Privacy Policy

Who we are

Ouroboro is operated by Hardest Cut Creations Inc ("we", "us"). We decide how the personal information described here is used, which makes us its controller. Our postal address is 3200 S Wadsworth Blvd, Num C29, Lakewood, CO 80227, and you can reach us at support@ouroboro.io. This policy covers the ouroboro.io website, the Ouroboro desktop app and its sidebar, the Ouroboro Chrome extension, the automation library, and our API at api.ouroboro.io (together, "Ouroboro").

Questions and requests go to support@ouroboro.io. We have not appointed a data protection officer; the same address reaches the people responsible for privacy.

Ouroboro is not offered in the European Economic Area, Switzerland or the United Kingdom yet.

The short version

• Every request the agent sends to an AI model goes through our servers at api.ouroboro.io and on to the AI provider under our account. That includes your messages, the text and screenshots of web pages the agent reads, the contents of files on your computer that it opens, and captcha images.
• Our servers pass those requests and replies through without storing their content. We keep a record of each request's cost and usage, because that is how your plan is billed.
• Your chats are saved on your computer. Unless you turn sharing off in Settings, a copy of each chat's text and activity timeline is also sent to our servers so our team can find and fix problems. Screenshots and images are never sent, and keys and passwords are masked first. Sharing is on by default, and you can turn it off at any time in Settings, then Sharing. See Your chats.
• The agent's memory, its logs and the files it makes are saved on your computer.
• When you give stars to one of the agent's runs, we store a record of that run on our servers, including what you asked and how the run went.
• We also keep your account, your plan and charges, your activity in the automation library, and anything you publish.
• We do not sell personal information or share it for advertising, and Ouroboro has no advertising trackers.

Information we collect

Your account

You can use Ouroboro with an anonymous device account or sign in with email or Google through our sign-in provider, Clerk.

• A device account is a random token kept in your Mac's keychain and in the app's local storage. We store only a hash of it, with an install ID and the date the token expires (180 days after it was last used).
• For a signed-in account we store your Clerk user ID, whether your email address is verified, a display name (your name, your username, or the part of your email address before the @), and your profile picture link. Clerk keeps your email address.
• When you link the app to a signed-in account, the app opens ouroboro.io with your email address filled in. We keep that email address on the link record, for display and to audit the link, and delete link records about 30 days after they expire.
• We also keep your account's status, and whether it has been suspended or closed for breaking our Terms of Service.

Network information

To prevent abuse and enforce limits we store a keyed, one-way hash of a shortened form of your IP address (the first three parts of an IPv4 address, or the first 48 bits of an IPv6 address). We keep it with your account, your library runs, ratings and reports, and our rate-limit counters. We do not store your IP address in our database.

Our hosting provider, Cloudflare, processes the IP address and browser details of every request. Our API's request logs in our Cloudflare account can include them and are kept for up to 7 days.

Runs you rate

After the agent finishes a request, you can give the run one to five stars. Rating is optional. When you rate a run, the app sends a record of it to our servers, and we keep it with your stars:

• what you asked the agent to do (up to 2,000 characters), the run's title, and its outcome (up to 1,000 characters);
• a short list of its steps, the websites it visited (domain names only), which tools it used and how often;
• the model and effort level, how long it ran, how many turns, tool calls and tokens it used, and how it ended;
• if the app includes them, up to four questions the agent asked you during the run and your short answers.

We index the text of these records so it can be searched, and we use them to study and improve how the agent works. Runs you do not rate are not sent as rating records, though their text goes with the copy of your chat described under Your chats while sharing is on. Deleting a chat in the app deletes the records of its rated runs from our servers, and deleting your account deletes all of them.

Your plan, billing and usage

• Your plan, your Stripe customer ID, and your subscription's status and dates.
• For every AI request the agent makes through our servers: the provider, the model, the kind of request, what it cost us and what we charged your plan, the provider's usage figures (such as token or character counts, never the content), the provider's request ID, and the time.
• While a request is running we hold an estimate of its cost against your allowance. Each hold is deleted when the request is charged or ends.

Stripe collects your card or other payment details and your email address directly when you subscribe. We never receive your full card number.

Your activity in the automation library

• Runs of library automations: which automation, a chat identifier, start and end times, how many turns and actions it took, how it ended, and where you started it from.
• Daily summaries for each automation you see or run: where its card was shown, whether you opened, ran or shared it, and how long you ran it.
• Ratings, reviews and reports you submit. Reviews are public and show your display name and picture.
• Shares you make, and when someone opens a link you shared (their account ID, or the hashed network information above if they are not signed in).
• The topics and apps you tell us you are interested in, used with your library activity to recommend automations.

Safety events

When the agent's safety guard blocks an action during a library automation, the app reports the blocked web address (up to 300 characters) or file name so we can improve the guard and warn other users about that automation.

Automations you publish

When you publish, we store the title, description, prompt, questions, apps, categories, trigger and thumbnail image, every version you submit, and the results of our safety review, including the web addresses, host names and email addresses it found in your text, and any strikes, bans, notices and appeals.

Publisher payouts

If you set up payouts, Stripe collects your identity, tax and bank details directly. We store your Stripe account ID and its status, the earnings we record for you, each transfer we send, and any amount taken back or held because a buyer's payment was refunded or disputed.

Messages to us

If you email us, we keep your message, your email address and our reply to answer you and to keep a record of requests.

What happens when the agent works

Ouroboro's agent is powered by AI models from other companies. Every request it sends to a model goes from the app to api.ouroboro.io, which checks your plan, forwards the request to the provider with our own account keys, and records the cost. The app does not use AI provider keys of your own. A request can carry:

• your messages and the instructions, memory and context the app adds;
• the text and screenshots of web pages the agent reads in your browser;
• the contents of files on your computer that the agent opens or that you give it, and the output of commands it runs;
• captcha images the agent is asked to solve, and screenshots it uses to find where to click.

Our servers forward this content and the replies without storing them. These are the providers that receive it:

• Anthropic, for the agent itself (Claude), chat titles and small helper tasks, and the safety review and showcase of published automations.
• OpenRouter, which passes requests on to other companies' models: chat titles, hints and page-watch checks (OpenAI's gpt-oss-20b and Mistral AI's Mistral Nemo), captcha reading (Google's Gemini, OpenAI's GPT and Xiaomi's MiMo models), video (Google's Veo), and any other model your plan lets you choose. OpenRouter picks which company runs each request. For open models these are hosting companies such as DeepInfra, Parasail, Groq, CoreWeave, Amazon Bedrock, Novita or the model's maker; for Google and OpenAI models they are Google, OpenAI, Microsoft Azure or Amazon Bedrock; for Claude they are Anthropic, Amazon Bedrock, Google or Microsoft Azure.
• fal, for generating images and video when you ask for them.
• ElevenLabs, for speech and video when you ask for them.
• TypeSafe, for solving captchas, finding where to click on a screenshot, checking whether a reply is a refusal before you see it, and checking parts of pages you asked the agent to watch.

We have not set up zero-data-retention or no-training arrangements with any of these providers, and our requests do not ask OpenRouter to avoid companies that may keep or train on prompts. Each provider handles the data under its own terms. When we checked their published policies on October 29, 2026, they said:

• Anthropic does not use inputs and outputs from its API to train its models by default. It deletes them within 30 days, except that it keeps inputs and outputs for up to 2 years (and safety scores for up to 7 years) when its systems flag them as violating its Usage Policy, and longer when the law requires.
• OpenRouter does not store prompts or responses unless the customer turns on logging, or the content is flagged as possible child sexual abuse material. It stores request details such as token counts and timing. Each company it routes to has its own retention and training policy, and some of those companies may keep or train on prompts.
• Google does not use prompts or responses to its paid Gemini API to improve its products. It logs them for a limited time to detect abuse and for legal disclosures, and may store them temporarily in any country where it has facilities.
• OpenAI does not use data sent to its API to train its models unless the customer opts in. It keeps abuse monitoring logs, which can include prompts and responses, for up to 30 days.
• fal's privacy policy says it collects the prompts and files you send and uses the information it collects to provide, maintain and improve its services. It does not say that data sent through its API is excluded.
• ElevenLabs' privacy policy says it may use inputs and outputs, including voice data, to research, build, train and improve its AI models, and that data it processes for a business customer is governed by that customer's contract.
• TypeSafe says it will not train or fine-tune AI models on what you send it and will not give it to third parties other than its service providers.

Your chats

The desktop app saves every chat on your computer. Unless you turn sharing off, it also sends a copy of each chat's text and activity timeline to our servers, so our team can find and fix problems with the agent. This copy is separate from the agent's requests to AI providers, which our servers do not store.

What we receive

• The text of the chat: your messages, the agent's replies, and what the agent read as text, such as the text of web pages, the contents of files and the output of commands.
• The chat's title, the task you gave the agent, and when the chat started and last changed.
• The chat's activity timeline: the steps the agent took, the tools it used, how long each took, and any errors.

What we do not receive

Screenshots, images, documents and other files that are not text are removed on your device before anything is sent. We keep only a note that something was left out. The app also masks API keys, passwords, tokens and similar secrets in the text on your device. It looks for the shapes such secrets usually take, so a secret in an unusual form could still get through. If you do not want us to see something, turn sharing off before you start.

When it is sent

Sharing is on by default, and you can turn it off at any time in Settings, then Sharing. While it is on, the app sends a chat about a minute after each run ends, and sends the chats you already have once when it first starts. If you turn sharing back on yourself, the app sends the chats not yet sent at once.

Who can read them

Chats are stored in private storage at Cloudflare (R2), with an index of the chat's ID, title, the start of its task, its size and its times in our database (D1). Only Ouroboro staff can read them, through an administrator login, and we read them to find and fix problems. We do not make them public or show them to other users. Chats you share may hold personal information about you and about the people in the pages and files the agent read for you; see Other people's information.

Turning it off and deleting a chat

• To turn sharing off, open Settings, then Sharing. The app stops sending chats at once. This does not delete what we already received.
• To delete a chat, delete it in the app. That deletes its copy from our servers too, and the app tries again later if it cannot reach us.
• To have everything deleted, email support@ouroboro.io or delete your account. We keep a chat's copy until you delete the chat or your account.

What stays on your device

Full chat transcripts and their activity timelines, the agent's memory, the app's logs, the files the agent creates and your settings are saved on your computer. The app's logs record only the length of text the agent types into pages, never the text itself. Dictation turns your speech into text on your Mac with Apple's speech recognition; the audio is not sent to us. We receive none of this, except the content the agent sends to AI providers as described above, a text copy of your chats and their timelines unless you turn sharing off (see Your chats), the records of runs you rate, and automations you choose to publish.

If you ask the agent to put a website it built live, it uploads the site to Cloudflare Pages using a Cloudflare account and token you give it. That site goes to your Cloudflare account, not ours.

Mac permissions

• Accessibility lets the app click and type for you.
• Screen Recording lets the app show your browser window live inside Ouroboro. Each frame is kept in memory only long enough to show it (or in a private temporary file that is deleted at once). Frames are not recorded and are not sent to us.
• Microphone access is used only for dictation, on your Mac.

The Chrome extension

The Ouroboro extension lets the agent work in its own tabs in Chrome. It has no server of its own: it sends what it reads only to the Ouroboro app on your computer, which uses it as described under What happens when the agent works. Its permissions and why it needs them:

• debugger: to read pages and take screenshots in the tabs the agent works in, and to click, type and scroll there.
• Access to all websites: so the agent can work on any site you send it to.
• tabGroups: to group and find the agent's tabs.
• scripting: to run the agent's page reading inside a page.
• nativeMessaging: to talk to the Ouroboro app on your computer.
• alarms: to try connecting to the app again every 30 seconds when the connection is down, such as after the app restarts.

The extension also runs a one-line script on ouroboro.io and singularitystudiosdev.github.io, the shared automation pages, and on localhost. The script only tells the page that the extension is installed and which version it is. It reads nothing else from those pages.

Other people's information

The pages, emails, messages and files the agent reads for you can show other people's information, such as names, addresses and the contents of their messages. That information goes to the AI providers as part of the agent's requests, and short parts of it can appear in the record of a run you rate and in the text of the chats you share. We use it only to do what you asked and to improve the agent, never to build profiles of those people. If you think we hold information about you because someone else used Ouroboro, contact us and we will help.

Publishing, safety review and the showcase

• When you publish an automation, we send its text and thumbnail to Anthropic for an automated safety review. We check the web addresses in it with Cloudflare, Google Web Risk and rdap.org. We also download public lists of dangerous web addresses; we send nothing about you to get them.
• Your published listing, your publisher page, your display name and picture, and reviews of your automations are public.
• Every 15 minutes our servers may take the top automation in the library and send its listing (title, description, prompt, questions and trigger) to Anthropic's Claude, which writes a short demonstration of it for the ouroboro.io home page. We store that demonstration with the listing.

Automated decisions

Our safety review decides automatically whether a published automation goes live, needs a person to review it, or is blocked. A blocked automation can lead to a warning, a strike, a ban from publishing or, after repeated serious violations, the closing of your account. Reports from other users, and safety events from runs, can send a listing back to review. You get a notice explaining each decision and the rule behind it, and you can appeal a strike in the app; a person decides every appeal. We also rank and recommend automations automatically from library activity and the interests you pick; this decides what you see first, not what you can use.

Website analytics

The ouroboro.io website counts how its pages are used with analytics of our own. They run on our own Cloudflare account, and no analytics company receives them. As you use the site, your browser sends us:

• The page you visit, the name of the site that sent you (never its full address), any campaign tags in the link, your screen size, language and time zone, and whether you use touch or reduced motion.
• Which parts of the page you look at and for how long, how far you scroll, and where you click: the name of what you clicked, its label, and where in that part of the page the click landed.
• Whether a download or an Open Ouroboro button worked, and any errors the page runs into.

Our server adds your country and the kind of device, browser and operating system you use, and records each app download with the build and the button it came from. It does not store your IP address or your full browser details. We never collect anything you type on the site, such as a search, and nothing from forms.

To tell visits apart, your browser keeps a random visit ID while you browse and a random visitor ID in its local storage, so that coming back counts as the same visitor. Neither is a cookie, and neither is tied to your name, email address or account. If your browser sends Global Privacy Control or Do Not Track, it keeps no visitor ID and each visit counts on its own. Clearing your browser's data for ouroboro.io removes the ID at any time. We delete website analytics after 400 days. Cloudflare, which hosts the site, may also measure page loads for us with its cookieless Web Analytics.

Cookies and storage

The ouroboro.io website sets no cookies of its own, and its analytics do not use cookies; they keep a random visitor ID in your browser's local storage, as described above. The Ouroboro sidebar keeps your device token, your settings, and ratings waiting to be sent in its local and session storage. When you sign in, Clerk sets the cookies it needs to keep you signed in, and the sign-in page uses Cloudflare Turnstile to check that you are a person.

Why we use information, and our legal basis

Where a law such as the GDPR requires a legal basis for using personal information, these are ours:

• To provide Ouroboro (your account, the agent's requests to AI providers, the library, publishing, billing and payouts): because it is necessary to perform our contract with you.
• To keep Ouroboro safe (the safety review, reports, strikes and bans, rate limits, the hashed network information, the safety guard, and preventing spam, fraud and payment abuse): our legitimate interest in protecting our users, the public and our service.
• To improve the agent from the runs you rate: our legitimate interest in making the agent work better. You choose whether to rate a run, and you can object or delete the records at any time.
• To find and fix problems from the chats you share: our legitimate interest in keeping the agent working. You can turn sharing off at any time, delete a chat, or object, and we will delete the copy.
• To rank and recommend automations and to show the showcase: our legitimate interest in running a useful library. The interests you pick yourself are used because you asked us to.
• Website analytics and fixing problems: our legitimate interest in understanding how ouroboro.io is used and keeping Ouroboro working.
• Billing, tax and payout records: to comply with our legal obligations.
• Answering your requests and messages: to comply with our legal obligations and our contract with you.

You do not have to give us any information, but without an account the app cannot send the agent's requests, the Free plan needs a verified email address, and payouts need the identity and tax details Stripe asks for.

Who we share information with

• Service providers that run parts of Ouroboro for us and may use the information only to do so: Cloudflare (hosting, our API, the D1 database, R2 storage for thumbnails and uploaded chats, request logs and the security checks above), Clerk (sign-in), Stripe (payments and payouts), and Tinybird (a copy of library events: your account ID, which automation, where its card was shown and when).
• The AI providers listed above, when you use the agent, and Anthropic and Google Web Risk for the safety review of published automations.
• Everyone, for what you make public: published automations, your publisher page, and reviews with your display name and picture.
• Authorities, when the law requires it or to protect people from harm.
• A buyer or successor, if Ouroboro is sold or merged, under this policy.

Where information is processed

We are based in the United States, and our providers process information there and in other countries; OpenRouter can route a request to a company in another country. Because Ouroboro is not offered in the European Economic Area, Switzerland or the United Kingdom, we do not rely on a transfer mechanism for those regions today. For reference, on October 29, 2026 the U.S. Department of Commerce's Data Privacy Framework list showed Cloudflare, Clerk, Stripe, Google and ElevenLabs as active participants in the EU-U.S. Data Privacy Framework, its UK Extension and the Swiss-U.S. Data Privacy Framework. Anthropic, OpenRouter, fal, TypeSafe, Tinybird and OpenAI were not on the list. If we offer Ouroboro in those regions, we will rely on the Data Privacy Framework for certified providers and on standard contractual clauses for the others, and update this policy first.

How long we keep information

• Your account, runs, ratings, reviews, reports, interests, shares, safety events and published automations: as long as your account exists.
• Records of runs you rate: until you delete that chat in the app, or your account.
• Uploaded chats (their text and activity timelines): until you delete that chat in the app, or your account. Turning sharing off does not delete them.
• Daily library summaries: 90 days.
• Rate-limit and abuse counters: deleted 2 days after the time window they count has ended. Most windows last between a minute and a day; the one that counts new device accounts per network lasts 30 days.
• Device link records, with the email address on them: about 30 days after the link expires.
• Holds on your allowance: until the request they belong to is charged or ends, and never more than about an hour after they expire.
• Billing, usage, earnings and payout records: as long as your account exists, and after that for as long as tax, accounting and payment law requires.
• Strikes for less serious violations expire after 180 days; strikes for the most serious violations do not expire. Safety review records: as long as the automation or your account exists.
• Website analytics: 400 days. Our API's request logs at Cloudflare: up to 7 days.
• Our database's backups let us restore it to any point in the last 30 days, so deleted information leaves them within 30 days.
• Content already sent to AI providers: under each provider's policy, as described above.

Your rights and how to use them

Wherever you live, you can ask us to:

• tell you what personal information we hold about you, including any chats we hold, and give you a copy, in a file you can take to another service;
• correct information that is wrong;
• delete your account and the information tied to it;
• stop or limit a use of your information that you object to, including uses based on our legitimate interests.

Email your request to support@ouroboro.io. To make sure the request comes from you, we will ask you to write from, or confirm from, the verified email address you sign in to Ouroboro with. If you use Ouroboro only with a device account, link it to a sign-in from the app first so we can verify you. We will complete your request within 30 days, using a tool that covers every table of personal information we hold.

If you are on a paid plan, the subscription has to be cancelled before we can delete your account, so that Stripe stops billing it. You can cancel it in the app under Settings, Billing, Manage subscription; a plan cancelled there ends at the close of the period you have paid for, and we can delete the account once it has ended. If you would rather not wait, say so in your deletion request and we will cancel the subscription with you right away, so the deletion is not held up.

When we delete an account, we delete its records from our database, and its thumbnails and uploaded chats from our storage, and remove what it published. We keep the billing, tax and payout records the law requires us to keep, with your name and other personal details removed where we can. Stripe keeps its own records of your payments and payouts under Stripe's privacy policy. Deleted information leaves our backups within 30 days, and content already sent to AI providers is kept under their policies. Deleting your Clerk account removes the email address Clerk holds.

You can also delete things yourself: deleting a chat in the app removes its uploaded copy and the records of its rated runs from our servers, and you can delete your reviews and published automations in the app. We will not treat you differently for using any of these rights.

Notice for California and other U.S. states

This section is our notice at collection under the California Consumer Privacy Act, as amended by the CPRA, and gives the disclosures other U.S. state privacy laws require. In the last 12 months we have collected these categories of personal information, from you, from your device and browser, and from Clerk and Stripe:

• Identifiers: account and device IDs, display name, email address on link records, and hashed network information.
• Customer records: your name and profile picture link, and payment details, which Stripe collects.
• Commercial information: your plan, subscription, charges, usage and payouts.
• Internet or other electronic network activity: your activity in the library and on ouroboro.io, safety events, the records of runs you rate, including the websites they visited, and your uploaded chats and their activity timelines.
• Audio, electronic and visual information: thumbnails you publish, and screenshots, captcha images and dictation that the app handles. Screenshots and captcha images go through our servers to AI providers without being stored, and are removed from your chats on your device before a copy is uploaded; dictation stays on your Mac.
• Contents of your chats: the text of the chats you share (see Your chats). It can fall in any category above, because it holds what you and the agent wrote and what the agent read, and it may hold personal information about you and about the people in the pages and files the agent read. We use it to find and fix problems, keep it until you delete the chat or your account, and store no screenshots.
• Geolocation: the country of website visitors only, never your precise location.
• Sensitive personal information: your account sign-in (we store device tokens only as hashes), and the contents of email, messages and files the agent reads for you, which go through our servers to AI providers without being stored. When that content appears as text in a chat you share, we store that text as described under Your chats. We use it only to provide Ouroboro, keep it secure and fix problems, so the right to limit its use does not apply.

We use each category for the purposes in Why we use information, disclose it for business purposes only to the service providers and AI providers in Who we share information with, and keep it for the periods in How long we keep information. We do not sell personal information, share it for cross-context behavioral advertising, use it for targeted advertising, or profile people in ways that produce legal or similarly significant effects; we have never done so. So there is nothing to opt out of, and we honor Global Privacy Control on ouroboro.io as described above.

Depending on your state, you have the right to know and access the personal information we hold about you, to get a portable copy, to correct it, and to delete it, as described in Your rights. You can use an authorized agent; we will ask the agent for proof of your permission and may ask you to verify your identity with us. If we decline your request, our reply will explain why, and you can appeal by replying to it with the word "Appeal". We will answer your appeal in writing within 45 days. If we deny your appeal, you can contact your state's attorney general.

Complaints

If something here worries you, please contact us first and we will try to fix it. If a data protection law such as the GDPR or UK GDPR applies to you, you also have the right to complain to the data protection supervisory authority where you live, where you work, or where you think the law was broken. In the United Kingdom that is the Information Commissioner's Office, at ico.org.uk/make-a-complaint. The European Data Protection Board lists the authorities in the European Economic Area at edpb.europa.eu.

Google sign-in and the Chrome Web Store

If you sign in with Google, Google shares your name, email address, profile picture and Google account ID with Clerk, and Clerk shares the account details listed above with us. We use this information only to create your Ouroboro account, sign you in, and show your name and picture where you choose to appear, such as on reviews and published automations. We do not request access to your Gmail, Drive, Calendar or any other Google data. We do not sell Google user data, use it for advertising, or use it to train AI models.

The use of information received from Google APIs will adhere to the Chrome Web Store User Data Policy, including the Limited Use requirements.

Children

Ouroboro is only for people 18 and older, as our Terms of Service say. We do not knowingly collect personal information from anyone under 18. If you believe someone under 18 has given us information, contact us and we will delete it.

Security

All traffic to Ouroboro is encrypted in transit. Device tokens are stored only as hashes, IP addresses only as keyed hashes of a shortened address, and our service keys are kept in protected secret storage. No system is perfectly secure, so please tell us at once if you think your account has been compromised.

Changes to this policy

If we change this policy, we will update the date at the top. If a change is significant, we will tell you in the app or on this site before it takes effect.

Contact

Hardest Cut Creations Inc, 3200 S Wadsworth Blvd, Num C29, Lakewood, CO 80227. Email support@ouroboro.io.