Privacy policy
Olympus is a knowledge engine that runs on your computer (a Mac, or a Linux computer running Omarchy or Ubuntu 24.04 LTS) and answers questions about your own mail, files and notes, including questions you ask in ChatGPT, Claude, Grok or another AI app you connect. It is published by Open Coordination Unlimited, Inc. ("OCU", "we", "us"), a United States company, which is responsible for the personal data described here. This policy explains what data Olympus processes, where, why, who receives it, how long it is kept, and how you control it. When it changes, the date above changes.
In short: Olympus stores your source collection and the index it builds on your computer; we do not maintain a hosted copy. When you ask a question in ChatGPT, ChatGPT receives the question it sent, plus answers and excerpts built from items Olympus has labeled Personal, with citations. Items labeled Private are excluded from what ChatGPT's model receives; answers from them are shown to you in an Olympus panel inside ChatGPT, encrypted between that panel and your computer. Requests between ChatGPT and your computer pass through our relay, which forwards them without intentionally storing their content. Claude, Grok and other AI apps you connect receive the same kinds of results from Personal items; for Private items they receive at most an answer written from them on your computer (by Olympus's private model, or by Venice if you set it up), never the items themselves, though that answer can repeat what the items say; otherwise they are told only that the answer is in Olympus on your computer. Olympus keeps the questions and answers of that private inbox on your computer until you delete them, revoke the app that asked, or delete a source they drew on. When you connect Google, your computer signs in with Google directly. If you set up anonymous questions, a question you choose to ask goes to zkAPI and an AI model provider without your identity attached, though what you ask can still identify you. Automatic labels can be wrong.
Privacy policy in brief
The same facts as the detailed sections below, under the usual headings.
Personal data we collect
OCU holds little personal data, because Olympus runs on your computer. Olympus does not require an OCU user account, and OCU does not currently take payments for Olympus. What we hold:
- Relay registry: for each install, an install identifier derived from the install's public key, the public key, and timestamps.
- Relay event logs: event types (for example "session ready"), times, fixed diagnostic fields and a shortened one-way hash of the install identifier.
- Usage totals: daily counts of how many installs connected to the relay, how many were new and how many requests it forwarded. They name no install.
- Download counts: daily counts of how often the installers and uninstallers on our website were downloaded. See Our website.
- In memory only: IP addresses and request counters used to limit abuse, and which installs are online.
- In transit: your questions, answers, evidence excerpts and the other data described in What ChatGPT receives, while the relay forwards them between ChatGPT, Claude, Grok, Grok Bot or Cursor and your computer.
- Google sign-in: when you connect Gmail or Google Drive through Olympus's shared Google app (the default), your computer signs in with Google and refreshes its tokens with Google directly; no Olympus service takes part. Olympus versions before 1.0.0 that used the shared Google app signed in and refreshed through a sign-in service we ran on Cloudflare at
auth.olympusplugin.ai; we retired it before releasing 1.0.0 and deleted the keyed account hashes and first-sign-in times it kept. The page that returns you to Olympus after you sign in to Dropbox is served by Cloudflare atauth.olympusplugin.ai, whose request logs can record the one-time sign-in code and state in that page's address; the code works only once, expires within minutes, and cannot be used without a one-time key that never leaves your computer. - Support correspondence: your email address, name if you give it, and anything else you send us.
On your computer, under your control and not collected by us, Olympus processes the content of the sources you connect, the index it builds, sign-in tokens for your accounts, your settings and local logs (see Categories of data).
How we use it
To answer your questions from your own sources when you ask, to connect ChatGPT, Claude, Grok or another AI app to your computer once you approve it there, to keep the relay secure and working, and to answer support requests. We do not sell personal data, use it for advertising, build profiles, or train models on it.
Who receives it
- OpenAI (ChatGPT) receives your questions and what Olympus returns: answers and excerpts from Personal items, citations, item names, status, and the data shown in Olympus panels (see What ChatGPT receives).
- Anthropic (Claude), xAI (Grok and Grok Bot), Anysphere (Cursor, and Grok Bot, which connects to Olympus through Anysphere's Cursor sign-in) and other AI apps you connect receive your questions and what Olympus returns to them (see What other AI apps receive).
- Google, Dropbox and other sources you connect are read directly by your computer with the access you grant.
- Services you choose to set up: Venice, for private answers and, if you choose, for labeling items, building search vectors and reading images and PDF pages; zkAPI and the AI model provider it reaches, for anonymous questions (see Anonymous questions); any other provider you configure.
- Our service providers: Hetzner Online GmbH (relay and website hosting, Germany), Cloudflare, Inc. (the page that returns you to Olympus after you sign in to Dropbox) and our email service provider (support email).
- Download hosts: Hugging Face, GitHub, the Python Package Index, the npm registry and, if you use those features, the Tor Project and Go's module proxy see your computer's IP address when it downloads models and tools (see Third parties and recipients).
- Update check: on a Mac or Linux computer, about once a day, Olympus checks olympusplugin.ai for a new version by downloading the public install script. The check sends no identifiers, cookies or query string; like any request to our website, it reaches our server from your computer's IP address. It is skipped when your computer uses a proxy, and
OLYMPUS_UPDATE_CHECK=offturns it off. - Authorities, where the law requires, limited to information we hold.
How long we keep it
- Relay registry: active registrations expire after about 90 days without a connection, or sooner if you ask (see Retention for what the relay keeps after that).
- Relay event logs: 14 days.
- Sign-in return page: Cloudflare's request logs for the page that returns you after Dropbox sign-in, including the one-time code they may record, are kept under Cloudflare's own retention for our account; we use them only to troubleshoot. The Google sign-in service used by Olympus versions before 1.0.0 is retired and its records deleted.
- In-memory data on the relay: only while needed; the relay's code does not write it to disk.
- Support email: until your request is resolved, then deleted unless the law requires longer.
- Account and billing data: none; there are no OCU accounts and we do not currently take payments.
- On your computer: until you delete it. Finished pending answers about 15 minutes; private answer jobs 10 minutes and private question jobs 30 minutes, in memory.
Your controls and rights
You choose which sources Olympus reads, and you can disconnect ChatGPT or any other AI app, export Olympus's data from your computer, or delete it, at any time (see Your controls). To get a copy of the personal data we hold, correct it, or delete it, email support@olympusplugin.ai. We respond without undue delay, normally within 30 days and within any shorter deadline the law sets. We may ask for information to confirm the request is yours. Rights you have under the law where you live are not affected.
Contact
Privacy questions: support@olympusplugin.ai. Open Coordination Unlimited, Inc., 30 N Gould St Ste R, Sheridan, WY 82801-6317, USA.
Where your data is processed
| Place | What is processed there | Who controls it |
|---|---|---|
| Your computer | Everything Olympus reads from the sources you connect; the search index, built-in models and labels it builds; sign-in tokens for the accounts you connect; Olympus settings and logs. | You. OCU has no access to it. |
The Olympus relay (mcp.olympusplugin.ai) |
Requests between ChatGPT, Claude, Grok, Grok Bot or Cursor and your computer, in transit, and the small registry and logs described below. See What the relay processes. | OCU, on a server rented from Hetzner Online GmbH in Germany. |
The sign-in return page (auth.olympusplugin.ai) |
The one-time sign-in code and state in the page's address after you sign in to Dropbox, which Cloudflare's request logs can record. See Personal data we collect. | OCU, on Cloudflare. |
| ChatGPT (OpenAI) | Your conversation, what Olympus returns to it, and Olympus panels. See What ChatGPT receives. | OpenAI, under OpenAI's own terms and privacy policy. |
| Other AI apps you connect (for example Claude, Grok, Grok Bot or Cursor) | Your conversation and what Olympus returns to it. See What other AI apps receive. | Their providers (for example Anthropic, xAI or Anysphere; for Grok Bot, both xAI and Anysphere), under their own terms and privacy policies. |
| Accounts you connect (for example Google or Dropbox) | Your computer reads your items directly from these services using access you grant. | Those providers, under your agreement with them. |
| Services you choose to set up (Venice, zkAPI) | What you choose to send them. See Third parties and recipients. | Those services, under their own terms. |
The three labels
Olympus labels every item, one by one, as Personal, Private or Secret. Item names and titles are labeled separately from their contents, and are Personal by default, so the name of a Private item may be visible in ChatGPT even though its contents are not.
- Personal items may be used to answer questions in ChatGPT and the other AI apps you connect.
- Private items (by default, for example health, money, legal and identity matters; once you describe in your own words what is private for you, only the kinds you call private, plus any folders, labels or senders you mark as always Private) are excluded from the answers and excerpts ChatGPT's model receives. They are answered on your computer, by the private model built into Olympus or by Venice if you choose to set it up, and the answer is shown to you in the private answer panel (see The private answer panel). Other AI apps you connect receive at most an answer written from them on your computer, never their text (see What other AI apps receive). Unless your privacy settings keep Private items to keyword search, Olympus also finds them by meaning. By default the model built into Olympus builds their search vectors on your computer; if you set up an embedding provider for search vectors instead, such as Venice, Olympus sends that provider the text of Private items and it builds the vectors (see Third parties and recipients).
- Secret items (for example passwords, recovery codes and private keys) that Olympus recognizes are never sent anywhere once recognized, and their contents are kept out of Olympus's search index; they are excluded from answer retrieval and model inputs. In the index, Olympus keeps where each one is: the source, its location, a screened title, the kinds of secret found and a one-way hash of the content. If an item becomes Secret after Olympus has indexed it, including when you mark it Secret, its contents are removed from the index, where Olympus keeps only that record of where it is. Telegram and WhatsApp messages stay in the raw message store on your computer until you remove that source, except that a message Olympus recognizes as Secret has its text (and any saved voice note) removed from that store. If you turn on Venice for reading files, Olympus first reads each image or PDF page on your computer and does not send it to Venice if it recognizes a secret in it or cannot read it; a secret that this local reading misses can still reach Venice. Recognition looks for common forms, such as private keys, API keys and labeled passwords, and can miss others; an item it misses is labeled like any other item.
Labeling is automatic and can be wrong. A new item with nothing that marks it as your own record is labeled Personal at once; Olympus then checks it on your computer and moves it to Private if the check finds it is, and an item whose check cannot finish is treated as Private. If an AI app read an item while it was Personal, what that app received stays with it. When an item moves to Private, Olympus removes the text and search vectors of its Personal copy from your computer, leaving only its name and similar details, which are Personal by default, and withholds any answer built from the item that has not yet been delivered.
Categories of data
Processed on your computer
- Content from sources you connect: mail, documents and files from accounts you connect, such as Gmail, Google Drive and Dropbox, and from other sources you set up on your computer, such as X bookmarks, Readwise, Telegram or WhatsApp. This can include any information those items contain, including text, images and audio.
- Derived data: the search index, text extracted from documents and recordings, vector representations made by the built-in models (by default for Private items too), labels and their reasons.
- Credentials: the sign-in tokens that let Olympus read the accounts you connect, the keys and tokens that link your computer to ChatGPT, and, if you set up anonymous questions, your zkAPI account keys.
- The private inbox: the questions AI apps sent that Private items may answer and the questions you ask in Ask a question, with their answers, the citations they drew on and what was sent to Venice or zkAPI (see The private inbox).
- Settings and logs: your Olympus settings and local log files. Olympus's logs are designed not to contain message content or tokens.
What ChatGPT (OpenAI) receives
Olympus returns data to ChatGPT in two ways: tool results that ChatGPT's model reads, and Olympus panels displayed inside ChatGPT. Panel-only data is not given to ChatGPT's model, but it passes through OpenAI's systems.
- Questions and answers: the questions ChatGPT sends to Olympus on your behalf; answers and evidence excerpts built from Personal items; citations with source name, title, date and link where available; item names that are Personal, which can include the name of an item whose contents are Private; how many items were searched or not yet read; and temporary identifiers used to collect an answer that is still being prepared.
- When Private items match: a fixed note to ChatGPT's model saying so, with no titles or content. For the private answer panel only: how many Private items matched, progress, a temporary job identifier and a public encryption key.
- The Olympus dashboard and status: connection and source status, when sources last synced and what was found, indexing and download progress, item counts, names of files Olympus could not read and why, model choices and their status, recent activity times, troubleshooting status and diagnostic references. When you set up remote access, it also shows the configured SSH user name or host name and local service port, to give connection instructions.
- Choosing folders and mail (panel only): folder names, identifiers or paths, how folders are nested, sizes and file counts; mail label names and identifiers, mail categories and counts; suggested sender addresses with how many messages each sent; the sender addresses you mark as always Private or skip; your saved choices; and technical revision markers that keep these choices consistent. Locations you keep Secret are left out.
- Connecting a source: the service you chose, a one-time sign-in link and when it expires.
- Your privacy settings: your own description of what is private for you, and how many folders, labels or senders you marked as always Private; the panel also shows which ones.
- Cited sources: when you open a cited source that is not on your computer, its web address, which ChatGPT opens for you.
- Anonymous questions asked in the conversation: the answer, the model that gave it, the exact question that was sent, the privacy level and preparation you chose, and whether network privacy was verified (see Anonymous questions).
- The private question panel: ChatGPT's model learns only that the panel opened, or why it could not. The panel receives a temporary job identifier, a public encryption key and your saved level; the question you type and its answer are encrypted between the panel and your computer.
- Errors: fixed error messages and, for some errors, an opaque diagnostic reference.
The private answer panel
When Private items match a question, an Olympus panel inside ChatGPT shows you an answer from them, marked "Not sent to ChatGPT". Your computer writes the answer and encrypts it with a key that only that panel holds; the panel fetches it from your computer through the relay, not through ChatGPT's conversation. ChatGPT's model does not receive it, and the relay forwards only encrypted data. The panel's code is served inside ChatGPT, so this protection also depends on your computer, the panel's code as ChatGPT delivers it, and the key exchange.
The panel keeps its encryption keys, job identifiers, creation times and question-panel progress in the browser storage ChatGPT gives it. Entries older than 24 hours are rejected or removed when the panel next uses or cleans that storage; unused entries can remain until a later cleanup or until you clear ChatGPT's site data. Answer text is not written to that storage.
What OpenAI does with your ChatGPT conversations, including Olympus results and panels, is governed by OpenAI's terms and privacy policy and your ChatGPT settings, not by this policy.
What other AI apps receive
Claude, Grok, Grok Bot, Cursor and other AI apps you connect through the relay receive the questions they send to Olympus on your behalf, and answers, evidence excerpts and citations built from Personal items, like ChatGPT's model. They do not receive the text of items labeled Private or Secret. Olympus panels, including the private answer panel, are shown only in ChatGPT. When Private items match a question, the app may receive an answer written from them by the private model on your computer, or by Venice if you set it up, checked for secrets, with each item's title, location, source and author, but never the items' text; that answer can repeat what the items say. When it does not receive one, Olympus records the question in the private inbox on your computer, where the private model built into Olympus answers it in the background, and tells the app only that the answer is in Olympus on your computer, with a link (olympus://open/private/ and a random identifier) that opens that answer there. The link carries no part of the question or the answer, and the answer is not sent to the app. This applies to both an app's questions and its searches. If Olympus cannot finish checking a question against Private items in time, the app is not told; Olympus keeps the question in the private inbox, checks it again later, and answers it there if Private items match.
Apps that run on your computer, such as Claude Code, Codex or Hermes, connect to Olympus there, without the relay, and receive the same. What any AI app's provider does with your conversations, including Olympus results, is governed by its own terms and privacy policy and your settings with it, not by this policy.
The private inbox
Olympus keeps an inbox on your computer, under On this computer in the Olympus dashboard (Inbox). It is shown only in an unlocked dashboard on your computer, Olympus gives no AI app a way to read it, and its contents are left out of exports, support bundles and telemetry. It is a file your computer account can open, so an app you let read files or run programs under that account, such as a coding agent, could open it like your other files.
- What it holds: each question an AI app you connect sent that Private items may answer, and each question you ask in Ask a question, with the answer, the citations it drew on (title, source, date, link and where the item is on your computer), and who asked. For a question you send to Venice it also records the titles, sources and dates of the items sent. For a question sent through zkAPI, including one an AI app asks for you with
ask_anonymously, it holds the question exactly as it was sent and the reply; a question you type in ChatGPT's private question panel is not kept. Answers can repeat what your Private items say. - Who answers: a question an AI app sent because Private items may answer it is answered only by the private model built into Olympus, on your computer. A question an AI app sends for you with
ask_anonymouslyis answered through zkAPI by the model provider it reaches (see Anonymous questions). In Ask a question you choose: the local model, a Venice model or a zkAPI model. The local model sends nothing off your computer. Venice receives the passages of your Personal and Private items that match the question, and only if your privacy settings allow Private items to go to Venice; connecting a Venice key alone does not allow it. zkAPI receives only the prepared question (see Anonymous questions). Answers shown in ChatGPT's private answer panel are not kept in the inbox, but if the panel does not collect an answer in time, Olympus may answer the question in the inbox instead. - How long: a question that is not answered is deleted after seven days. An answer stays until you delete it. You can delete answers one by one or all at once in the inbox. They are also deleted when you delete the data of a source they drew on, when you revoke the AI app that asked, and when you delete Olympus's data.
Anonymous questions (optional)
Anonymous questions are available on a Mac and on supported Linux x64 computers, such as Omarchy or Ubuntu 24.04 LTS, once Olympus's network isolation checks pass. If you set it up on your computer, Olympus can ask a frontier AI model a question for you through zkAPI, using your own zkAPI account. It does this only when you ask for an anonymous or private question, whether you type it in Ask a question in Olympus on your computer or an AI app sends it for you.
- What leaves your computer: the prepared question and the technical information needed to route and run it, such as the chosen model. Olympus does not attach source documents or conversation history. In Strict mode, your own model on your computer first rewrites the question into general questions to remove identifying details; in Standard mode, by default, your model removes names, contact details and account numbers, or, if you choose, it goes as you wrote it or is prepared by your own instructions, which then decide what is removed; either way it can keep identifying details. Olympus checks it before it leaves.
- Who receives it: the zkAPI service and the AI model provider it reaches, which can read the question. zkAPI is designed to separate who pays from what is asked, so your questions are not attached to your account. Olympus sends the question through Tor unless you turn Tor off, but cannot currently verify network anonymity on macOS. The content of a question can itself identify you.
- Where you ask it: a question you write in the ChatGPT conversation, and its answer, are seen by OpenAI and pass through our relay readable, like other Olympus results. A question you type in the private question panel, and its answer, are encrypted between the panel and your computer. A question you type in Ask a question in Olympus on your computer goes from your computer to zkAPI without passing through ChatGPT or our relay.
- What stays on your computer: your zkAPI account keys; a usage record (counts, costs and technical checks for each session); and, in the private inbox, each question exactly as it was sent and the reply (except a question you type in ChatGPT's private question panel), for as long as the inbox keeps answers (see How long).
- Your zkAPI account is your own, and is set up and managed in Olympus on your computer, outside ChatGPT. OCU does not sell zkAPI access or receive, hold or have custody of your zkAPI funds. Adding to the account is an Ethereum transaction, which is public on that network; zkAPI is designed so that your questions cannot be linked to it.
zkAPI and the model provider keep what they receive according to their own terms. OCU does not receive anonymous questions you type in the private question panel, or their answers, in readable form.
What the relay processes
ChatGPT, Claude, Grok, Grok Bot and Cursor reach your computer through the relay at mcp.olympusplugin.ai. Your computer keeps an encrypted, outgoing connection open to the relay, and the relay forwards their requests along it.
- In transit: the relay ends the encrypted connection from the AI app (TLS), so request and response contents pass through its memory while being forwarded. Its code does not intentionally store them.
- Logs: the relay's operational logs record event types, times, fixed diagnostic fields and a shortened one-way hash of your install's identifier. Its code does not intentionally log request or response content, tokens or credentials, query strings, or caller IP addresses. The web server in front of the relay is configured with access logging off.
- Registry: for each install, an install identifier derived from its public key, the public key, and timestamps. It holds no name, email, IP address, account or token.
- In memory only: IP addresses and counters used to limit abuse, and which installs are online.
- Tokens: the relay does not issue or check the tokens these apps use. Your computer does.
- Private answers and private questions: the relay forwards the panel's requests and your computer's replies encrypted, without the key to read them, and does not log them or their identifiers.
Our website
Our website, olympusplugin.ai, uses no cookies, analytics or trackers, loads nothing from other sites, and keeps no access log for its pages. To count downloads, the web server records each download of an installer, uninstaller or release file with only the address requested, the time and whether it succeeded: no IP address, browser or other detail. Once an hour these records become daily counts and are deleted.
On a Mac or Linux computer, Olympus also checks olympusplugin.ai about once a day for a new version, by downloading the public install script. That check sends no identifiers, cookies or query string; our server sees your computer's IP address, as with any request to the website. The check is skipped when your computer uses a proxy; set OLYMPUS_UPDATE_CHECK=off to turn it off.
Why we process data
- To answer your questions from your own sources, as you ask.
- To connect ChatGPT, Claude, Grok or another AI app you approve to your computer, and only to your computer, once you approve it there.
- To keep the relay secure and available: rate limits, abuse prevention, and fixing faults.
- To respond when you contact support.
We do not sell your data, use it for advertising, build profiles of you, or use your content to train models.
Where the EU or UK GDPR applies, we rely on: performance of our agreement with you, to run the relay and connect the AI apps you approve to your computer; our legitimate interests in keeping the relay secure and working; and our legal obligations where they apply. The content on your computer is processed by Olympus on your own computer under your control.
Third parties and recipients
- OpenAI provides ChatGPT, which you use to ask Olympus questions. It receives what is described in What ChatGPT receives.
- Anthropic, xAI and Anysphere: Anthropic provides Claude, xAI provides Grok, and Anysphere provides Cursor. Grok Bot connects to Olympus through Anysphere's Cursor sign-in, so both xAI and Anysphere may receive what it sends and gets. If you connect these apps, their providers receive what is described in What other AI apps receive.
- Google and Dropbox are read directly from your computer through their official APIs, with access you grant using your own accounts; Google sign-in and token refreshes go directly between your computer and Google (see Google sign-in). Olympus asks for read-only access: for Google, read-only access to Gmail, Google Drive and, if you choose, Google Calendar; for Dropbox, read access to file content, file metadata, sharing information and your account's basic information. You can withdraw that access in your account settings with those providers at any time. Olympus's use of information received from Google APIs adheres to the Google API Services User Data Policy, including the Limited Use requirements. Data from Google is used only to answer your own questions, is not transferred to others except as described in this policy (for example to ChatGPT, or another AI app you connect, when you ask), is not used for advertising, and is not read by people at OCU.
- Other sources you set up on your computer are read as follows: X bookmarks and Readwise through their official APIs; Telegram through its client API, signed in with your own account; WhatsApp through an unofficial client that connects as a linked device. WhatsApp's terms do not allow third-party clients, and WhatsApp may restrict or ban accounts that use one.
- Venice (optional). If you choose to connect a Venice account, Olympus may send item names and excerpts to Venice to label them as Personal or Private (Secrets are recognized on your computer and never sent), and the passages of your Personal and Private items that match a question to produce an answer, for example when you choose a Venice model in Ask a question. Private items go only if your privacy settings allow it; connecting a Venice key alone does not. If you choose Venice for search vectors, the text of each item Olympus indexes with Venice goes to Venice to build them; which items those are depends on how you set it up, and can include Personal items as well as Private ones. If you turn on Venice for reading files, Olympus also sends images and the first rendered page of PDFs to Venice to extract their text. Venice's handling, including how long it keeps what it receives, is governed by Venice's own terms and privacy policy.
- zkAPI and AI model providers (optional). If you set up anonymous questions, the prepared question and the technical information needed to run it go to the zkAPI service and the AI model provider it reaches (see Anonymous questions). Their handling is governed by their own terms.
- Other providers you configure yourself, for example an embedding provider, receive what that setting sends.
- Downloads: your computer downloads Olympus's built-in models directly from Hugging Face, and the runtimes and tools that run them from GitHub, the Python Package Index (
files.pythonhosted.org) and the npm registry (registry.npmjs.org); if you turn on anonymous questions, Tor comes from the Tor Project (dist.torproject.org); if you pair WhatsApp, its bridge is built on your computer with Go, which downloads its modules through Google's Go module proxy (proxy.golang.org) or their source repositories. Those services see your computer's IP address, as for any download. - Hosting: the relay and website run on a server rented from Hetzner Online GmbH in Germany. Email to support is handled by our email service provider, which processes messages on our behalf.
- Legal requests: we do not maintain a hosted copy of your source collection. We may disclose information we hold where the law requires it.
Retention
- On your computer: until you delete it. Uninstalling Olympus stops it but keeps its data until you delete that too (see Your controls).
- Answers in progress: a finished pending answer stays available on your computer for about 15 minutes, so the AI app that asked can collect it.
- Private panels: private answer jobs expire 10 minutes after they are created and private question jobs after 30 minutes. They are held in memory on your computer and removed by periodic cleanup. An answer already shown can stay visible while its panel is open. For the panel's browser storage, see The private answer panel.
- Relay registry: active registrations expire after about 90 days without a connection. Earlier registry entries stay on disk until the relay restarts and compacts its records. The relay keeps up to 20,000 expired install identifiers, with no fixed time limit, to recognise returning installs, and keeps the identifiers of revoked installs until they are restored, so that a blocked install cannot reconnect.
- Relay event logs: kept for 14 days, then deleted.
- Usage totals and download counts: kept for about two years. Download records become counts within an hour and are then deleted.
- Sign-in return page: Cloudflare's request logs for the page that returns you after Dropbox sign-in, including the one-time code they may record, are kept under Cloudflare's own retention for our account; we use them only to troubleshoot. The Google sign-in service used by Olympus versions before 1.0.0 was retired before 1.0.0 was released, and the keyed account hashes and first-sign-in times it kept were deleted.
- Support email: kept as long as needed to resolve your request, then deleted unless the law requires us to keep it longer.
- Anonymous questions: Olympus keeps a usage record on your computer (counts, costs and session checks) until you delete Olympus's data. The question as sent and the reply stay in the private inbox, as below.
- The private inbox: unanswered questions, seven days; answers, until you delete them or Olympus removes the oldest ones at its size limits (see The private inbox).
- Account and billing data: Olympus does not require an OCU account and OCU does not currently take payments for Olympus, so we hold none.
- OpenAI, Anthropic, xAI, zkAPI, Venice and the providers you connect keep data according to their own policies.
Your controls
- Choose sources: connect and disconnect sources in the Olympus dashboard.
- Disconnect an AI app: remove Olympus in ChatGPT, Claude or Grok, or revoke the app's access on your computer from the Olympus dashboard or with
olympus connections listandolympus connections revoke <id>. After revoking, the app cannot reach your computer until you approve it again on that computer. - Privacy settings: tell Olympus in your own words what is private for you, and mark folders, labels or senders as always Private, in the Olympus panel.
- Anonymous questions: off unless you set them up; turn them off at any time in the Olympus dashboard on your computer, under On this computer, Keys, in Anonymous answers.
- The private inbox: read it in Inbox on your computer, and delete answers there one by one or all at once.
- Revoke account access: remove Olympus's access in your Google or Dropbox account settings.
- Stop Olympus:
olympus engine uninstallstops Olympus and removes its login item (on Linux, its user service). It keeps your settings and data. - Export your data:
olympus data export --output <folder>exports supported indexed data and sanitized settings. Credentials and some operational state, including the private inbox, are left out; the export's manifest lists what was included and skipped. - Delete your data:
olympus data delete --alldeletes the index and data Olympus stored on your computer, including the private inbox (add--dry-runfirst to see what it will remove). This cannot be undone. - Relay registry: ask us to remove your install's registry entry by emailing support.
Depending on where you live, you may have rights to access, correct, delete or object to processing of personal data we hold, and to complain to a regulator. Because your source collection stays on your computer, the data we hold about you is limited to what is described above. To exercise a right, email support@olympusplugin.ai. We respond without undue delay, normally within 30 days and within any shorter deadline the law sets; where the law permits an extension, we will tell you why within the first deadline. We may ask for information to confirm the request is yours.
EU and UK users have the right to access, rectify, erase, restrict or object to processing of their personal data, to data portability, and to lodge a complaint with their local data protection authority.
California residents, where applicable law grants these rights, may ask to know, delete or correct personal information and may exercise these rights without discrimination. We do not sell personal information or share it for cross-context behavioral advertising, and we do not use sensitive personal information to infer characteristics about you.
Sensitive information
Connected sources can contain sensitive information, such as health or financial records, government identifiers, or passwords. Olympus processes the sources you select on your computer and uses automatic labels to limit what is disclosed; labels can be wrong. Olympus does not ask you for this information. Do not submit payment card data, protected health information, government identifiers, passwords, API keys or authentication codes through the ChatGPT plugin, its panels, or another AI app connected to Olympus.
Security
Connections are encrypted in transit. An AI app must be approved on your computer before it can reach it, and through the relay only the published clients of ChatGPT, Claude, Grok, Grok Bot and Cursor are accepted. The relay cannot issue tokens or approve new connections. No system is perfectly secure; if we learn of a breach affecting you, we will tell you as the law requires.
Children
You must be at least 13 to use Olympus, the same minimum age as ChatGPT. If you are under 18, you need permission from a parent or guardian. Olympus is not directed at children under 13, and we do not knowingly process personal data from them. If you believe a child under 13 has used Olympus, contact us and we will delete any data we hold about them.
International transfers
OCU is based in the United States. Our relay is hosted in Germany. OpenAI, Anthropic, xAI and the providers you choose may process information in other countries under their own policies.
Changes
We will post changes on this page with a new date, and for material changes we will give notice in the Olympus dashboard or by other reasonable means before they take effect.
Contact
Open Coordination Unlimited, Inc.
30 N Gould St Ste R
Sheridan, WY 82801-6317
USA
Email: support@olympusplugin.ai (write to us)