Agent Modeler Privacy Policy

1. Who we are and what this Policy covers

Agent Modeler is a browser-based software service provided by Marcozza Inc. (“Marcozza,” “we,” “us,” or “our”). This Privacy Policy explains how Marcozza collects, uses, communicates, retains, and protects personal information when you visit the Agent Modeler website, create or use an account, use its services, or contact us.

This Policy is intended to describe our practices under Québec's Act respecting the protection of personal information in the private sector and applicable Canadian privacy law, including the Personal Information Protection and Electronic Documents Act where it applies.

This Policy does not govern a third-party website or service operating under its own privacy notice. Section 6 explains the principal providers involved in Agent Modeler.

2. Person in charge of personal information

The person exercising the highest authority at Marcozza Inc. acts as the Person in charge of the protection of personal information.

Questions, privacy complaints, and requests to exercise privacy rights may be sent to:

Person in charge of the protection of personal information
Marcozza Inc.
support@marcozza.com

3. Personal information we handle

We limit collection to information reasonably necessary for the purposes described in this Policy.

3.1 Account and identity information

When you register or sign in, we may process:

• your email address, display name if supplied, preferred interface language, and account creation and update times;
• identifiers supplied by Microsoft Entra External ID, including the identity provider, issuer, tenant, and subject identifiers;
• internal user and workspace identifiers, account status, authentication events, and last authenticated activity.

The signup notice does not create a versioned Terms-acceptance receipt or a separate Privacy Policy acknowledgment record. Purchase records may include consent collected at checkout, its timestamp, and the related checkout reference. This is separate from registering or signing in to Agent Modeler.

Microsoft Entra External ID manages authentication. We do not receive or store your identity-provider password or one-time sign-in code.

3.2 Plans, subscriptions, and transactions

We process your Plan, capabilities, quotas, complimentary or trial grants, subscription status, billing period, cancellation status, and related account history to provide and control access to the Service.

Stripe Managed Payments, presented as Sold through Link, acts as merchant of record for purchases made through that service and collects payment and billing details at checkout. Stripe handles the transaction, applicable transaction taxes, billing support, and refunds. Marcozza provides Agent Modeler and may receive a Stripe customer identifier and limited transaction information such as the purchased Plan, price, currency, payment status, billing period, applicable tax information, and the last four digits or type of a payment method where Stripe makes them available. Marcozza does not store complete payment-card numbers or card security codes.

Stripe’s processing of purchase and payment information is described in the Link privacy policy. Marcozza remains responsible for its own handling of Agent Modeler account and Customer Content information under this Policy. Deleting an Agent Modeler account does not itself delete records separately held by Stripe or close a Link account.

3.3 Customer Content

Depending on your Plan and the features you choose, we may process the content you create, import, or save, including:

• projects, models, diagrams, requirements, tools, skills, libraries, and version history;
• prompts, instructions, AI conversations, specifications, test information, trace-analysis results, and other generated or imported material; and
• messages and attachments you deliberately send to support.

Customer Content may contain personal information about you or another person. You are responsible for having authority to provide that information and for avoiding unnecessary personal information, confidential information, and secrets. Agent Modeler is not designed to collect sensitive personal information such as health or biometric information, government identifiers, financial-account information, precise geolocation, criminal history, or information about children. Do not place such information in Customer Content, prompts, traces, or support messages.

Cloud Customer Content is encrypted at rest. Agent Modeler's backend can decrypt content when necessary to provide an authorized feature; cloud storage is therefore not zero-knowledge storage.

3.4 Usage, security, and diagnostic information

We process information needed to authenticate users, operate the Service, enforce tenancy and Plan limits, prevent abuse, diagnose failures, and maintain security. This may include login and authentication records, timestamps, request identifiers, API route and method, response status, duration, error type, storage operations, and entitlement or quota usage.

Our application diagnostic logs are designed not to contain request or response bodies, model content, prompts, provider responses, API credentials, authorization values, cookies, or browser history. They are also designed not to record IP addresses. Infrastructure and identity providers, including Microsoft Azure and Microsoft Entra, may nevertheless process IP address, browser, and device information for network delivery, authentication, fraud prevention, and security under their own service terms.

3.5 Information stored only in your browser

Agent Modeler uses browser storage, including local storage, session storage, and IndexedDB, for functional purposes. Depending on the features you use, this may include interface and language preferences, local projects and libraries, chat history and settings, temporary authentication state, and AI-provider credentials.

An AI-provider credential is kept in session storage by default. If you explicitly choose to remember it, Agent Modeler stores an encrypted copy in local storage on that browser and device. Your encryption password is not stored and cannot be recovered by Marcozza. A usable unlocked credential is held in browser memory. Your encryption password and derived encryption keys are not sent to Agent Modeler's servers. The Test AI and Findings AI review relays are currently disconnected and do not accept new requests.

Browser-only information remains under the controls of your browser and device. You can remove it using Agent Modeler's available controls or your browser's site-data settings. Clearing it may permanently remove unsaved local work or a remembered provider connection.

3.6 Cookies and similar technologies

Agent Modeler uses only storage or similar technologies needed for the application, authentication, security, and user preferences. Microsoft Entra's authentication pages may use their own necessary cookies. We do not use Google Analytics, advertising cookies, tracking pixels, behavioural advertising, remarketing, or cross-site profiling.

Blocking necessary browser storage or identity-provider cookies may prevent sign-in or parts of the Service from functioning. Because we do not currently use optional analytics or advertising cookies, Agent Modeler does not presently offer a separate advertising-cookie consent control.

MCP authorization also uses essential cookies on Agent Modeler's connection service to protect the sign-in and consent exchange. Connection credentials and permission records are held by the server; they are distinct from the AI-provider keys you configure in your browser.

3.7 How we collect information

We collect information directly from you when you register, configure or use a feature, save Customer Content, make a purchase, or contact us. We receive limited identity information from Microsoft Entra External ID and, when applicable, transaction information from Stripe. The application and its infrastructure generate the limited usage and security records described above as the Service operates. Built-in AI chat sends applicable content directly from your browser when you initiate an AI operation. Separately, an external assistant you authorize through MCP can request permitted workspace information from our servers as described in section 5.2.

4. Why we use personal information

We use personal information to:

• create, authenticate, secure, administer, and support accounts and workspaces;
• provide the features you request, including cloud storage, project and model management, validation, exports, and optional AI connections;
• assign and enforce Plans, entitlements, quotas, trials, and promotions;
• process subscriptions, payments, cancellations, and account deletion when those functions apply;
• communicate necessary account, security, billing, support, service, and policy notices;
• diagnose errors and improve the reliability, accessibility, and security of the Service using limited operational information;
• prevent, investigate, and respond to fraud, misuse, security incidents, and violations of our Terms; and
• comply with legal obligations, respond to lawful requests, resolve disputes, and protect legal rights.

We do not sell personal information. We do not use Customer Content for advertising or to train a general-purpose AI model. We do not send promotional marketing unless you separately choose to receive it, and you may withdraw that choice without losing necessary service communications.

We obtain consent where required and may also handle information where the law otherwise permits or requires it. Withdrawing consent may prevent us from providing a feature that necessarily depends on the information, but does not affect handling already permitted by law.

5. AI providers

5.1 Built-in AI chat

Optional AI features let you connect an enabled provider such as OpenRouter, OpenAI, or Anthropic. Connecting a provider alone does not send your model. Design chat, including Review design spec, sends the relevant design context, instructions, applicable discussion and any attached specification draft directly from your browser to your selected provider.

Test chat and experiment generation also connect directly from your browser to your selected provider. They send public authoring guidance, your diagram, relevant experiment definitions, pending drafts and the applicable Test discussion. Recorded execution evidence and private analysis instructions are not attached automatically.

Agent Modeler's server checks the request and proposed definitions for format, references and duplicates without calling an AI provider or receiving your provider key. The request context and returned proposals are processed without saving their content in this checking service. Saving your workspace and optional conversations is a separate action governed by your saving settings.

Findings AI review remains unavailable. The old Test AI and Findings review relays remain disconnected, including for previously saved permissions. Manual test design, Test Lab and deterministic validation remain available without an external AI model.

Before this change, enabled relays forwarded request content and provider credentials through our server to the selected provider. Disabling the relays does not remove information previously sent to a provider. Saved discussions and designs continue to follow the storage and retention provisions of this Policy.

The selected provider processes this information under its own terms and privacy notice and may process it outside Québec or Canada. Review the provider's practices and avoid sending information you are not authorized to disclose. You are also responsible for managing and revoking your provider credential and for charges imposed by that provider.

5.2 External assistants and MCP connections

You may connect a supported external assistant using the Model Context Protocol (MCP). After you authorize a connection, that assistant can request permitted content from Agent Modeler's servers within your workspace and Plan. Workspace access may include models created later. Requested content can include model definitions, requirements, tool and skill libraries, specifications, and, where your Plan and separate permissions allow, test definitions and execution evidence. This transfer is from our server to the external assistant and its provider; it does not require the Agent Modeler browser tab to remain open.

Read access does not grant editing authority. Authoring, Testing, and deletion access are controlled separately by the connection's granted permissions, applicable approvals, and server-side checks. Permitted operations can create, change, or remove saved content. A connection does not grant access to your billing details, browser-stored AI-provider keys, or saved Design/Test chat history. Review the consent screen, connection settings, and your assistant's tool approvals before allowing it to act.

We retain connection identifiers, client identity, granted permissions, authorization and expiry times, protected connection credentials, and revocation or audit records needed to operate and secure the connection. These are authentication records, not an automatic copy of the conversation you hold with the external assistant.

You can disconnect an assistant in Agent Modeler's MCP integration settings. Disconnecting revokes that connection's further access; requests already admitted may finish. It does not delete copies already received by the assistant or its provider. That provider's own privacy, retention, training, processing-location, and deletion terms apply to those copies. Review those terms and send only information you are authorized to share.

6. Service providers, access, and transfers

We communicate personal information only as reasonably necessary for the purposes described in this Policy, including to:

• Microsoft Azure, for application hosting, databases, encrypted object storage, key management, monitoring, and related infrastructure;
• Microsoft Entra External ID, for registration and authentication;
• Stripe / Link, for Stripe Managed Payments purchases, merchant-of-record payment, subscription, tax, fraud-prevention, billing support, and refund services, under its own privacy notice; and
• OpenRouter, OpenAI, or Anthropic, when you select that provider and initiate a built-in AI operation; and
• the external assistant and its provider that you authorize through MCP, for the permitted content and operations described in section 5.2.

Providers may use their own subprocessors. Their processing locations and retention practices may differ from ours. Information may therefore be processed outside Québec or Canada and may be subject to the laws and lawful access procedures of another jurisdiction. Where applicable, Marcozza assesses privacy factors before entrusting personal information outside Québec and uses contractual and other safeguards required by law.

Within Marcozza, access is limited to the operator and systems that require it for support, security, account administration, or legal compliance. Customer Content is not made available to other customers. Authentication alone does not authorize access to another workspace.

We may also disclose information when required or permitted by law, to respond to valid legal process, to protect safety or legal rights, or as part of a proposed financing, merger, reorganization, acquisition, or sale of business assets. In a business transaction, information remains subject to applicable confidentiality and privacy requirements.

7. Retention and deletion

We retain personal information only for as long as reasonably necessary for the purposes described above and any applicable legal requirements.

• Account, workspace, Plan, and cloud-content records are generally kept while the account remains active or while needed to provide the Service.
• Application operational logs are generally retained for 30 days. Security or incident records may be retained longer when reasonably necessary to investigate an event or comply with law.
• Active database backups are retained for up to seven days under the current production configuration and then expire through the provider's normal backup lifecycle.
• Browser-only information remains until you or your browser removes it. Marcozza cannot delete browser storage on a device it does not control.
• AI providers and Stripe retain information according to their own notices and legal obligations. Deleting Agent Modeler data does not automatically delete records independently held in your provider accounts.

Cancelling a paid subscription stops renewal but does not delete the account immediately. Paid access continues until the end confirmed by the payment provider. When paid or trial access ends, we retain the account and Customer Content in a read-only archive for 90 days so that the customer may review and export eligible existing content or resubscribe. Editing, paid analysis, and AI-provider connections are disabled during that period.

We send necessary lifecycle notices when the archive begins, 60 days before scheduled deletion, and 7 days before scheduled deletion. If the customer resubscribes before the deadline, we cancel the subscription-ended deletion. Otherwise, when the 90-day period expires, we delete the account profile, identity mapping, workspace metadata, Plans and entitlements, projects, models, libraries, versions, and other customer-owned cloud records from active systems. This schedule also applies when the payment provider makes a subscription terminal after its payment-recovery process.

An explicit account-deletion request is separate. When deletion is requested, access ends at the later of the requested closure time and any already-paid period. A 30-day recovery period begins when access ends. A scheduled deletion may be cancelled during that recovery period. This request does not extend a shorter subscription-archive deadline already communicated to the customer. Deleted encrypted objects may remain recoverable in restricted Azure soft-delete storage for up to seven additional days before permanent expiry. The separate Microsoft Entra customer identity is also deleted or its deletion is verified as part of completing the process.

After completion, we may retain a minimal deletion receipt containing only an internal closure identifier and lifecycle dates and status. It does not retain the customer's email address, user identifier, identity-provider subject, Customer Content, or cloud-object address. We may retain other limited records where required by tax, accounting, fraud-prevention, dispute, or other legal obligations. Information subject to a valid preservation requirement may be held until that requirement ends.

Until self-service deletion is made available in the account interface, send a deletion request to support@marcozza.com.

8. Security

We use administrative and technical safeguards appropriate to the nature of the information, including encrypted network transport, encryption of cloud content at rest, managed key storage, scoped service identities, authentication, server-side authorization, workspace isolation, access controls, and sanitized operational logging. We test account isolation and require server-side Plan and quota enforcement; browser controls are presentation only.

No system can guarantee absolute security. You are responsible for protecting your account, device, browser profile, provider credentials, and exported or locally saved files. Contact us promptly if you believe your account or personal information has been compromised.

If a confidentiality incident creates a risk of serious injury, we will notify affected individuals and the competent regulator as required by law.

9. Your privacy rights and complaints

Subject to applicable law, you may request:

• confirmation of whether we hold personal information about you and access to that information;
• information about how it has been used and communicated;
• correction of inaccurate or incomplete information;
• withdrawal of consent for an applicable use or communication;
• deletion of information that is no longer required, subject to lawful exceptions; and
• a copy of eligible computerized information collected from you in a structured, commonly used technological format.

Send a written request to the Person in charge at support@marcozza.com. We may take reasonable steps to verify your identity and authority before acting, but will not ask for more information than reasonably necessary. We aim to respond within the period required by applicable law; Québec access and correction requests generally require a response within 30 days.

You may also submit a privacy complaint to the same address. We will acknowledge it, investigate it proportionately, communicate the outcome, and take appropriate corrective action where warranted. If you are dissatisfied, you may contact the Commission d'accès à l'information du Québec or, where applicable, the Office of the Privacy Commissioner of Canada.

10. Children

Agent Modeler is intended only for individuals who are at least 18 years old. We do not knowingly collect personal information from anyone under 18. If you believe a minor has provided personal information, contact us so that we can investigate and take appropriate action.

11. Changes to this Policy

We may update this Policy when our practices, providers, features, or legal obligations change. We will publish the updated Policy and its revision date at https://agentmodeler.com/privacy. We will provide an appropriate notice of a material change before it takes effect where required.

The Privacy Policy is a notice about our handling of personal information. We may ask you to acknowledge that it was provided, but we do not treat acceptance of this Policy as blanket consent for unrelated purposes. Where separate consent is required, we will request it distinctly.