Privacy Policy

This Privacy Policy explains how Honeybase ("we", "us") collects, uses, and safeguards personal data when you visit honeybase.ai, request early access, use the Honeybase application at app.honeybase.ai, or fill in a form an organization shared with you at forms.honeybase.ai. It also describes the rights available to individuals in the European Union (EU), European Economic Area (EEA), and United Kingdom (UK) under the General Data Protection Regulation (GDPR) and UK GDPR.

1. Who is responsible for your data

Honeybase is the data controller for the personal data described in this policy. For any privacy question or to exercise your rights, contact us at privacy@honeybase.ai.

One exception: if you filled in a form that an organization shared with you as a public link, that organization — not Honeybase — decides what it asks and why, and is the data controller for your answers. We only store them on its behalf, as its processor. See “Responses to forms shared with you” below, and section 7 for how to exercise your rights over those answers.

2. What we collect

Several items below say that content is sent to the AI provider you connected. A new workspace starts with Honeybase AI credits, and while it runs on those credits rather than on a provider of your own, the AI provider in every one of those items is the one we choose — OpenRouter and OpenAI, listed in section 4. As soon as you connect your own provider and use it, that provider is yours again.

• Early-access form: your work email, and optionally your company name and team size.
• Account data: your first and last name, email address, and the organization and team you belong to.
• Authentication data: a securely hashed password (we never store it in plain text) or, if you sign in with Google, your verified Google account email.
• Content you create: the workflows, tasks, and data you and your team enter into the product, which may include content sent to AI providers when you run AI-powered steps, when you use the Barry AI assistant in the workflow editor, when the assistant needs to tell two of your connections apart and reads their non-secret connection settings (which server, database, mailbox address, web address or AI provider a connection points at — never its password, key or token, which it has no way to read), when you paste an image into a Barry AI conversation (the image is then sent to your chosen AI provider, and re-sent each time that conversation continues), or when you ask it to debug a workflow run (the run's step outputs are then shared with your chosen AI provider, with credential-like fields redacted), or when you ask it to build a web app (the app’s own source code and a list of your organization’s endpoints are then shared with your chosen AI provider).
• What your AI agents read and do through their tools: when a member of your organization attaches one of your connections to an AI agent and ticks a tool, that agent can use the connection on its own. Depending on which tools are ticked, it can read messages in a connected mailbox, read the cells of a connected Google spreadsheet, and read the text of documents in a connected Google Drive folder — and that content is then sent to the AI provider you connected, as untrusted material, so that the agent can answer with it. Google Sheets and Google Drive are two separate connections asking for two different things: a Sheets connection reaches your spreadsheets and can see only the names of your other files, while reading the contents of a document needs a Drive connection, which you grant separately. A Google connection made before we separated the two keeps the wider access it was originally granted, is labelled “Google Sheets + Drive” wherever you see it, and can be narrowed to spreadsheets only in one click. Documents are read as TEXT only, are size-limited, and PDFs, Office files and images are refused rather than guessed at. It can also act: send email as your organization, post to a connected Slack workspace, send an approved WhatsApp template to a phone number, write or clear rows in a connected spreadsheet, and create or change tasks. Which tools an agent holds is always a decision a member of your team makes — the tools that read someone’s correspondence or documents are never switched on for you — and every call an agent makes is recorded in its run history. Tools that write ask for approval by default, but an agent you have released to run on its own cannot pause for approval, so on a released agent those actions happen without a person in the loop.
• Conversations with agents inside your apps: an app you deploy can offer a chat with one of your AI agents. What people type there, and the agent's replies, are stored with that agent's run history and sent to your chosen AI provider, in the same way as everything else above. An agent is reachable from an app only if it has been released, explicitly allowed to be used by apps, and listed by that app — three separate decisions your team makes. Apps require a Honeybase sign-in, so today the people using them are members of your organization.
• Feedback you give on an AI agent’s work: when an agent has worked on one of your tasks, the task’s owner (or their team, or an agent administrator) can rate the result and write what should have happened. That rating and comment are stored with the agent’s work on the task. If you then ask Barry AI to turn the feedback into a lesson, the conversation — which includes the task, the agent’s report and your comment — is sent to your chosen AI provider like any other Barry chat, and a lesson you approve is stored on the agent and included in that agent’s future task runs, so it can reach your AI provider again on later tasks until someone in your organization removes it. Only people can add these lessons; agents cannot write their own.
• Your organization’s directory and work, when you use the Barry AI assistant on your home page: to answer questions like “who is on the sales team?” and to build an automation from a sentence, the assistant can read your organization’s members (names, email addresses, roles, team membership), teams, tasks, processes and workflows, forms, the names of your secrets, the names of your integrations together with their non-secret connection settings (which server, database, mailbox address, web address or AI provider each one points at, so it picks the right connection rather than guessing from its name), and — if you have granted it — your organization database. That content is sent to the AI provider you connected, under your agreement with them. It is limited to what your own account can already see in the app, requires the “build workflows” permission to use at all, and secret values are never readable by the assistant, whatever permissions you hold. The assistant can also ask you to fill in a setup form inside the chat — to connect an integration, store a secret, invite a colleague or turn on your organization database. It chooses which form appears; you type every value, and those values go straight to secure storage on the same path the equivalent settings page uses. Credentials, secret values and invitation details never enter the conversation and are never sent to the AI provider: afterwards the assistant is told only what happened (for example an integration’s name and non-secret settings, a secret’s name, or whether an invitation now exists). Finally, the assistant can carry out a small number of actions in your organization — starting a run of one of your processes, pausing or resuming a process or an AI agent, and changing whether a person is given work from a process. It never does any of these on its own: each one shows you a confirmation card naming exactly what it will do and what follows from it, and nothing happens unless you approve that card. Starting a run means the automation’s own steps really execute, so any emails, records or calls to services you have connected happen for real. These actions are carried out with your own account’s permissions, and are recorded in the same history as if you had done them from the app yourself. To report back on a run it started — or on one that failed — the assistant can also read your organization’s run history from the home page, limited to the runs your own account can already see; those step outputs are then shared with your chosen AI provider, with credential-like fields redacted. The assistant can also read through the systems you have connected, to build from what is really there rather than a guess: the tables and sample rows of a connected database, messages in a connected mailbox, the channels of a connected Slack workspace, the cells of a connected Google spreadsheet and the text of documents in a connected Google Drive folder. These are reads only — it cannot write to a connected system from the chat — and what it reads is sent to the AI provider you connected, as untrusted material, and kept in that conversation’s transcript, which only the member who had the conversation can open. Reading needs the “manage integrations” permission on your own account; an administrator can exclude any connection from the assistant on the Integrations page, and you can switch a connection off for one conversation in the chat itself. You can also paste an image into the conversation — a screenshot, for example. Pasted images are stored in our private file storage and sent to your chosen AI provider along with your message; because the conversation is replayed on each turn, an attached image is re-sent for the rest of that conversation. They are deleted after 30 days.
• Connected AI apps (MCP): you can connect an AI assistant of your choice (for example Claude) to your Honeybase account through our MCP connector. After you sign in and explicitly approve the connection for one of your organizations, that assistant can read (and, with the scopes you approve, create and edit) your processes, workflows, run results, tasks, schedules, AI agents and Agent tools on your behalf, and can see which external MCP servers your administrators have connected and which of their tools were approved (names and settings only — never the credentials for those servers) — so that content flows to the AI app you chose to connect, under that provider's own terms. Reaching the source code of your apps, their file lists and their build logs requires separate app permissions that you approve on their own; a connection you approved before those existed cannot use them. Credential-like fields are redacted and access is limited to what your own account can see. You can revoke a connection at any time, and revocation takes effect within seconds.
• Files from services you connect: when a workflow step reads files from a service you have connected (for example a Google Drive folder), we copy those files into our own private storage so that later steps can use them. Their contents are whatever your own documents contain. They are stored privately, are never included in the workflow’s saved run results, and can only be downloaded through a short-lived link by a member of your organization. They are shared with an AI provider only if you build a step that asks for it — an AI step whose “files” setting names them — in which case the file is sent to the provider you connected, as part of that step’s request. We delete them automatically after 30 days.
• External MCP servers you connect: an administrator in your organization can connect Honeybase to an external MCP server that your organization chose (for example your issue tracker or an internal tool). Once connected, and only for the specific tools an administrator has reviewed and approved, your AI agents can call that server, and so can the Barry AI assistant on your home page when you ask it to (a tool approved as read-only runs directly; one approved as mutating only after you approve the exact call on a card) — so the arguments those calls carry, which may include workflow, task or chat content, are sent to that third party under its own terms. Honeybase does not vet those servers or their operators. If the server requires you to sign in, the access and refresh tokens it issues are held in our secrets vault for that connection only, are never shown in the product or included in any AI context, and are deleted when the connection is deleted. Every call is recorded in an access log kept for 365 days.
• Files from services you connect: when a workflow step reads files from a service you have connected (for example a Google Drive folder), we copy those files into our own private storage so that later steps can use them. Their contents are whatever your own documents contain. They are stored privately, are never included in the workflow's saved run results and are not shared with any AI provider, and can only be downloaded through a short-lived link by a member of your organization. We delete them automatically after 30 days.
• In-app notifications: messages we generate for you in the product (for example, when a workflow run fails), stored against your account so you can see them in the notification centre. These stay within our own systems and are not shared with any third party.
• Availability status & history: whether you are currently available to work, which processes you accept tasks for, and a change history of those settings (who changed them and when), used to route work within your team. Other members of your organization can see this status; changing it for someone else requires a permission granted by your administrator. This data stays within our own systems and is not shared with any third party.
• Responses to forms shared with you: an organization that uses Honeybase can share a form as a public link that anyone can fill in without an account. If you submit one, we store your answers for that organization, together with the version of the form you answered. We deliberately store nothing else: no IP address, no browser or device details, and no name or identifier of any kind beyond whatever the form itself asked you for. The organization that shared the link chooses the questions and how long to keep the answers. If the form asks for a file, the file you attach is stored in our private storage for that organization on the same terms. The page you fill the form on sets no cookies at all; while you are filling it in, your unsent answers are held in your own browser tab so a dropped connection does not lose them, and they are removed as soon as you submit or close the tab.
• Organization audit history: we record who created, changed or deleted core workspace objects, the channel used (such as the web app, Barry AI, an API key or a workflow), and bounded records of what changed. These records include workflow code, prompts and configuration. Dedicated credential fields, integration configuration and task or form response data are excluded from these change records.
• Technical logs: limited operational logs (e.g. request method, path, and timing) used to keep the service reliable and secure.
• Product usage: when you are signed in to the application, we record which screens you visit and which controls you use, and any answers you choose to give to an in-app feedback prompt. These events are tied to a pseudonymous identifier derived from your account (a one-way hash — your email address and name are not sent) and to your organization’s numeric identifier, together with technical details such as browser type, screen size, language and timezone. The text of what you type, the content of your workflows and tasks, and your IP address are not included, and we do not record your session. We do not run advertising trackers.

3. Why we use it and our legal bases

• To provide the service (managing your account, running your workflows — whether you start a run yourself or approve the Barry AI assistant’s request to start one) — performance of a contract.
• To respond to early-access interest and contact you about the product — legitimate interest.
• To keep the service secure and reliable (authentication, rate limiting, fraud and abuse prevention) — legitimate interest.
• To understand how the application is used and improve it (product usage events, in-app feedback prompts) — legitimate interest; you can object at any time via the contact address below, and we honour your browser’s Do Not Track and Global Privacy Control signals automatically.
• To comply with legal obligations where applicable — legal obligation.
• To store responses to a publicly shared form — we process these on the instructions of the organization that shared the link; that organization is responsible for having its own legal basis for collecting them.

4. Service providers (sub-processors)

We share personal data only with the service providers needed to operate Honeybase, each under a data processing agreement:

• Google Cloud Platform — application and database hosting (United States).
• Resend — transactional and notification email delivery.
• Google (Sign in with Google) — optional authentication; used only to verify your identity at sign-in.
• Google reCAPTCHA — protects our forms from automated abuse; collects device and usage signals to distinguish humans from bots.
• PostHog (PostHog Cloud US, hosted in the United States, under the same safeguards as our other US providers) — product analytics for the signed-in application: receives the pseudonymous usage events described in section 2 and any in-app feedback you submit. Not used on this website or on publicly shared forms.
• Infisical — secure storage of secrets and integration credentials.
• Meta (WhatsApp Business Cloud API) — if you connect a WhatsApp Business number, sends and receives WhatsApp messages on your behalf; processes sender/recipient phone numbers, message content and contact profile names, including data of the people who message your business.
• AI providers (via our model gateway) — process the content of AI-powered workflow steps you choose to run (including any files that step attaches), of your Barry AI assistant conversations (including workflow definitions, images you paste into a conversation and, on the home page, your organization’s directory and work items), and of workflow run outputs when you ask Barry AI to debug a run in the workflow editor or to report on a run from your home page, and of the source code of apps you build with Barry AI together with a list of your organization’s endpoints, and of conversations people have with agents inside your deployed apps. When you bring your own AI provider key, the provider is your choice and is not our sub-processor (see below). Only when an organization uses Honeybase AI credits is the provider our choice — the two entries that follow.
• OpenRouter, Inc. — AI model gateway for Honeybase AI credits, only when an organization uses Honeybase AI credits (United States); reached through our own model gateway, which instructs OpenRouter to send the request to OpenAI itself and to no other host, and not to let OpenAI retain your content or train on it.
• OpenAI, L.L.C. — model provider reached through OpenRouter for Honeybase AI credits, only when an organization uses Honeybase AI credits (United States).
• Neon (a Databricks company) — if your organization activates its dedicated database, hosts that PostgreSQL database on AWS (us-east-1, N. Virginia); processes whatever operational data your organization chooses to store in it, which may include personal data of your own users or contacts.
• Exa, Firecrawl and Browserbase — provide web search, page retrieval and browser interaction for AI agents. They process the queries, URLs, page content and browser instructions needed for the tools you enable, which may include information from conversations, tasks or form values. With your own vendor key, the vendor is your chosen processor. Browserbase's Stagehand model calls use your selected AI connection through Honeybase by default; selecting its Model Gateway sends that inference to Browserbase's gateway.

AI apps you connect yourself — whether by bringing your own AI provider key or by connecting an assistant through our MCP connector — are your chosen providers, not our sub-processors: they receive data only because you connected them, and under your agreement with that provider. Honeybase AI credits are the exception: there we choose the provider, which is why OpenRouter and OpenAI are listed above. The budget token that meters those credits for your organization is ours, not yours — you never enter it, never see it, and it is never shared with any third party.

We do not sell your personal data.

5. International data transfers

Honeybase is currently hosted in the United States. If you access the service from the EU, EEA, or UK, your personal data is transferred to and processed in the United States. We rely on appropriate safeguards for these transfers, including the EU–U.S. Data Privacy Framework where available and the European Commission's Standard Contractual Clauses with our providers.

6. How long we keep it

• Account data is kept for as long as your account is active, and deleted on request.
• Organization audit history is kept for 365 days, including after individual objects are deleted. Current creator and updater attribution remains with each object. Remaining legacy internal change logs are deleted after 7 days.
• Files copied from services you connect are automatically deleted after 30 days.
• Async endpoint calls: when you explicitly request a background endpoint run, we store its request and workflow results to execute it and let you retrieve the answer. Result access expires seven days after acceptance; request and response payloads are removed on the next hourly cleanup, and the starting user or API-key identifier is kept one extra day. Run lifecycle metadata may remain in Run History. Synchronous endpoint calls are not recorded in Run History.
• Authentication tokens are short-lived and expire automatically.
• Early-access enquiries are kept only as long as needed to follow up with you.
• Responses to a publicly shared form, and any files attached to them, are kept until the organization that collected them deletes them. Closing or revoking the link stops new responses but does not delete the ones already given. These files are not covered by the 30-day rule above, which applies only to files copied from services you connect.

7. Your rights

Subject to applicable law, you have the right to access, correct, delete, or export your personal data, to object to or restrict certain processing, and to withdraw consent where processing is based on it. To exercise any of these rights, email privacy@honeybase.ai and we will respond within 30 days. You also have the right to lodge a complaint with your local data protection authority.

If your request concerns answers you gave to a form an organization shared with you, address it to that organization: it is the controller, and it decides whether to correct or delete them. Write to us at the address above if you cannot reach them, or do not know who they are, and we will help identify and contact them. Because we deliberately store no identifying information alongside such answers, we may need details from the form itself to locate them.

8. Cookies and local storage

The Honeybase landing page sets no tracking or advertising cookies. The application uses strictly necessary storage to keep you signed in and to remember preferences such as your language and theme. Our product analytics (PostHog, section 4) is configured so that it sets no cookie and stores nothing on your device — it keeps its state in memory for the duration of the page only — and it respects your browser’s Do Not Track and Global Privacy Control settings. We do not use advertising trackers. Where a form is protected by Google reCAPTCHA, Google may set cookies and process device signals for the sole purpose of fraud and abuse prevention on our early-access, sign-up and sign-in forms and on publicly shared forms; this use is governed by Google's privacy policy.

The page where you fill in a form an organization shared with you sets no cookies of its own. It loads Google reCAPTCHA only at the moment you submit — not when you open the page — so simply opening a link someone sent you does not contact Google.

9. Changes to this policy

We may update this Privacy Policy from time to time. Material changes will be reflected by updating the "Last updated" date above and, where appropriate, by notifying you directly.