Privacy Policy

Effective: July 19, 2026.

This Privacy Policy ("Policy") describes how CADABRA AI Inc. ("CADABRA," "we," "us," or "our") collects, uses, discloses, and protects information when you:

• visit our website at cadabrai.com, including the marketing and pricing pages (the "Website");
• use the CADABRA dashboard at the /dashboard/ path, including sign-in and the post-payment welcome flow (the "Dashboard");
• install or use the CADABRA AI CAD copilot, including our SolidWorks add-in and related desktop software (the "Software"); and
• contact us through any of our forms, community channels, or email addresses.

The Website, Dashboard, and Software are referred to together as the "Services." By using the Services, you agree to this Policy. If you reach CADABRA through an Enterprise account that your employer or institution set up, that organization decides how your data is used inside its account, and we process data on its behalf under our agreement with it.

1. Information we collect

1.1 Information you give us directly.

• Account information. When you create an account, your credentials are managed by our identity provider (Microsoft Entra External ID, and, during our platform migration, Amazon Cognito). Your password is stored and verified by that provider. We do not store it, and we never store it in readable form. During the migration window our sign-in service does receive your password in transit in order to pass it to the identity provider and, on your first sign-in after the migration, to create your Entra credential; it is held only for the moment needed to complete that exchange and is never written to our databases or logs. In our records we keep your name, email, phone number if you provide one, plan, your Stripe customer/subscription details, your organization membership and administrator status, your preferences, your data-retention preference, and the version of the Software you last used.
• Demo and contact forms. When you submit a "Get a demo" or contact form, we collect the name, email, company name, and any message you provide, plus the time and page of submission.
• Newsletter signup. If you join our newsletter, we collect your email address.
• Pricing and student signup. When you start a paid plan, we collect your name, email, and (for Student plans) your school or university.
• Support and community messages. If you email team@cadabrai.com or reach us through our community channels, we receive whatever you send us.

1.2 CAD and design data you submit through the Software.

When you use the AI CAD copilot, the Software sends the content needed to answer your request to our hosted backend. The categories below describe what can be transmitted; the exact set depends on the request you make and the features you use.

• Your prompts and chat messages.
• CAD content read from your active model: feature-tree structure; feature, face, edge, and vertex identifiers; geometry, dimensions, dimension tolerances, and mass properties; sketch geometry; equations and global variables; document units; materials and appearances; configuration names, per-configuration suppression states, and dimension overrides; custom and configuration-specific file properties (which commonly carry part numbers, revisions, authors, project names, and descriptions); component and assembly structure and mates; sheet-metal parameters including K-factor and bend allowance; rebuild-error and feature-error text; and your current selection.
• Drawing content, when you work on a drawing: sheet names and scales, view names, the text of notes and annotations, title-block field names and their values, bill-of-materials table rows, and dimension text.
• Document and file identifiers: the name and the full local path of the active document and of referenced components and drawings, and the titles of other documents you have open. The full path is part of the context sent on each turn, not only of occasional diagnostic records, and a local path can contain your folder names and your Windows user name. See §4 for how paths are handled in telemetry.
• When you ask the copilot to find or open a document, the folders SolidWorks lists as recently used are enumerated to locate it, and the candidate file paths considered are included in the request.
• Screenshots captured by the copilot to understand your model visually. These are produced by SolidWorks' own image export and contain only the model view. The Software does not read your screen: it captures no part of your desktop, no other application's window, and no part of the CADABRA panel. If the export fails, no screenshot is taken and the copilot continues without one.
• Files you attach to a chat. The Software accepts images (PNG, JPEG, GIF, WebP) and PDFs up to 20 MB. Files above roughly 5 MB for an image or 4.5 MB for a PDF are still sent to the model to answer your request, but are not saved to your chat history. It also accepts SolidWorks drawing files (.slddrw, .drwdot); these are converted on your machine to an image of the drawing, and only that image is transmitted; the drawing file itself is not uploaded.
• Drawing Standards reference files. If your organization uses the Drawings Reference Learning feature, the reference files you submit are uploaded and stored so the learned standard can be reproduced and audited. This feature accepts raw SolidWorks files, including part (.sldprt), assembly (.sldasm), and drawing (.slddrw, .drwdot, .slddrt) files, as well as .dwg, .dxf, .step/.stp, .pdf, and .png, up to 25 MB each. This is the one feature in which a complete CAD file, rather than data derived from it, leaves your machine, and it operates only on your instruction. See §3.

We call the prompts, CAD content, attachments, reference files, and screenshots you submit, and the outputs we generate, your "Content." How we process, retain, and (unless you opt out) train on Content is described in §3.

1.3 Information we collect automatically.

• Technical / web-log data. When you load the Website or Dashboard, our servers receive your IP address (often shortened), browser type and version, operating system, time zone, the page you came from, and the pages you visit. We use this for operating the Services, security, and basic analytics.
• Software operational data. The Software transmits operational telemetry, logs, and diagnostics so we can keep it working and fix problems (see §4).
• Local data on your device. The Software stores certain data on your own computer, such as encrypted sign-in tokens, local caches, and working copies of unsaved documents (see §5).

1.4 Information from third parties.

• Stripe. When you complete a checkout, Stripe sends us a confirmation from which we record your email, the plan and price you selected, your billing interval, your subscription identifier and status, the current billing period, the number of seats, and whether the subscription is set to cancel. Card numbers stay with Stripe and we never receive them; if you view your payment method in the Dashboard, we retrieve the card brand, last four digits, and expiry date from Stripe to display to you.
• Cloud infrastructure. Our hosting providers may pass us standard request metadata (like your IP address and request headers) so the Services can respond to your device.

2. How we use information

We use information to:

• provide and operate the Services, including signing you in, remembering your plan and preferences, and generating responses from the AI CAD copilot to your prompts;
• process payments through Stripe and prevent fraud;
• reply to your messages, demo requests, and support tickets;
• send you transactional communications (receipts, account and security notices) and, only if you opt in, newsletters and product updates, which you can unsubscribe from at any time;
• keep the Services secure: detect abuse, debug errors, rate-limit traffic, and investigate suspicious activity;
• understand how the Services are used so we can improve them;
• improve and, subject to the opt-out in §3, train our artificial-intelligence and machine-learning models; and
• comply with legal obligations (such as tax recordkeeping) and enforce our Terms of Service.

2.1 Our legal bases, if you are in the EU, the UK, or Switzerland

Data-protection law in those regions requires a specific legal basis for each thing we do with your personal information. Ours are:

• Running the Services for you: signing you in, remembering your plan and preferences, passing your prompts to the copilot, storing your chats so history works, taking payment: performance of our contract with you (Article 6(1)(b)).
• Keeping the Services working and secure: operational telemetry, crash and update diagnostics, error investigation, rate limiting, abuse detection: our legitimate interests (Article 6(1)(f)) in running software that is reliable and not abused. We balance that against your interest in keeping your design data to yourself by limiting what telemetry carries (§4) and by giving you a switch that removes model details from it entirely.
• Improving the Services and training our AI models on your Content, where you have not opted out: our legitimate interests (Article 6(1)(f)) in improving a product our customers depend on. This is the most intrusive processing described in this Policy, which is why it is the one you can switch off at any time. The setting in §3, or an objection under §10, both stop it. If you want to see the balancing assessment we carried out before relying on this basis, email us and we will send it.
• Newsletters and product-update emails: your consent (Article 6(1)(a)), which you can withdraw at any time without affecting anything you received before.
• Tax records, accounting, and responding to lawful requests from authorities: compliance with a legal obligation (Article 6(1)(c)).

We do not set out to collect special-category data (Article 9), and nothing in the Services asks for it. Engineering content is commercial information rather than personal information most of the time, but it is not always. If your designs, file paths, or prompts contain someone's personal details, they are covered by this Policy too.

3. Your CAD designs, AI processing, retention & training

Your CAD designs and engineering data are among your most sensitive information, and we treat them that way.

Processing to answer your requests. To generate an output, your Content is transmitted over an encrypted connection to our hosted backend and passed to the AI model provider that serves your request (see §6). Our agreements with the model providers we use prohibit them from using your Content to train their own models and require them to delete it after processing. Those are contractual commitments from the provider, not technical guarantees we can enforce from our side.

These things are still collected or kept even when you opt out:

• Operational diagnostics: crash, installation, and update logs, and your device's machine name, so we can keep the Software working and secure (see §4).
• Feedback you explicitly submit. If you rate or report a response, we retain the report and, where your version of the Software includes them, that turn's prompt and output, because that is the substance of what you sent us.
• Drawing Standards reference files and learned standards, which the product exists to store for you.
• Usage records used for billing, quota, and abuse prevention. These identify you and record the model used, the mode, token counts, and the identifier of the chat, but not the content of your prompts or outputs.
• De-identified, aggregated metrics that no longer identify you or your designs.

At-rest protection of your stored Content. Where your Content is retained, it is encrypted at rest by our cloud provider's storage-level encryption, which is applied to all of our databases and object storage and cannot be disabled. Under this protection, authorized CADABRA personnel can, under least-privilege access controls, reach stored Content in order to operate and support the Services.

To be precise about what that control is and is not: access is limited to the people who need it and is granted through our cloud provider's permission system, for a narrow set of reasons: investigating a failure you reported, keeping the Services running, or responding to a security or legal obligation. They do not access it for any other reason. We do not use your designs, drawings, or prompts in demonstrations, sales material, screenshots, conference talks, case studies, or public benchmarks. If we ever want to show your work, we will ask you first and you can say no.

We do not today keep a per-record log of which staff member read which customer's Content, so we cannot produce a per-record access history on request, and we will not describe our access controls as audited until we can. Our internal console shows account details, plan and subscription state, plugin version and devices, and token totals. It cannot open your account, read your chats, or open your files, and there is no impersonation path. If your organization requires access logging as a condition of deployment, raise it with us and we will tell you where it stands rather than imply it already exists.

We have additionally built an application-layer encryption mode, in which stored chats, attachments, and feedback text are encrypted with AES-256-GCM before being written to storage, each account's data key is wrapped by a separate key-encryption key held in a managed key vault, and decryption requires a distinct key-access permission that storage access alone does not grant. This mode is not enabled in our production environment today and is therefore not in effect for any account. We will update this Policy before it is switched on. If your organization requires application-layer encryption, customer-managed keys, or a deployment in which no data leaves your network, tell us what you need and we will tell you honestly where that stands. We would rather scope it with you than imply it is a setting we can switch on today.

If we switch that mode on, it will be a strong administrative and technical safeguard, not a representation that we are technically incapable of decryption. We would still operate the key vault. Encryption at rest, of either kind, also does not change the fact that to answer your request your Content is processed in plaintext by the AI model provider. No at-rest measure can remove that step.

Regardless of the setting above, the third-party model providers we use are contractually barred from training on your Content.

4. Telemetry, logs & diagnostics

To keep the Software reliable, the desktop Software sends operational signals to our backend rather than keeping them only on your machine:

• Application and event logs: what the Software did, including error messages and, where an error occurred, the underlying exception text. Before upload, absolute file paths in these logs, including any inside exception text and stack traces, are replaced with a truncated SHA-256 token, and your Windows user name is removed, so your folder names and your account name are not transmitted. As with the tool-audit records below, that token is a stable pseudonym rather than an anonymization. The text of the request you type in Agent mode, and of any web-search or feature-search query the copilot runs, is reduced to a length indicator rather than sent.
• Tool-audit records: what the copilot did, such as tool names and parameters (dimensions, feature names, component names) and success/error results. Before upload, values in these records that are recognizable as local file paths are replaced with a truncated SHA-256 token so that folder names and your Windows user name are not carried in them. That token is a pseudonym, not an anonymization: it is stable, and someone who could already guess a specific path could confirm it by hashing that guess. Path-shaped values nested inside lists, and document names that carry no file extension, are not currently recognized by this replacement.
• Drawing-operation records: the stages of a drawing run. Model and drawing paths in them are reduced to the file name; the file name itself is transmitted.
• Macro-execution records: when the copilot generates and runs a macro, the record includes the macro name, timing, and any error messages. The text of your prompt is reduced to a length indicator in these records, and your Windows user name is not a field in them.
• Model-quality traces: to improve the copilot and diagnose failures, we record the text of your prompt, the model's output, the model's own reasoning about which tool to use, and the input and output of each CAD operation the copilot ran during the turn. Those operation records carry the parameters the copilot worked with: dimensions, coordinates, feature names, and component names. These traces go to our observability provider, Langfuse (see §6), which is hosted in the United States and outside our own cloud accounts. It is the one place your Content is processed by a third party for a purpose other than generating your answer. By default, this text is not masked. If you have turned on "Don't use my data to train AI" (§3), or your organization is on a zero-retention arrangement, the prompt, output, reasoning, plan, and per-operation text are all replaced with a placeholder before the trace leaves our backend, and only non-content metadata (token counts, latency, model identifier, and whether the step succeeded) is sent.
• Diagnostics: crash reports, and installation and update logs, together with the Software version and a one-way hash of your device's machine name (we use the hash to group one device's reports together; we do not transmit the machine name itself). The contents of these log files are uploaded (up to the last 256 KB of each) exactly as written, without redaction, and can contain file paths and your Windows user name. We do not collect your hardware identifiers, network addresses, Windows security identifier, or your SolidWorks serial or license information.

Turning off model details in diagnostics. In the CADABRA panel, under Settings → Account → Data & privacy, you can switch off "Include model details in diagnostics." With it off, records that would describe your model (prompt text, tool arguments such as dimensions and feature names, and CAD file names) are not transmitted at all. We still receive that an operation ran, its type, its severity, and when, so we can see failure rates; we do not receive anything describing your design. The setting is stored on your computer and takes effect immediately, including for the current session. It is on by default, and every feature works identically with it off.

If you have opted out / are on a zero-retention arrangement (§3), telemetry and traces that contain your Content are dropped rather than stored, and model-quality traces keep only non-content metadata (tokens, latency, model id). Diagnostics are retained regardless, solely for operational and security purposes, and are not used to train models. User-submitted feedback is stored only when you explicitly send it.

5. Cookies & local storage

We keep cookies and trackers to a minimum. We do not sell ad space, and you will never see third-party advertising inside CADABRA.

We do advertise CADABRA on Google. The Google Ads tag runs on our Website so we can see which ads brought you here. It reports your visit to Google and sets Google cookies, which Google may use for conversion measurement and to show you CADABRA ads elsewhere. It does not run in the Dashboard or in the desktop Software. You can turn off personalised ads at Google Ads Settings, and your browser settings let you block or clear these cookies.

We do not use mobile advertising identifiers, and we do not run any session-recording product on the Website or Dashboard.

• Website / Dashboard. The Dashboard uses a small amount of essential browser local storage to keep you signed in and to carry checkout state across the Stripe redirect (for example, your session tokens and a cached copy of your profile). This lives in your browser and, briefly during signup, also includes the email and password you entered, until sign-in completes. Signing out clears your session tokens and cached profile; other values, such as a plan you selected before checkout and interface preferences, remain until your browser clears them or you clear them yourself. Stripe sets cookies on its own checkout domain, and Stripe's payment script also runs on our dashboard and can set cookies there. Our pages load the Inter typeface from Google Fonts, so Google receives your IP address and browser user-agent when a page loads. Embedded third-party widgets (such as a scheduling widget or a linked video) may set their own cookies under their own domains.
• Desktop Software. On your computer, the Software stores your sign-in tokens encrypted with the Windows Data Protection API (DPAPI) under your user account; a local cache of per-document known-entity information, keyed by a hash of the file path, to speed up the copilot (expires after 30 days); a queue of pending telemetry awaiting upload; local operation-audit and build-transcript report files under your temporary folder; macros the copilot generates, written under your user profile and removed after use; and temporary screenshots in your temporary folder. When you run a drawing operation against a document you have not saved, the Software saves a working copy of that part or assembly under your local application-data folder so the operation has a file to reference; these working copies stay on your machine and are not uploaded. Because the CADABRA panel is a web view, Windows also keeps a browser profile for it under your local application-data folder, containing the panel's cache and local storage. The Software additionally keeps crash, installation, and update logs, downloaded update installers, and a short-lived cache of the current plan (including prompt text, for up to 48 hours). Configuration endpoints are read from the Windows registry under HKLM\SOFTWARE\CADABRA; the installer, not the Software, sets a CADABRA preference value on some install channels.

Your browser and operating-system settings let you block or clear cookies and local storage; doing so may prevent parts of the Services from working. If we add further analytics or marketing cookies, we'll update this section and, where required, obtain your consent first. Our Website does not currently respond to "Do Not Track" signals.

6. Who we share information with

We do not sell your personal information or your Content, and we do not share it for cross-context behavioral advertising. We share information only with:

• Payment processing: Stripe, to process payments and refunds;
• Cloud infrastructure: our hosting and storage providers (Microsoft Azure, and Amazon Web Services during our platform migration) that run the Services and store our data;
• AI model providers: the providers whose APIs generate outputs for you. We are part-way through a migration between two clouds and both are in service, so which one answers your request depends on the deployment you are connected to: Amazon Web Services (Amazon Bedrock), which serves the Anthropic Claude models, and Microsoft Azure AI Foundry, which serves both the Anthropic Claude and the OpenAI GPT models we offer. We do not silently re-route a failed request to an undisclosed vendor. Two narrower paths are the exception, and we name them rather than fold them into that sentence. If our primary provider is rate-limited or unavailable, your request may be completed through Cerebras, which receives the same request content as a normal turn. If the primary provider cannot process an image, the screenshot of your model, never your screen, may be sent to OpenAI. If you have turned on "Don't use my data to train AI" (§3), or are on a zero-retention arrangement, both fallbacks are refused before any of your Content is sent, and the request simply fails instead. We do not route your Content to any model vendor beyond these four. We work only with providers that contractually agree not to retain or train on your Content and to delete it after processing;
• Web search: when the copilot performs a web search, the query is sent to our web-search provider, either Perplexity AI or the web-search service of our cloud provider Microsoft Azure, depending on the deployment serving your request. We do not attach your CAD content or conversation history to that request; note that the query itself is composed by the model and can therefore mention a part name or dimension you asked about, and that web search is refused outright for opted-out and zero-retention accounts;
• Fallback model providers: OpenAI, which serves a fallback path for image understanding and certain auxiliary requests, and Cerebras, which serves certain auxiliary requests when our primary provider is unavailable. Both receive request content. Both are blocked outright for opted-out and zero-retention accounts;
• Observability: Langfuse, a US-hosted service, which stores the quality traces described in §4. Trace metadata includes your account identifier always, and your name and email address where retention is enabled. Unless you have opted out under §3, Langfuse also receives your prompt text, the model's output and reasoning, and the parameters of each CAD operation in the turn, and our staff can read those traces in Langfuse's interface. We also use Microsoft Azure Monitor / Application Insights, which receives request and performance traces of our own services rather than your Content;
• Identity: our authentication provider (Microsoft Entra External ID, and Amazon Cognito during migration), which stores and verifies your password;
• Email, forms, and scheduling: Google, which hosts our email, so anything you send to team@cadabrai.com is stored in Google Workspace, and which also hosts the spreadsheet that receives demo and contact form submissions, provides the fonts our pages load, and provides the scheduling calendar you may use to book a demo (which receives the name and email you enter); Amazon Web Services (SES), which delivers account verification and password-reset messages; and Cloudflare and Resend, which deliver our contact-form, newsletter, and welcome email;
• Enterprise administrators, if your account is part of an organization, who can see, for each member of their workspace, the member's name, email address, account identifier, administrator status, plan, subscription status, installed Software version, and individual usage and cost, including token counts, request counts, and dollar spend, over any date range they choose. Administrators cannot read your chats, prompts, or outputs;
• Other members of your organization: where your organization uses the Drawings Reference Learning feature, the drawing standard is shared across the organization. Every member, not only administrators, can see the names of the reference files uploaded to it and which member uploaded each one, and can download the active drawing template. Do not upload a reference file you are not willing to share with your colleagues;
• Law enforcement or other authorities, when we believe we are legally required to (for example, a valid subpoena), or to protect the rights, property, or safety of CADABRA, our users, or the public; and
• a successor entity in connection with a financing, merger, acquisition, or sale of assets, in which case we'll require the recipient to honor this Policy or notify you of any change.

We may also share information that has been aggregated or de-identified so that it can no longer identify you; we will not attempt to re-identify it except as permitted by law.

7. How long we keep information

• Account data: for as long as your account exists. If you ask us to close your account, we delete your account record and your identity with our authentication provider.
• Your Content (prompts, attachments, outputs), for retention-enabled accounts: until you delete it or ask us to delete it. Deleting a chat removes that chat and its attachments. Closing your account does not by itself delete Content already stored in the copilot backend. If you want that erased, ask us before or at the same time as closing the account, and see §10. For opted-out / zero-retention accounts, Content is not retained (see §3).
• Drawing Standards reference files and learned standards: retained until you remove them; there is no automatic expiry. Removing a file from a standard, or deleting the standard, erases the stored copy (see §3).
• Semantic-search index entries derived from your Content: the Software does not currently build or maintain a remote semantic index of your CAD content. That capability is present in the codebase but disabled, and nothing derived from your models is indexed into a searchable store. If we enable it, we will update this Policy first.
• Billing records and invoices: up to 7 years, as tax and accounting laws require.
• Usage records used for billing, quota, and abuse prevention: retained for as long as we need them for billing and audit purposes.
• Operational telemetry, logs, and diagnostics: retained for operational and security purposes until we delete them. No automatic expiry is in force today. A second copy of these records lands in our cloud provider's logging service and is kept on the same basis, which means general access to those logs is a way to read them. We are working toward an automatic deletion rule and we will say so here once it is actually in force, not while it is planned.
• Backups: our databases are backed up continuously. Deleted records can persist in those backups for the length of the restore window before they age out.
• Marketing subscriptions: until you unsubscribe.
• De-identified / aggregated data: kept indefinitely, because it no longer identifies you.

8. Data security

We use commercially reasonable physical, technical, and administrative safeguards, including TLS in transit; encryption at rest across our databases and object storage; DPAPI encryption of sign-in tokens stored on your device; managed authentication through our identity provider (Microsoft Entra External ID or, during migration, Amazon Cognito), which stores and verifies your password so that we do not have to store it; separation of secret-management permissions from storage permissions; and least-privilege access for our staff. No system is perfectly secure, so we cannot guarantee absolute security. If you discover a vulnerability or believe your account has been compromised, email team@cadabrai.com.

If we determine that a security breach has compromised your personal information, we will notify you and any regulator that applicable law requires us to notify, within the time that law requires, and we will tell you what we know about what happened and what you can do about it.

9. Where we process information

CADABRA AI Inc. operates from the United States. Our infrastructure (our servers, databases, and object storage) is located in the United States, and our identity, payment, and observability providers operate in the United States.

One exception is worth stating plainly: our AI model deployments use a global routing tier, which means the inference request that generates your output may be served by our model provider's capacity in a data center outside the United States. Your Content is processed in transit for that request and, under our agreements with the provider, is not retained by it afterward.

If you access the Services from outside the United States, your information will be transferred to and processed in the United States, where data-protection laws may differ from those in your country.

10. Your rights & choices

Subject to your local law, you may have the right to:

• access the personal information we hold about you and obtain a copy (including a portable copy);
• correct information that is inaccurate or incomplete;
• delete your account and the information tied to it;
• opt out of the use of your Content to improve or train our models (see §3);
• opt out of marketing communications;
• limit the use of sensitive personal information;
• object to or restrict certain processing, including direct marketing and certain automated decision-making;
• withdraw consent you previously gave (without affecting prior processing); and
• confirm whether we process personal information about you, and obtain the categories of third parties we disclosed it to.

To exercise any of these rights, email team@cadabrai.com from the address tied to your account. We may need to verify your identity before acting, and we'll respond within the time required by applicable law (and within 30 days where no specific deadline applies). You may appeal a decision by emailing us with the subject line "Appeal of Decision Related to Privacy Rights Request," and you may use an authorized agent supported by a signed authorization or power of attorney. Exercising any of these rights is free, and we will not degrade the Services or charge you differently because you did.

If you are not satisfied with how we handled it. Come back to us first, at team@cadabrai.com, and we will try to put it right. You do not have to. If you are in the European Economic Area, the United Kingdom, or Switzerland, you have the right to lodge a complaint directly with your data-protection supervisory authority: in the EEA, the authority in the country where you live, work, or where you believe the problem occurred (the list is published by the European Data Protection Board at edpb.europa.eu); in the UK, the Information Commissioner's Office at ico.org.uk; in Switzerland, the Federal Data Protection and Information Commissioner. If you are in California, you may complain to the California Privacy Protection Agency or the Attorney General. Contacting a regulator does not require you to have contacted us first.

Deleting your Content. You can delete an individual chat in the CADABRA panel, which removes that chat and its attachments. For anything broader, such as erasing all of your stored Content and your uploaded reference files, email team@cadabrai.com and we will carry it out and confirm when it is done. Three limits are worth knowing before you ask. First, we may keep what we are required or permitted to keep: billing and tax records, records we need to detect and prevent fraud, abuse, or security incidents, and anything a legal obligation makes us hold. If we rely on one of those, we will tell you which. Second, we cannot yet pull one account's records out of our operational telemetry, logs, and diagnostics. We can stop collecting them for you, but per-user erasure of that data is not something we can honestly promise today; if that matters to you, switch off "Include model details in diagnostics" (§4) so nothing describing your designs is collected in the first place. Third, deleted records can persist in backups for the restore window described in §7. Please make deletion requests before closing your account: closing it removes the identity we use to locate your data.

California residents. The California Consumer Privacy Act ("CCPA"), as amended by the CPRA, gives you the rights above plus the right to know the categories and specific pieces of personal information we collect and the right to opt out of the "sale" or "sharing" of personal information. We do not sell or share personal information as those terms are defined under the CCPA, and we will not discriminate against you for exercising your rights.

11. Children

The Services are not directed to children under 16, and we do not knowingly collect personal information from anyone under 16. If you believe a child has provided us personal information, contact team@cadabrai.com and we'll delete it.

12. Changes to this Policy

We may update this Policy as the Services evolve. The "Effective" date at the top reflects the most recent revision. If we make a material change, we'll notify you by email or an in-app/in-product notice before the change takes effect.

13. Contact