the storage you use are independent of each other. Everything is kept in a folder called Coburg AI with the folders Threads, Documents, Pictures and Projects inside. Each chat is one file in Threads that you can open, read and delete directly in your storage. Files you attach to a message go to Documents, photos and pictures you attach or that an AI model draws for you go to Pictures. A project is a folder in Projects with a small description file.
| Storage | Where Coburg writes | What Coburg can access |
|---|---|---|
| Google Drive | Folder Coburg AI in your Drive | Only the folders and files Coburg itself created (Google permission drive.file), never your other files |
| Microsoft OneDrive | App folder Apps/Coburg AI | Only its own app folder (Microsoft permission Files.ReadWrite.AppFolder) |
| Dropbox | App folder Apps/Coburg AI | Only its own app folder (Dropbox “App folder” access) |
| GitHub | Your private repository Coburg-AI, which you create yourself | Only that one repository, and only the contents of its files (GitHub permission “Contents: read and write”). Coburg cannot create, manage or delete repositories. |
| GitLab | Your private project Coburg-AI, which you create yourself | GitLab offers no narrower permission than its general API access (api), which technically covers all your projects. Coburg works only with the private project Coburg-AI in your own namespace and does not read or change any other project. |
- To keep working with your storage while you use Coburg, we store a long-term access key from the storage provider (a “refresh token”) on our server, encrypted with a key that is kept separately from our database and its backups. Short-lived access keys are kept only in memory. We also store which storage you use and the identifiers of the Coburg AI folder, repository or project, so that Coburg can find them again.
- Coburg uses only the storage account you connected. If the Coburg AI folder or a chat file is shared with anyone else, or a GitHub repository or GitLab project is not private or not yours, Coburg stops writing to it and tells you.
- You can remove Coburg’s access at any time in the settings of your storage account (for example, Google Account → Security → Third-party apps). Your files stay where they are.
- If you rename or move the Coburg AI folders, Coburg puts the names back; if you delete one of them, Coburg creates it again, empty. Chats you delete are not restored.
- Switching storage: when you switch to another storage, Coburg offers to copy your chats and files there. It copies only if you agree, and only what is missing. So that the same chat is not copied twice and chats you deleted do not come back, we keep one-way fingerprints (hashes) of the chats and files involved, never their names or contents, together with the counts of copied and skipped items.
- Deleting chats and files: when you delete a chat or a file in Coburg, we keep a one-way fingerprint of its identifier, so that it is not copied back from another storage later. While a deletion or a save is still being confirmed with your storage, we keep a fingerprint of the chat (and of the saved content) until it is confirmed.
- Projects: the list of your projects is read from your storage. On our server we keep only a keyed fingerprint of the name of the project you last opened.
- Coburg’s use and transfer to any other app of information received from Google APIs will adhere to the Google API Services User Data Policy, including the Limited Use requirements. For every storage provider, we use the access only to save, show and copy your Coburg chats and files; we do not use it for advertising, do not sell it, and do not use it to train AI models.
While you are signed in
When you sign in, your browser receives a random session identifier in a cookie. On our server we store only a one-way fingerprint (hash) of it, together with the time the session started, when it was last used and when it ends. A session ends when you log out, after 7 days without use, after 30 days at the latest, or when you sign in on another browser or device (Coburg allows one signed-in browser at a time).
While a page is open, your browser or app keeps the chats you open in its memory so they appear instantly, and it may keep small previews of your pictures and documents on your device so they do not have to be downloaded from your storage again. These copies stay on your device; nothing from your chats is kept in our servers’ storage.
Security and abuse prevention
- To limit repeated sign-in and connection attempts, we count attempts per network address. We do not store your IP address for this: only a keyed fingerprint from which the address cannot be read back, kept for at most two hours.
- Our servers keep short technical error logs. These may contain your IP address, the name of a service you connected or the name of a tool an AI model used. They never contain your messages, answers, attachments, the content of connected services, voice audio or sign-in codes and cookies, and they are deleted after at most 90 days.
Invitation list
During early access we keep a list of the email addresses of invited users, so that only they can sign in. If you are on the list, your email address is stored there even before you sign in for the first time.
When you use Chat: the AI provider you choose
Coburg does not build AI models. When you send a message, Coburg sends it, together with the earlier messages of the same conversation and any attachments you added, to the provider of the model you chose; you can see which model answered under each answer. Together with your message, the provider receives short instructions from Coburg (the model’s name, today’s date, formatting rules, the time zone your device reports so that “tomorrow” means the right day, and, if any, your personalization text — see below). Coburg’s servers only pass your conversation through: we do not store the content of your messages or of the answers on our servers, and they never appear in our logs. The conversation is saved only in your own storage (see above): your message before it goes to the AI provider, the answer when it is finished or when you press Stop. If you leave the page while an answer is being written, the answer is still finished and saved.
Which company receives your conversation depends on the model you pick. The providers, where they process your data and under what terms are listed under “Who processes it for us”. Models from providers outside the European Union, Canada and the United States — today those in Singapore, Hong Kong and China — are listed by country in the table below; please read the section on transfers before choosing them.
Attachments. Photos and PDFs you attach are sent to the AI provider as they are. Word, Excel and PowerPoint files, ZIP archives and PDFs for models that cannot read PDFs are opened by Coburg in a separate, isolated process on our server, and only their text is sent to the model. A scanned PDF with no text layer is first transcribed by a small model from Google (Gemini) at Coburg’s expense; the transcript is kept in our server’s memory for the length of the conversation, so the scan is not read twice, and is never written to disk. An attachment you remove before sending leaves nothing behind; one you send is saved in your storage.
Links. If your message contains web links, Coburg’s own server fetches those pages (identifying itself as “CoburgAI-LinkReader” and respecting the site’s robots rules) and sends only their text to the model. The website you linked sees a request from Coburg’s server, not from you. Coburg does not sign in anywhere or use your cookies for this.
Chat titles. To give a new chat a short title, Coburg sends the beginning of your first message (at most 2000 characters) to a small model — today Claude from Anthropic — at Coburg’s expense. The title is saved with the chat in your storage.
Images. When you ask a drawing model for a picture, the provider receives only your request and any pictures you attached, without Coburg’s instructions or your personalization text. The finished picture is saved in your storage, in Pictures. Some picture-generation providers embed a marking in the image that identifies it as AI-generated; Coburg does not add or remove such markings itself.
Personalization. You can write up to five short lines about yourself and how you want answers to look (Settings → Personalization). We store them with your account. They are placed at the beginning of your first message of each conversation when it is sent to the AI provider, so the provider receives them like any other message. They are not written into the chat file in your storage. Drawing models and research models that search the web do not receive them.
Your choices. We store with your account the model, version and effort you last chose (per mode: Chats, Images, Coding), your interface language, and the voice you picked, so they are the same on any device you sign in on.
Internet search
Many questions need current information from the web. How Coburg gets it depends on the model:
- Research models — ChatGPT Astra and Codex (OpenAI), Claude Fable, Opus and Claude Code (Anthropic), Gemini Pro (Google), Perplexity and Consensus — search the web through their own provider. Your conversation does not go anywhere else than to that provider.
- Other models use “Internet via Perplexity” (this feature is being introduced and may still be switched off for your account). Coburg sends to Perplexity (USA) only a short, standalone search query of a few words — for example “Madrid metro timetable October 2026” — together with today’s date, the time zone and language of your device. It does not send your account, your name, your conversation, your attachments or the content of connected services. Coburg refuses to send a query that looks like it contains an email address, a key, an IBAN, a long number or a password. Perplexity returns short findings with their sources, which Coburg adds to your question for the model you chose, and the sources are shown under the answer.
- For models that can use tools (OpenAI, Anthropic, xAI, Alibaba Qwen, Z.ai, BytePlus), the model itself decides when to search and writes the query.
- For the remaining models (for example Gemini Flash, Kimi, Meta, DeepSeek, Mistral, Cohere, StepFun), a small model from OpenAI (GPT-6 Luna) first reads the recent part of your conversation to decide whether a search is needed and to write the query. That part of the conversation goes to OpenAI under the same terms as any other message sent to OpenAI.
Voice
Coburg can take dictation, read answers aloud and hold a voice call with an AI model. In all three cases your microphone audio or the text to be read is streamed from your device through Coburg’s server to a speech provider and back. Audio and transcripts are not stored on our servers and are not written to our logs. Which provider is used depends on your language and on the voice you choose: dictation uses OpenAI, xAI, Google, Alibaba, Meta or Mistral; reading aloud uses Google, OpenAI, xAI, StepFun or Mistral; a voice call uses xAI (Grok), Google (Gemini Live) or OpenAI. During a call, the model can also use your connected services and, for Grok and Gemini, search the web through its own provider. After a call, its transcript is written into the chat in your own storage, like a typed conversation. For voice, as for chat, we store only numbers: the seconds of audio sent, the cost, the provider and when it happened.
Connections: your other apps and devices
In Connections you can let the AI read your data in other services you use, and in a few cases control a device. Today these are: Zoom, Slack, Notion, Asana, Atlassian (Jira, Confluence), GitHub, Canva, Linear, Todoist, Trello, Supabase, Vercel, Figma, Sentry, DocuSign, Miro, Monday.com, Box, Salesforce, Zendesk, Dropbox, GitLab, HubSpot, Zapier, Google Drive, Gmail, Google Calendar, Google Docs, Google Sheets, Google Slides, Google Classroom, OneDrive, Outlook, Outlook Calendar, SharePoint, Microsoft Teams, Spotify, Sonos and LG ThinQ. Nothing is connected unless you connect it yourself, and you can disconnect each service at any time in Connections or in the settings of that service.
- What we store: for each connected service, your identifier at that service and the account label it shows (usually your email address or user name), the permissions granted, and a long-term access key (“refresh token”) encrypted in the same way as your storage key. For Sonos we also remember the room you chose; for Zendesk the address of your company’s help desk; for LG ThinQ the country of your LG account and your personal LG key, encrypted. Short-lived access keys live only in memory.
- What the AI can do: read only — search and read your messages, files, pages, tickets, boards, mail, calendar, records and so on, and only while it is answering your question. The AI cannot send, write, change or delete anything in these services. The exceptions you explicitly allow are Spotify (play, pause, skip, volume, add a track to a playlist — never delete or rewrite), Sonos (play, pause, volume, favourites) and LG ThinQ (read the state of your appliances and send them commands). Zapier is limited to its read actions.
- What the AI receives: only what it asks the service for in order to answer the current question. Coburg gives the model the list of services you connected; the model decides which to use. If you name one service in your question, the model gets only that one. The content the AI reads passes through to the answer and, as part of the answer, into the chat in your storage. Coburg does not store it; our logs record only the service and the name of the tool used, never its content.
- Who talks to the service: for models from OpenAI, Anthropic and xAI, the AI provider’s own servers connect to the service (through the service’s official connector) using a short-lived access key that Coburg gives it for that one answer; the content of the service therefore goes to that AI provider, like your messages. For models from Alibaba, Z.ai and BytePlus, and for Spotify, Sonos, LG ThinQ, Google Classroom, GitLab, Figma, Microsoft services and a few others, Coburg’s own server reads the service and passes the result to the model — your access key never leaves Coburg. In all cases the data ends up at the AI provider of the model you chose.
- Managed tools that need no connection: Context7 (Upstash, USA) receives only the name of a programming library and the topic the model asks about, so it can return public documentation; the Shopify shopping assistant lets the model look up public products and policies of a Shopify store you name — the store sees a request from Coburg’s server; and “Internet via Perplexity” is described above.
- For Google services connected here, Coburg’s use of information received from Google APIs will adhere to the Google API Services User Data Policy, including the Limited Use requirements. We request read-only permissions wherever the service offers them.
Plan, credit and usage
To keep track of your credit, we store for every answer, search, image or voice session only numbers: your account, the provider and model, the number of tokens (pieces of text) or seconds of audio it used, its cost in USD and EUR, whether it finished, and when it started and how long it took. We also keep your plan, its currency (EUR or USD), its period and the credit added to it. When paid plans are introduced, payment card details and invoices will be handled by a payment provider under its own terms; this policy will be updated at that time.
When you write to us
If you email us, we receive your email address and whatever you write, and use it only to answer you.
Why we use it
| Purpose | Data | Legal basis (GDPR) |
|---|---|---|
| Letting you sign in and keeping you signed in, on the web and in the apps | Account record, session data, one-time app sign-in codes | Performance of our agreement with you (Art. 6(1)(b)) |
| Saving your chats, files, pictures and projects in your storage, and copying them when you switch storage | Encrypted storage access key, folder identifiers, fingerprints of copied and deleted items, fingerprint of your last project | Performance of our agreement with you (Art. 6(1)(b)) |
| Answering your messages through the AI provider of the model you chose, including attachments, links, titles, images and personalization | Your conversation, attachments, link text, personalization text, device time zone | Performance of our agreement with you (Art. 6(1)(b)) |
| Finding current information on the web for your question | A short search query, date, time zone and language; for some models the recent part of the conversation (query planning) | Performance of our agreement with you (Art. 6(1)(b)) |
| Dictation, reading aloud and voice calls | Microphone audio, text to be read, call transcript | Performance of our agreement with you (Art. 6(1)(b)) |
| Letting the AI read your connected services and control the devices you allow | Encrypted connection keys, service identifiers and labels, content the AI reads while answering | Performance of our agreement with you (Art. 6(1)(b)) — you connect each service yourself and can disconnect it at any time |
| Remembering your choices | Chosen model, version and effort per mode, language, voice, personalization text | Performance of our agreement with you (Art. 6(1)(b)) |
| Allowing only invited users during early access | Invitation list, email address | Our legitimate interest in running a controlled early-access phase (Art. 6(1)(f)) |
| Protecting the service against abuse and attacks, and fixing faults | Attempt counters, technical error logs | Our legitimate interest in keeping the service secure (Art. 6(1)(f)) |
| Keeping track of your plan and credit, and accounting | Usage records (numbers only), plan, credit | Performance of our agreement with you (Art. 6(1)(b)); keeping accounting records: legal obligation (Art. 6(1)(c)) |
| Answering your messages to us | Your email and message | Our legitimate interest in responding to you (Art. 6(1)(f)) |
We do not sell your personal data, we do not use it for advertising, we do not use your data to train AI models, and we do not make decisions about you based solely on automated processing. Please do not put other people’s sensitive data (health, political opinions, religion and the like) into Coburg unless you are allowed to; Coburg does not ask for such data and does not need it.
Cookies
Coburg uses only the two cookies it needs to work. There are no analytics, advertising or third-party cookies, and no tracking of any kind. The apps for iPhone and Mac show coburg.ai and use the same two cookies.
| Cookie | Purpose | Duration |
|---|---|---|
| __Host-coburg_login | Links your browser to a sign-in or a connection that is in progress, so it cannot be completed from another browser | 10 minutes, deleted when sign-in finishes |
| __Host-coburg_session | Keeps you signed in | Until you log out; at most 30 days |
These cookies are strictly necessary, so they do not require consent. They are sent only to coburg.ai, only over an encrypted connection, and cannot be read by scripts on the page. Your device also keeps your interface language and small previews of pictures in its own storage; this never reaches our servers.
Who processes it for us
Our infrastructure
- Hetzner Online GmbH (Germany) hosts our servers in Nuremberg and Falkenstein, Germany, and stores our encrypted database backups in Helsinki, Finland.
- Google and Microsoft provide the sign-in. When you sign in, you interact with them directly, and they process your data under their own privacy policies. Our business email is hosted by Google Workspace.
- The storage you choose — Google Drive (Google), OneDrive (Microsoft), Dropbox (Dropbox, Inc.), GitHub (GitHub, Inc.) or GitLab (GitLab Inc.) — keeps your chats and files under your own agreement with that provider and its privacy policy.
- The services you connect in Connections process your data under your own agreement with each of them; Coburg only reads them on your instruction.
AI providers
Each provider receives only what is described above for the model you chose — the conversation, attachments and instructions — and nothing else about you (no account record, no email address). We use each provider’s business API, not its consumer app. The following summarises each provider’s published API terms as we understand them at the date of this policy; the provider’s own terms prevail.
| Provider (models) | Where | Use of your data under its API terms |
|---|---|---|
| OpenAI (ChatGPT, Codex, GPT Image; search-query planning, dictation, reading aloud, voice calls) — OpenAI Ireland Ltd / OpenAI OpCo, LLC | USA | Not used to train its models; kept up to 30 days to detect abuse, then deleted. |
| Anthropic (Claude; chat titles) — Anthropic PBC / Anthropic Ireland | USA | Not used to train its models; deleted within 30 days. |
| xAI (Grok, Grok Imagine, Grok voice) | USA | Not used to train its models; deleted within 30 days. |
| Google (Gemini, Gemini Image, Gemini voice; scanned-PDF transcription; Consensus summaries) — Gemini API, paid tier | USA / Google data centres worldwide | Not used to improve Google’s products; logged for a limited time to detect abuse. Google may process it in any country where it has facilities. |
| Perplexity (Perplexity Sonar; “Internet via Perplexity”) | USA | Not used to train its models; may be kept temporarily under its API terms, which include its data processing addendum. |
| Consensus (scientific search) — receives your last message as a search query | USA | Not used to train AI models; search processed without storing the query (per Consensus). |
| Meta (Meta Muse, Muse Image; dictation) | USA | Not used to train its models (per Meta). |
| Mistral AI (Mistral; dictation, reading aloud) | France, EU | Not used to train its models; kept up to 30 days to detect abuse, then deleted. |
| Cohere (Command, North) | Canada | Kept up to 30 days; Cohere’s default settings allow training on API data unless the customer opts out in its dashboard. |
| Alibaba Cloud (Qwen, Qwen Coder, Qwen Image; dictation) — Model Studio, international region | Singapore | Not used to train its models; requests processed and stored in the Singapore region. |
| BytePlus (BytePlus Seed, Seedream) — ModelArk, Asia-Pacific region | Singapore / Malaysia | Processed in its Asia-Pacific data centres, which BytePlus states are separate from China; retention and training terms per BytePlus’ legal terms. |
| Z.ai (zAI GLM, ZCode) — Jingsheng Hengxing Technology Pte. Ltd. | Singapore | Not used to train its models; states that API content is processed in real time and not stored, except flagged content for up to 30 days. |
| Moonshot AI (Kimi, Kimi Code) — Moonshot AI Pte. Ltd. | Singapore | API data kept up to 30 days; Moonshot states it does not train on API data. |
| StepFun (Step; reading aloud) — StepFun Technology (HK) Limited | Hong Kong | Deleted after the service is provided, except where law or abuse prevention requires keeping it (per StepFun’s data processing agreement). |
| DeepSeek (DeepSeek) — Hangzhou DeepSeek Artificial Intelligence Co., Ltd. | China | Stored in China for as long as DeepSeek considers necessary; DeepSeek’s policy allows it to use data to improve and train its models. We recommend not sending personal or confidential data to this model. |
Transfers outside the European Union
Everything Coburg stores on its own servers is stored in the European Union. Your chats and files are kept wherever your chosen storage provider stores them. Data leaves the EU only when you choose a model, a search or a voice feature from a provider outside the EU, or connect a service hosted outside the EU; what is sent is only the content described above.
- United States (OpenAI, Anthropic, xAI, Google, Perplexity, Consensus, Meta, Context7, and the US services you connect): transfers rely on the EU–US Data Privacy Framework where the provider is certified under it, and otherwise on the European Commission’s Standard Contractual Clauses included in the provider’s data processing terms.
- Canada (Cohere): the European Commission has recognised Canada’s commercial data protection law as adequate.
- Singapore, Hong Kong and China (Alibaba Cloud, BytePlus, Z.ai, Moonshot AI, StepFun, DeepSeek): there is no adequacy decision for these countries. Transfers rely on the Standard Contractual Clauses in the provider’s data processing terms, where the provider offers them, together with the safeguards described in this policy: you choose these models deliberately, only the conversation is sent, your account data never is, and you can avoid these providers entirely by choosing a model from the EU, Canada or the USA. Local law in these countries may give authorities access to data that EU law would not allow. If you would rather not have any of your data processed there, do not choose these models.
You can ask us at support@coburg.ai for more information on the safeguards for a specific transfer.
How long we keep it
- Account record: as long as your account exists. You can delete your account yourself in Settings → Delete account; it is deleted right away (or, if you write to us, without undue delay).
- Sessions: until they end (see above); a record that a session was ended is kept for one more day, so that the browser concerned can be told why it was signed out. App sign-in codes: at most 2 minutes.
- Sign-ins and connections in progress: at most 10 minutes.
- Chat messages, answers, attachments, pictures, voice audio and transcripts, content of connected services: not stored on Coburg’s servers. Chats, attachments, pictures and call transcripts are in your own storage until you delete them. An attachment you did not send is dropped from our server’s memory after a short time.
- Encrypted storage and connection keys and identifiers: as long as your account exists and you have not removed Coburg’s access. If the provider tells us the access was removed, the key can no longer be used and the connection is dropped; all keys are deleted with your account.
- Fingerprints of copied and deleted chats and files, and of your last project: as long as your account exists. Fingerprints of saves and deletions still being confirmed: until they are confirmed.
- Your choices and personalization text: as long as your account exists, or until you change or clear them.
- Usage records, plan and credit: as long as your account exists, and afterwards as long as accounting law requires.
- Technical error logs: at most 90 days.
- Backups: our database is backed up every night. Backups are encrypted and deleted automatically about a month after they are made (after 30 days they are released, and deletion follows within two days). Data you have asked us to delete disappears from backups when they expire.
- At the AI providers: see the table above — typically up to 30 days, longer only where the provider says so.
Your rights
Under the GDPR you have the right to access your personal data, to have it corrected or deleted, to restrict or object to its processing, and to receive it in a portable format. Your chats and files are already in your own storage in an open format, so you can take them with you at any time. You can delete your account yourself at any time in Settings → Delete account. To exercise any of the other rights, write to support@coburg.ai. We may ask you to confirm that the request comes from you.
You also have the right to lodge a complaint with a data protection authority: in Spain, the Agencia Española de Protección de Datos (www.aepd.es), or the authority in the country where you live or work.
Age
Coburg is intended for people aged 18 and over. We do not knowingly collect personal data from anyone younger.
Security
All connections to coburg.ai are encrypted. Session identifiers are stored only as fingerprints, access keys for your storage and connected services are encrypted with a key kept outside the database, the database is reachable only from our own servers, documents are opened in an isolated process, and backups are encrypted before they leave the server. Provider keys stay on our servers; your browser never receives them.
Changes to this policy
We will update this policy when Coburg starts to process other data, for example when new features, providers or connections are added. The date at the top shows when it last changed.