Roboty Privacy Notice

Who we are

Main Character LLC operates Roboty. This notice covers roboty.dev, browser experiments, and the Roboty Mac and iPhone software. The initial interactive software and beta are offered in the United States. Use Contact Roboty, Privacy topic for access, correction, deletion or other privacy requests. You do not need a Roboty account to make a request.

Reading, local experiments, pairing, diagnostics, public sharing and research are different activities. Permitting one does not permit the others. Optional hosted research collection is not active. Its sections below describe prepared policies, not an enrollment or a promise that the service is available.

Website requests and performance

Our hosting and network providers process requests needed to deliver and protect the website. These can include IP address, request time, requested URL and browser/device information. The editorial website uses Vercel Speed Insights for page performance measurements where enabled. That instrumentation is excluded from the experiment and Help surfaces. We do not use those measurements to add your private experiments to a research corpus.

On the public website and Roboty Guide, Google Analytics helps us understand visits and navigation. Google receives website usage and device/browser information, including visited page addresses and referrers. Its tag and analytics cookies load automatically after the page loads, unless you have turned analytics off for this browser. Advertising storage, advertising user data, personalization and Google signals are disabled by this integration. We do not add account identities, contact or beta form answers, private experiment records or native app activity to custom analytics events.

Choose Turn off analytics under Website analytics on Privacy to stop future measurement. Your choice is saved in this browser's local storage when available. You can change it through the Website analytics control on Privacy. Turning it off stops future measurement in this browser. It does not recall information already sent to Google or remove every previously stored cookie. Browser cookie and storage controls remain available. Google processes analytics data under its applicable service terms and privacy policy. Retention also depends on the property's configured settings.

Local browser storage retains preferences, reading progress and supported experiments. The software may also retain the version and time of your explicit terms agreement locally. That record is not an account, is not a research choice and is not automatically uploaded. Clearing storage may remove it.

Contact

The contact form sends your selected topic, message and optional reply email to our private support inbox through the configured email service, Resend. We use them to handle your request, reply and maintain the related support record. Please do not send credentials or unnecessary personal or experiment data. Contact messages are not automatically public posts or training material.

Beta applications

When beta applications are enabled, our website stores your email, optional name, selected platforms, intended use and submission date privately in a dedicated PostgreSQL database provided by Neon in the United States. Vercel processes the form request and sends it to that database. We use these to select testers and contact them about access. Applying does not create an account, subscribe you to marketing or grant research/training permission. Invitations are separate. Repeat applications retain the original answers. Contact us to request a correction or removal.

We retain applications while selecting and contacting testers, and support correspondence while needed to handle the request and related legitimate legal or security obligations. These forms are available only when their hosted intake is enabled. We do not represent local testing as completed hosted collection.

Browser experiments

Drafts, unfinished edits, attempts, numerical evidence and comparisons are saved in browser storage. It can be evicted or cleared. Export selected work to keep another copy. Explore does not capture camera/microphone data, private-world geometry or physical robot commands.

Copying an article shares its address. A compact setup link carries supported inputs. If hosted snapshot sharing is available and you publish after review, the selected public projection is uploaded. Anyone with its link can read and forward it. Unlisted is not private. Keep the separate management receipt to withdraw it. Withdrawal disables Roboty's corresponding page/data/preview/artifact routes. Downloaded copies and third-party caches cannot all be recalled.

Mac and iPhone records

Native workspaces keep supported drafts, experiments, results, preferences and recovery records in local device or Mac installation storage. Paired operations send the selected task and necessary protocol information to the Mac you trust. Pairing credentials and installation trust are separate from website accounts. Using the phone does not automatically synchronize your entire library or upload it to the website. Local and paired execution retain their respective owners.

Camera/AR access is requested for the feature using it. Viewing an AR camera does not by itself record or upload its feed. Recordings, calibration, spatial data, datasets, imports and exports have their own selected scopes. Inspect a selected file or disclosure preview before sharing it. A separate provider feature can send the approved input to its selected provider only through its own review and permission controls. These terms do not activate a provider or authorize a transfer.

We cannot restore every lost local file or recover ownership from device fingerprinting. Removing a download, unpairing a computer, deleting local records and withdrawing a public share are different actions. Their actual controls determine which copy is affected. Keep backups of important work.

Optional personal iCloud sync

Personal-library sync is optional and is not enabled in current development builds. When an eligible build offers it and you turn it on, supported saved projects, reusable setup descriptions, experiments, notes, approved personal knowledge, completed discussions and result summaries are sent to your private Roboty library in Apple iCloud using CloudKit. Apple processes this storage and delivery under your Apple Account and its applicable terms. Roboty does not use the website database for this copy.

The choice covers the saved content described in the app. Large private recordings, raw streams and datasets require separate explicit inclusion. Their descriptions can be present without the files. Downloads occur when selected. Credentials, pairing, active operations, local calibration validity, recovery permissions and live provider continuation are excluded. A saved discussion may contain information you wrote or chose to retain. Enabling sync does not authorize later disclosure to a model, another person, research or training.

Local work remains usable when offline. Turning sync off stops future sync on that device, retains local work and does not delete the iCloud copy. Changing Apple Accounts pauses the previous library and does not automatically transfer its content into the new account. Deleting an item, removing a local download, removing a cloud copy and forgetting eligible personal knowledge are different actions. Read the selected control before proceeding. Retained operation or evidence records may limit immediate erasure, and unavailable cloud-only files may not be recoverable after an account encryption-data reset.

Cloud records are private to the selected account. We do not promise that every cloud metadata field is end-to-end encrypted or that deletion instantly removes every backup or already received copy. Apple service availability, account storage and the app's supported schema affect delivery. Actual cloud activation, retention/processor review and signed-device verification remain separate from this prepared notice.

Your own operating agent

Reading public Help or its Markdown does not give an agent access to your work. Connecting an operating agent is a separate choice. Before granting access, review the selected experiments, operations and recipient. The local interface can let that client read the permitted records. A client running on your Mac may send what it receives to its own model provider or hosting service, according to that provider's policies and your configuration. Running locally does not mean its reasoning or storage stays local.

Review those recipients and their policies before permitting transmission. Share only the records needed for the task. Removing the agent's access blocks later authorized reads and operations, but cannot recall data it already received or forwarded. The operating-agent guide explains the separate access and permission steps.

Feedback and diagnostics

Send feedback and Report a bug deliver your message, optional reply email and basic app information to Roboty's private support inbox through Resend. Bug reports include error details by default. General feedback leaves them off. You can change Include error details and optionally preview the selected details. Pressing Send submits your message and the selected information. Viewing the preview is optional, and there is no second confirmation. These messages are used to understand problems, improve Roboty and reply when requested, not as research or training contributions.

The apps keep a bounded recent-action summary and the latest structured software failure locally for up to seven days. Selected report details can include version, OS and device-family information, screen/settings, error codes, timings, connection/readiness state and relevant recorded task outcomes. They exclude camera images, audio, room geometry, calibration files and credentials. Native and browser messages awaiting delivery are stored locally for best-effort delivery. Retry attempts stop within 23 hours or after eight attempts, and terminal or expired messages are removed. Delivery receipts retain hashes and delivery/budget bookkeeping, rather than message contents. Existing support-record retention applies to messages in the inbox.

Extended diagnostics

The iPhone diagnostic controls can record bounded local technical events for troubleshooting. Reports exclude camera images, audio, room geometry, drafts and credentials. Prepare a report, inspect it and choose Copy or Export only when you intend to disclose the selected file.

Sharing errors through Sentry is a separate timed opt-in during recording. It sends new scrubbed technical errors and supported crash/hang details during the consent period, not previously recorded local history. The control expires after 15 minutes. Turning it off cancels pending sends and discards unsent SDK reports. It cannot recall reports already sent. Other platforms do not gain permission to send diagnostics because you agreed on the phone. Read the current controls and diagnostic Help for supported routes.

Prepared research and training choices

Optional browser contribution is off by default and inactive pending the focused live privacy/license review. When separately activated and enabled, it covers future eligible synthetic evidence in that browser, including approved setups, revisions, applied controls, measured states, outcomes and recording gaps. It does not upload an old library or enroll another device. Automatic projections exclude free text, names, emails, account identifiers, host details, raw URLs/referrers, media, credentials and private-world coordinates.

Permitted contributions support workflow, controller and evaluation improvement. Model-training reuse, including commercial Roboty products, requires a separate default-off permission. Public dataset release and independent third-party training are not enabled. Hashing and account-free use do not make every network request anonymous. Hosting/security records must not be joined to reconstruct contributors. Contribution terms explain the narrower permissions.

Stop future uploads, withdraw research permission, withdraw training permission and delete records are separate choices. A pending request without acknowledgment is not confirmed server deletion. Re-enabling does not resurrect withdrawn material or expand earlier rights. Keep the private management receipt. If it and browser storage are lost, we cannot reconstruct ownership by fingerprinting you.

The prepared retention classes have different purposes and clocks:

Raw research and source-dependent reports. 90 days from server receipt of the original contribution. A report depending on several sources uses the earliest source expiry.
Selected reviewed numerical datasets. Up to 365 days from the original source receipt. Requires the separate longer-retention choice made when the evidence was captured and an exact dataset review. Converting it later does not restart the clock.
Reviewed nonidentifying technical findings. Up to 730 days from the exact output/privacy review. Requires the original permission for that reuse. It cannot retain the contributor's trace, coordinates or source identifiers.
Unsent browser uploads. 24 hours in the pending queue. Expired pending uploads are not later revived by a more permissive choice.

For example, turning a contributed trace into a report does not give the report another 90 days. A dataset selected months later still expires against its original receipt date. A separately reviewed finding may state a nonidentifying software limitation without retaining the underlying trace. It may outlive the source only under its original reuse permission and review. Longer-retention and training choices never broaden older captures.

Before live research collection, the actual processors, regions, backup deletion, security-log retention and minimal consent/deletion records must be reviewed and disclosed. Creating these notices does not complete that review. Withdrawal blocks governed future uses and source-dependent selections. Already trained outputs, legal retention and onward copies require their actual reviewed policy rather than a universal erasure promise.

Find the relevant data choice

Local experiments and reading. No experiment upload is required. Website delivery still involves hosting requests. Keep browser/device backups. Delete local work through its local library.
Optional iCloud sync. When available, supported saved content goes to your private Apple iCloud library only after opt-in. Large private files need selection. Turning off, changing account, removing a download and deleting cloud work affect different copies.
Paired work. Selected task and protocol information goes to the Mac you trust. Review pairing and the selected workspace. Removing pairing does not delete that Mac's records.
Feedback or bug report. The message and selected app details go through Resend to Roboty's support inbox. Change details before Send. Preview is optional. Use the Privacy contact topic for a sent-message request.
Your operating agent. Permitted records go to the selected client and may go to its own model or host. Review recipient and transmission before access. Revocation cannot recall received copies.
Public sharing, if available. Only the material in the reviewed public snapshot is uploaded for anyone with its link. Review before publication. Keep the private management receipt to withdraw Roboty's copy.
Prepared research. Selected future synthetic evidence would go to Roboty's contribution service. Collection is currently inactive. Research, training and longer dataset retention remain distinct choices. Prior-record withdrawal tools remain separate from enrollment.

Providers, retention and requests

Roboty uses Apple iCloud/CloudKit for separately enabled personal-library sync, Vercel for website hosting/performance, Resend for the configured contact email route, Neon for beta application storage, Google for public-website analytics, and Sentry for separately enabled technical error sharing. Selected local providers and separately activated database/object-storage services depend on their actual feature configuration and disclosure. Service processing does not grant independent model-training rights. We do not import Main Character's other product integrations or data into Roboty because it has the same operator.

We may retain or disclose information when required by law or reasonably necessary to protect rights and security. We do not sell personal information or use it for cross-context behavioral advertising. Retention depends on the purpose and the actual selected service. We do not promise immediate removal from every backup, third-party copy or already disclosed report. We must still comply with applicable deletion obligations.

You can manage local data through the app/browser controls and submit requests through Contact Roboty. We may need limited information to verify a request concerning material we actually hold. Never send a pairing credential or secret management capability in an ordinary support message.

Children and notice changes

Public educational reading is available to learners of different ages. Interactive software, beta applications and submissions are for people 13 and older. Minors need parent/guardian agreement and supervision. We do not knowingly collect personal information from children under 13. If you believe a child has submitted it, use the contact page so we can take appropriate action. Educational use does not itself establish parental permission for collection or third-party transfer.

The date below shows when this notice was last updated. Material new processing needs its appropriate notice and permission before it happens. A new notice does not silently broaden old research/training choices. See Terms of Use and Contact Roboty.

Last updated: October 26, 2026.