Privacy Policy

1. Scope and our role

This Privacy Policy ("Policy") applies to personal data processed by Cerebro Labs Inc., the company developing and operating AgentNava ("AgentNava", "we", "us") through the AgentNava website, console, API, and SDK (collectively, the "Service"). It is incorporated into our Terms of Service.

For most purposes we act as the data controller of personal data we collect about you as our customer. For Customer Content that you upload or transmit through the Service, we act as a data processor on your behalf, processing such data per your instructions as set out in the Terms of Service, this Policy, and our Data Processing Agreement.

If you are an end user of an agent built by one of our customers, that customer, not AgentNava, is the controller of any personal data you provide. Please contact that customer for their privacy practices.

2. Data we collect

Account data

When you sign up, our identity provider Clerk collects your email address, name, profile image (optional), and authentication credentials. We mirror your Clerk user ID, email, and name into our database so the console can display the correct workspace and route notifications.

Billing data

When you subscribe, Stripe collects your payment method, billing address, and tax identification (where applicable). We never see or store your full payment card number. Stripe shares with us a customer ID, last-four digits, brand, expiration, and the subscription lifecycle events necessary to operate billing.

Customer Content

You provide content when you push specs, configure agents, run chat conversations, and upload knowledge files. We store this content to operate the Service for you. We do not train models on identifiable Customer Content and we do not sell it.

Usage and operational data

We collect API request logs, conversation counts, latency metrics, error logs, IP addresses, browser/SDK versions, and similar operational telemetry. This data is necessary for billing accuracy, abuse detection, debugging, capacity planning, and improving the Service.

Communications

If you email us or open a support thread through our Intercom customer-support Messenger, we retain the content of those communications, your contact details, and any attachments. For signed-in customers, we securely provide Intercom with your AgentNava account ID, email, and name so our support team can verify who they are assisting. We do not automatically send prompts, agent or workspace content, secrets, or API keys to Intercom; only account identity and content you choose to submit in the support conversation are sent.

Cookies and similar technologies

We use essential first-party cookies set by Clerk for authentication and conversation continuity. We also use first-party Intercom Messenger cookies to maintain a customer-requested support conversation and prevent another user on a shared browser from inheriting it after logout. These functional technologies are necessary to provide the features you request and are always on when the Messenger is configured.

We use Mixpanel for privacy-friendly product analytics and masked conversation replay across our marketing site, the AgentNava app, and the developer console. We also use PostHog for product analytics on the marketing site and AgentNava app; PostHog conversation recording, heatmaps, performance capture, and automatic interaction capture are disabled. We record product events (for example, "signed up", "message sent", "agent activated", "API key created") tied to an account identifier so we can understand which features are used and diagnose problems. Mixpanel replays are fully masked: all text and input fields are hidden and images/videos are blocked, so message contents, names, emails, API keys, and other sensitive content are never captured in a replay. PostHog receives a stable account identifier but not your name or email, and neither service receives prompts, messages, secrets, or API keys in analytics event data. These technologies are never used for advertising, ad-targeting, or selling your data. Mixpanel replays auto-expire after 30 days.

Your consent choices depend on your region. If you are in the EU, EEA, UK, or Switzerland, analytics and conversation replay stay off until you accept the cookie notice (explicit opt-in). Everywhere else, they are on by default to improve the product, and you can opt out at any time from the notice we show or by contacting us; we honor your browser's Do Not Track and Global Privacy Control signals as well. If you decline or opt out, the product works exactly the same. See Your rights.

3. How we use data and lawful bases

We use personal data only as needed to operate the Service and run the business. Where the GDPR applies, our lawful bases are:

• Performance of a contract: to provide the Service you signed up for, including authentication, running your agents, returning model output, and billing.
• Legitimate interests: to secure and improve the Service (abuse detection, anti-fraud, latency telemetry, capacity planning), to communicate essential service notices, and to enforce our Terms. We balance these interests against your rights.
• Legal obligation: to comply with tax, accounting, anti-money-laundering, sanctions, and law-enforcement obligations.
• Consent (EU/EEA/UK/CH) or legitimate interests with an opt-out (elsewhere): for non-essential technologies such as product analytics (Mixpanel and PostHog), masked conversation replay (Mixpanel), and optional product emails. In opt-in regions these stay off until you accept; elsewhere they run by default and you may opt out at any time.

4. Sub-processors

To deliver the Service we share personal data with carefully selected sub-processors, bound by contract to confidentiality and equivalent data-protection standards:

• Our cloud-infrastructure provider: application hosting, edge delivery, and storage.
• Clerk, Inc.: user authentication, conversation and identity management.
• Stripe, Inc.: payment processing, subscription management, and tax determination.
• OpenRouter, Inc.: routing inference requests to large-language-model providers (Anthropic, etc.). Prompts and conversation history necessary to fulfill a chat request transit OpenRouter to the selected provider.
• TypeSafe AI: the Jev decision model, reached through OpenRouter, for guardrails and Reflect (see Your prompts and the models). The text being checked transits OpenRouter to TypeSafe AI, which returns a decision.
• Composio, Inc.: connections to other apps. It holds the access tokens for the apps a customer connects to an agent, such as Gmail or Slack, and makes the requests the agent's tools send to those apps.
• Mixpanel, Inc.: product analytics and masked conversation replay (see Cookies and similar technologies). Recording is fully masked (all text/inputs hidden, media blocked); never receives message contents, secrets, or API keys. In the EU/EEA/UK/CH it runs only after opt-in; elsewhere it is on by default with an opt-out.
• PostHog, Inc.: product analytics (see Cookies and similar technologies). It receives a stable account identifier but not your name or email; conversation recording, heatmaps, performance capture, and automatic interaction capture are disabled. It never receives prompts, messages, secrets, or API keys. In the EU/EEA/UK/CH it runs only after opt-in; elsewhere it is on by default with an opt-out.
• Intercom, Inc.: customer-support Messenger and support conversation management. For signed-in customers it receives a securely verified account ID, email, and name, together with the messages and attachments you choose to send to support. It does not automatically receive prompts, agent or workspace content, secrets, or API keys.

We will provide reasonable advance notice of material changes to this list. Under our Data Processing Agreement, customers get at least fourteen (14) days' notice before we add or replace a sub-processor, and can object.

5. Your prompts and the models

To generate responses, we transmit your system prompt, conversation history, and user message to the model provider via OpenRouter. We do not authorize providers to train models on your prompts; OpenRouter's terms forward this restriction to upstream providers and the major commercial providers (Anthropic, OpenAI, Google) honor it.

Two features also send text to a decision model, TypeSafe AI's Jev, through OpenRouter. Guardrails, on for every agent unless its owner turns them off, check each message, and each reply as it is written: we send that text with the agent's instructions and workflows and the conversation so far, and, for a reply, what the agent's tools returned during that turn (for example the part of a knowledge file it read), so a rule is judged against what the agent actually saw. The actions an agent tries (an email, a record update) are checked by OpenRouter itself, inside the model request already described above, against the owner's rules. Reflect, on Team plans and above, judges how an end user's conversation with an agent went: once the conversation has been quiet for 30 minutes, we send an excerpt of it (the first message and the most recent part), and store the resulting verdict (good, okay or bad) in your workspace. Reflect does not judge conversations of your own workspace members.

Despite this, please do not submit personal data, secrets, or other sensitive content that you would not want logged by a third party. Model providers may retain prompts for abuse-detection windows (typically 30 days) regardless of training restrictions.

6. Retention

We retain personal data only as long as needed to provide the Service, comply with legal obligations, resolve disputes, and enforce our agreements:

• Account, workspace, billing rows: while your account is active; deleted within thirty (30) days of account closure, except where law requires longer retention (for example, transaction records for tax, typically seven years).
• Conversation transcripts and chat history: while the parent conversation exists. You may delete a conversation in the console to remove the underlying transcript.
• Operational logs and request telemetry: thirty (30) days, then aggregated and de-identified.
• Encrypted backups: rolling thirty (30) days for disaster recovery.
• Stripe transaction records: per Stripe's policies and applicable tax/anti-money-laundering law.
• Support conversations: while needed to resolve and document your request and operate your account, then deleted or de-identified in line with account closure and applicable legal obligations.

We may retain de-identified or aggregated data indefinitely for product analytics and improvement.

7. Security

We maintain administrative, technical, and physical safeguards designed to protect personal data, including: TLS in transit; encryption at rest for databases at the infrastructure layer; salted SHA-256 hashing of API keys (we never log full keys); principle-of-least-privilege access for staff; and review of sub-processor security practices.

No system is completely secure. By using the Service you accept the residual risk inherent in any internet-based service. If we discover a security incident materially affecting your personal data, we will notify you without undue delay and consistent with applicable law (and within seventy-two (72) hours where required by Article 33 of the GDPR or equivalent).

8. Your rights

Subject to applicable law, you may have the right to access, correct, port, restrict the processing of, object to the processing of, or delete the personal data we hold about you. You may also withdraw consent (where consent is the lawful basis), lodge a complaint with your local data-protection authority, and opt out of the "sale" or "sharing" of personal information under California law (we do not sell or share personal information as those terms are commonly defined).

You can exercise most rights directly in the app: Settings exposes account data; "Delete workspace" cascades a deletion across our D1 database and cancels any active Stripe subscription. For other requests, email contact@agentnava.com or use the contact form on our Support page; we will respond within thirty (30) days, subject to identity verification. We will not retaliate for the exercise of any privacy right.

9. International transfers

We are based in the United States and process personal data in the U.S., as do our sub-processors. If you are in the European Economic Area, the United Kingdom, or Switzerland, your personal data may be transferred outside that region. We rely on Standard Contractual Clauses, adequacy decisions, or other lawful transfer mechanisms where applicable, and our sub-processors maintain similar commitments.

10. Children

The Service is not intended for, and we do not knowingly collect personal data from, anyone under the age of eighteen (18). If you believe a minor has provided us personal data, please contact us and we will delete it.

11. Do Not Track and Global Privacy Control

For essential service functions there is no industry consensus on the meaning of browser "Do Not Track" signals. Our non-essential product analytics (Mixpanel and PostHog) is off by default in opt-in regions and available with an opt-out elsewhere. Both services additionally honor your browser's "Do Not Track" setting, so analytics will not run for browsers that send it. We honor the Global Privacy Control (GPC) signal where required by applicable law.

12. Shopify stores

This section applies when a store installs the AgentNava app from Shopify. It covers the personal data of that store's shoppers.

Who decides

The store is the controller of its shoppers' personal data. The store decides why and how that data is used. We process it for the store, as its processor, and only to run the app.

What the app reads

• The store's pages and policies, once, when the owner sets the app up, through Shopify's Admin API. They become the store assistant's knowledge in the owner's AgentNava workspace, where the owner can see and change them.
• Products, prices and stock, at the moment a shopper asks, from the store's own public catalogue. They are not copied or kept.
• One order at a time, and only when a shopper asks about it and gives both the order number and the email used at checkout. The app answers only when both match the order, checked on our servers. It shares the order's date, payment and delivery status, items, carrier and tracking. It never shares an address, a phone number or payment details, and it never says which of the two did not match.

The app never reads the store's customer list or payment details.

Shoppers' personal data

The shopper data the app handles is what a shopper types into the store's chat, including an order number and an email when they ask about an order, and the order details listed above. We use it only to answer that shopper in that store's chat. We never use it for advertising. We never sell it. We never use it to train models.

The store's access

To look up orders, we keep the store's Shopify access token. It is encrypted, bound to that one store, and used only for that store's lookups. Each store's assistant has its own key, so one store's assistant can never reach another store's orders. We delete the token as soon as the store uninstalls the app.

Who else handles it

• Our cloud-infrastructure provider hosts the Service and the data it stores.
• OpenRouter, Inc. and the model provider it routes to receive the conversation text, so the model can write each reply.
• TypeSafe AI receives the text that guardrails check.

Each one is on our list of Sub-processors.

What we keep

We do not copy the store's orders or customer records into our database. A shopper's messages, and the order details the assistant read back to them, are kept only in the transcript of that conversation. The store owner can see that transcript and can delete it. We keep a shopper's chats while the store uses the app, and delete them within thirty (30) days of Shopify's request to delete the store's data (see below). Encrypted backups roll off as described in Retention.

Shopper requests

Shopify forwards a shopper's request to see their data or to delete it. We complete each forwarded request within thirty (30) days. For a request to delete, we delete every chat with the store's assistant in which the shopper's email or phone number appears. A shopper can also ask the store, or email us at contact@agentnava.com.

When a store uninstalls the app

Our access to the store stops at once, and we delete the store's access token. Shopify sends us a request to delete the store's data forty-eight (48) hours after the uninstall. Within thirty (30) days of that request, we delete every chat with the store's assistant and the app's records of the store.

Security

• Data is encrypted in transit with TLS.
• Data is encrypted at rest, including backups.
• Test and production systems and their data are kept apart.
• Staff get only the access their work needs, and their access is logged.
• If a security breach affects a store's data, we tell the store within seventy-two (72) hours of learning of it.

Consent and automated decisions

The chat starts only when a shopper writes to it. The app makes no automated decisions that have legal or similarly significant effects on shoppers.

Our agreement with stores

Our Data Processing Agreement puts these commitments in a contract between the store and Cerebro Labs Inc. It forms part of our Terms of Service.

Contact

For questions about a store's data, email contact@agentnava.com.

13. Wix sites

This section applies when a site owner adds the AgentNava AI Assistant app to a Wix site from the Wix App Market. It covers the personal data of that site's visitors.

Who decides

The site owner is the controller of their visitors' personal data. The site owner decides why and how that data is used. We process it for the site owner, as their processor, and only to run the app.

What the app reads

• The site's published pages, business details, policies and FAQ, when the owner sets the app up and each time the owner chooses "Read my site again". Published pages, the FAQ page among them, are read as any visitor sees them. Business details (name, description, contact email and phone, address, opening hours) and checkout policies are read through Wix's APIs. They become the site assistant's knowledge in the owner's AgentNava workspace, where the owner can see and change them.
• Services, if the site uses Wix Bookings: each service's name, description, length, price and booking page, at the same times. Open booking times are read only at the moment a visitor asks. The app never books anything.
• Products, prices and stock, if the site uses Wix Stores, at the moment a visitor asks, through Wix's APIs. They are not copied or kept. At setup the app only counts the products.
• One order at a time, and only when a visitor asks about it and gives both the order number and the email used at checkout. The app answers only when both match the order, checked on our servers. It shares the order's date, status, payment and delivery status, items, carrier and tracking. It never shares an address, a phone number, a name or payment details, and it never says which of the two did not match.

The app does not ask Wix for access to the site's contact list or its members.

Visitors' personal data

The visitor data the app handles is what a visitor types into the site's chat, including an order number and an email when they ask about an order, and the order details listed above. We use it only to answer that visitor in that site's chat. We never use it for advertising. We never sell it. We never use it to train models.

The site's access

The app keeps no Wix access token. Each time it needs to read the site, it asks Wix for a token for that one site, which lasts at most four hours, and holds it only in memory. The live lookups (products, open times and orders) work only with a key that belongs to that site's assistant, and we keep only a hash of that key, so one site's assistant can never reach another site's data. To build the assistant and read the site again, the app keeps a key to the owner's AgentNava workspace, encrypted and bound to that one site. We delete that key as soon as the owner removes the app.

Billing

Wix bills the site owner for the app's plans. We do not receive the owner's payment details. Wix tells us only which plan the site is on.

Who else handles it

• Our cloud-infrastructure provider hosts the Service and the data it stores.
• OpenRouter, Inc. and the model provider it routes to receive the conversation text, so the model can write each reply.
• TypeSafe AI receives the text that guardrails check.

Each one is on our list of Sub-processors.

What we keep

We do not copy the site's orders, products or customer records into our database. A visitor's messages, and the order details the assistant read back to them, are kept only in the transcript of that conversation, in the owner's AgentNava workspace. The owner can see that transcript and can delete it. Conversations stay there until the owner deletes them, also after the app is removed.

The app's own record of a site holds its Wix site ID, its name, its published address, its plan and the hash of its assistant's key. Encrypted backups roll off as described in Retention.

Privacy requests

Wix does not forward privacy requests to apps. A site owner, or a visitor of a site, asks us directly at contact@agentnava.com.

• A site owner asks us to delete their data. We delete the app's record of the site and reply within seven (7) days. The assistant and its conversations are in the owner's AgentNava workspace, where the owner can delete them, or we delete the assistant for the owner on their written request.
• A visitor asks us to delete what they said in a site's chat. The conversation belongs to the site owner's workspace. We forward the request to the site owner, who deletes the conversation in AgentNava, and we tell the visitor we did so. If the owner does not act within thirty (30) days, we delete every conversation with the site's assistant in which the visitor's email address or phone number appears.
• A request to see data. We send what the app's record holds about the site within thirty (30) days. Conversation transcripts are in the site owner's workspace and come from the owner.

When a site owner removes the app

The chat leaves the site and the live lookups stop at once. We delete the key to the owner's AgentNava workspace that the app held for the site. The assistant and its conversations stay in the owner's AgentNava workspace until the owner deletes them. To have the app's record of the site deleted as well, the owner emails contact@agentnava.com (see Privacy requests above).

Security

• Data is encrypted in transit with TLS.
• Data is encrypted at rest, including backups.
• Test and production systems and their data are kept apart.
• Staff get only the access their work needs, and their access is logged.
• If a security breach affects a site's data, we tell the site owner within seventy-two (72) hours of learning of it.

Consent and automated decisions

The chat starts only when a visitor writes to it. The app makes no automated decisions that have legal or similarly significant effects on visitors.

Our agreement with site owners

Our Data Processing Agreement applies between the site owner, as our customer, and Cerebro Labs Inc. It forms part of our Terms of Service.

Contact

For questions about a site's data, email contact@agentnava.com. How to use the app is described in AgentNava on Wix.

14. Changes

We may update this Policy. We will update the "Effective" date at the top and, for material changes, provide notice through the console or by email at least fourteen (14) days before the change takes effect. Continued use of the Service after the change constitutes acceptance.

15. Contact

Cerebro Labs Inc. develops and operates AgentNava. Our principal business location is Milpitas, California, United States. Business contact: contact@agentnava.com. About the company.

For privacy questions or requests, email contact@agentnava.com or use the contact form on our Support page.

For EU-related data-protection matters, you may contact your local supervisory authority. We have not appointed an EU representative under Article 27 of the GDPR; if you believe one is required for our processing of your data, please contact us at the address above.