Privacy & data handling

AgentHub AI LLC, a limited liability company registered in Florida, operates agenthub.ai and the AgentHub application. This notice covers information handled through both services.

We use information to provide accounts and workspaces, run the assistant and its AI features, keep each client’s requests, approvals, checks and updates, process subscriptions, apply plan and AI usage limits, send account and onboarding emails, respond to inquiries, protect the service, and understand use of the public website. We do not sell personal information.

We collect the account and workspace information you provide, including your name and email, client records, workflow records and safe structural details such as step names and provider types, requests, approvals, changes, verification notes, client updates, and evidence you choose to upload or paste. Step and node names can contain client information and are treated as private workspace data.

When you connect another service, we also store the connection details needed to provide that feature, such as the provider, account or workspace identity, authorized scopes, sync status, and encrypted access credentials.

If a workspace member emails a published client update through AgentHub, we store the recipient name and email, the email subject, who initiated the send, the provider message reference, and the send status as a delivery receipt.

When an account owner receives the new-account onboarding sequence, we store the message type, frozen message copy, recipient address, send status, provider message reference, and whether the owner stopped the sequence. The messages are text-only, and their links are not routed through email open or click tracking. AgentHub does not add open or click tracking.

If a workspace member creates a client approval link, we store a frozen copy of the requested scope, an expiration time, and the time the approval page is first opened successfully. The open time does not identify the person who opened the page. If someone responds through that link, we store the name they provide, an optional email address, their decision, any response note, and the response time. The raw link token is not stored; only its security hash is retained. We do not collect an opener identity, IP address, device string, or open count for the approval record.

If a workspace member creates a standing client request link, we store the client it belongs to and the link’s security hash. A person using that link provides a name, an optional email address, the requested change, and a submission time. The submission remains a draft until a workspace member confirms or dismisses it. Dismissing it removes the submitted name, email, and request text. The raw link token is not stored.

When a workspace reaches a managed-client limit, we store the plan, managed-client count, limit, and the action taken, such as starting checkout or choosing managed clients. If you answer the optional plan feedback question, we also store the reason you provide. AgentHub does not automatically include client names, request text, or other workspace content in these plan events.

If your workspace joins the Slack pilot, we record that it is in the pilot, when it joined and left, and when its free access ends after notice. If an older Free account joins, we also record when its owner agreed that the pilot terms replace that plan.

When AgentHub runs an AI step, we record the workspace, the member who asked, the type of step, the AI model, what the step could cost and what it did cost, and when it ran. We use these records to apply plan limits and control costs. They do not contain the text sent to the AI model.

If an owner or admin invites someone to a workspace, we store the invited email address, the role, who sent the invitation, and whether it was accepted, withdrawn or expired.

Our contact form collects your name, email address, selected topic, an optional answer to “Which best describes you?”, and your message. We use this information to respond to your inquiry. Messages pass through Lovable’s connector gateway and Resend to our inbox.

The application uses cookies to maintain your signed-in session and to remember choices such as hiding a card on Home, and browser storage to remember interface preferences. Blocking required cookies may prevent sign-in from working.

The public website uses Google Analytics through Google Tag Manager to measure visits and interactions, including call-to-action clicks. Analytics can collect browser and device information, approximate location, referral information, and pages visited, and uses cookies to distinguish visits. You can manage cookies in your browser and use Google’s Analytics opt-out browser add-on. The website and application do not change how they work in response to browser Do Not Track signals.

When you arrive through a tagged campaign link, we keep its source, medium, campaign and creative labels in a first-party cookie for up to 30 days. This cookie works across our website and application. We save those labels with your account when you sign up so we can measure signups and subsequent product use by campaign. We do not collect advertising click IDs for this purpose or send this account record to Reddit. You can clear or block this cookie in your browser; doing so does not prevent signup.

Public website fonts load from Google Fonts. Authentication pages and the website chat use Cloudflare Turnstile to help detect automated abuse. These providers receive connection or browser information needed to deliver their services. Hosting and security providers may also process technical logs. The approval-record limits described above do not mean that infrastructure providers never process IP addresses.

The website chat, “Talk to AgentHub”, is optional. You can type or talk. A chat starts only when you send a message or press Talk, and Cloudflare Turnstile runs a bot check first. Your browser asks for your microphone only if you press Talk. An AI from xAI answers, not a person. xAI receives your messages, your voice if you talk, and the conversation, and may record and transcribe them. For each chat, AgentHub keeps the time, the length, why the chat ended, how many messages you typed and how many searches the AI ran. To limit use, it also keeps a keyed hash of your IP address (for IPv6, of its /64 prefix): a scrambled value made with a secret key, not the address itself. A daily job deletes these chat records before they are 40 days old. AgentHub does not store your messages, the audio or the transcript of a chat.

AgentHub reads new messages only from Slack channels that an authorized workspace member explicitly selects and maps to an AgentHub client. The AgentHub Slack app must already be a member of the channel. Monitoring never reads direct messages between people, backfills channel history, or reads unselected channels.

A selected-channel message is temporarily stored while AgentHub decides whether it may be a workflow request. If it is not a candidate, the message text is removed after classification. If it is a candidate, the message and an AI-prepared draft remain in the Intake inbox for a person to confirm, edit, or dismiss. If Slack is disconnected before a message is sorted, the message stays, unsorted, for a person to review. Dismissing a candidate removes the stored message text. AgentHub never treats a Slack message as approval and never changes a connected workflow.

A workspace member who links their Slack account can message AgentHub privately in Slack, or choose a message with the Create a request shortcut. AgentHub stores what that person sends or chooses (including a client’s message they chose), the answers, request drafts and the requests they create, as part of their assistant conversations. A client update they ask for in Slack is a draft in the workspace’s client updates, like one made in the app. Messages from people who are not linked, Slack guests and people from other organizations are not stored. AgentHub replies only in the person’s own private messages with the app and never posts in a client channel. Word or Excel files AgentHub makes or changes in Slack are uploaded into those private messages too. Its replies and files stay in Slack, under your Slack workspace’s own retention settings; AgentHub does not delete them there. Slack data is never used to train AI models.

Disconnecting Slack in AgentHub, or removing the AgentHub app in Slack, stops AgentHub in Slack for everyone in that AgentHub workspace. AgentHub stops using Slack and ends the Slack links as soon as it learns of the removal (Slack tells it, or its daily check finds it, which can take up to a day). Within 10 days after that it deletes the copies it made only to run the app in Slack: the messages sent to the app that are waiting to be processed and the replies waiting to be sent, the Slack names it keeps for people and selected channels, link codes, questions waiting for an answer, Slack file references and the access Slack gave it. Disconnecting does not delete saved work. It stays in the workspace’s AgentHub records and can contain wording, names and links copied from Slack: channel drafts, with the message, its author’s name, its Slack link and channel name, and AgentHub’s suggestion (a draft waiting for review can still be confirmed or dismissed); conversations, with the messages people sent to the app, and their files; requests with their approvals and history, channel messages attached to a request, and any draft preparation that used a Slack message; client updates; answers that quote those requests; and contacts and workflow names added through the Slack request form. AgentHub also keeps the Slack workspace’s name and ID, the app and bot IDs, the IDs of selected channels, the Slack IDs and times on channel drafts, and a marker of which records came from Slack, to trace a record to its source and let a reinstall reconnect. The person who started a conversation can delete it in AgentHub; that removes its messages and drafts and starts a 30-day removal of its files. To ask for deletion of anything else, write to info@agenthub.ai.

When a workspace member asks AgentHub a question, in the app or in Slack, AgentHub stores the conversation: the questions, the answers, request drafts and any Word or Excel files it makes. Other members and the workspace owner cannot open the conversation. What the member chooses to share into a job, and any request created from a draft, become part of the workspace’s records, where other members can see them according to their role.

A member can share one version of a Word or Excel file by private link. Anyone with the link can read and download that version until the link ends. For these file links, AgentHub stores only a fingerprint of the link, not the link itself. A file link ends when the member revokes it, deletes the conversation or leaves the workspace, or when the workspace is archived.

Deleting a conversation removes its messages and ends its file links at once. The files themselves are deleted from storage about 30 days later. Copies already downloaded stay with the people who have them.

When a member forwards a client’s email to the workspace’s forwarding address, our email provider Resend receives it. AgentHub stores the forwarding member’s name and email address, the original sender named inside the message, the subject, the message ID, the result of the email’s sender checks and up to 12,000 characters of its text, as a draft for a person to review. The sender checks confirm the forwarding email came from the member’s mailbox; they do not verify the original sender quoted inside it.

When a client replies to a client update sent by email, Resend receives the reply. AgentHub stores the name and email address the reply shows, the subject, the message ID and up to 12,000 characters of its text, as a draft on that client’s record linked to the update it answers. If the workspace owner has client request notifications on, AgentHub also emails them a copy of the message.

Google connections are opened one workspace at a time. An owner or admin can ask for access on the Google connections page. Once a workspace has access, a member can connect a Google account for one client. The connection runs through Pipedream, our connection provider, which holds the Google sign-in. When you connect, Google asks only for access to the files you use with AgentHub and for your Google account’s email address. You choose files with Google’s own file picker. AgentHub stores the connection’s name, client, who made it, its permissions, the files the maker lists for teammates, and the Google account’s ID and email address, which only the maker sees. AgentHub does not store your Google password or access token; a short-lived token is used to open Google’s file picker.

AgentHub reads Google only when a member asks for it on a request: a range of up to 200 cells from a spreadsheet, or the text of a document. Nothing is read in the background. What is read is saved on the request, where the workspace’s owners, admins and operators can see it, and it can be sent to OpenAI when a member asks AgentHub to prepare that request. When a member with permission confirms the exact preview, AgentHub creates a new private Google Sheet from a saved request in the Google Drive of the person who connected that account. It does not edit or delete existing Google files.

The maker can share a connection with teammates, who can read only the files the maker lists. Removing a connection stops it at once. The person who connected a Google account can also remove AgentHub’s access at Google, which stops every connection that uses that account. What was already read stays on the request, and files created in Google stay in that Google Drive.

AgentHub’s use and transfer of information received from Google APIs follows the Google API Services User Data Policy, including its Limited Use requirements. We use Google data only to provide the features you ask for. We do not use it to train AI models, sell it, or use it for advertising. AgentHub staff do not read it unless you ask for help, to protect the service, or where the law requires it.

Each time AgentHub lists a connected n8n account’s workflows (when it is connected, when its workflow list is opened, and on Assign and Refresh), n8n sends every workflow the account’s API key can see, whole: including workflows that are never assigned, their settings, any code and credential references, and data a workflow saves between runs. AgentHub asks n8n to leave out pinned sample data, the run results a workflow can keep for testing. AgentHub uses only each workflow’s id, name, status and last change time, and the names and types of its steps, which steps are turned off, and how they connect, to show the list and import the ones a member assigns, and does not keep the rest. Using a key from an n8n user who can see only the intended projects limits which workflows n8n sends.

Setup reading is off unless an owner, admin or operator turns it on for an n8n account. Then AgentHub reads how each workflow assigned to a client is set up: when the setting is turned on, when a workflow is assigned, and when a member chooses Refresh n8n. n8n sends the whole workflow, including its settings, any code and credential references, and data it saves between runs; AgentHub again asks n8n to leave out pinned sample data. When that answer does not include the steps of the published version, AgentHub reads that version on its own. n8n offers no way to leave sample data out of that read. n8n’s current documentation for it lists the version’s steps, how they connect, and its name, description, authors and dates, but not sample data.

AgentHub keeps only summaries: what starts the workflow, the checks it makes, the fields it writes, the steps it runs and how they connect. A published workflow with unpublished edits gets one summary of each version. Each comes with n8n’s version ids, whether the workflow is published or archived, and when it was read. Earlier summaries stay as the workflow’s history until setup reading is turned off or the account is disconnected. AgentHub does not read, keep or analyze code, and does not keep credentials, webhook addresses, request bodies, data saved between runs, or sample data. Values typed into fields named like personal or secret data, such as keys, emails or phone numbers, are withheld. A sensitive value in a field with a plain name can still be kept. Only the workspace’s owners, admins and operators can see a summary.

In a member’s conversation about that client, the current summaries of the client’s workflows are sent to OpenAI with the question, so the answer can explain how a workflow is set up, and AgentHub records which summaries the answer was given. A saved answer keeps a short citation, not the summary. A summary is the configuration as read, not a record that a workflow ran or works.

Turning setup reading off, or disconnecting the n8n account, deletes the summaries at once and stops answers using them; a read already under way stores nothing. Answers already given stay in their conversations, and messages and files already sent are not recalled.

AgentHub uses AI to answer members’ questions and Help questions, prepare request drafts and Word or Excel files, write the summary of a client update a member asks for in Slack (from what the client will see in it), read an uploaded workflow file to describe its steps, run workflow checks a member starts, organize submitted evidence, prepare operational recommendations, identify likely workflow requests in selected Slack channels, and, with n8n setup reading on, explain how a client’s assigned n8n workflows are set up. AI output is a draft. A person decides whether to create a request, approve a change, or record verification.

AgentHub does not use your workspace content to train AI models.

Content sent to the OpenAI API is not used to train OpenAI models by default. AgentHub sends Responses API requests with store: false, which prevents application state from being retained for later retrieval. This is not the same as Zero Data Retention: standard API abuse-monitoring logs may retain content for a limited period under OpenAI’s API data controls.

Providers used by the website and application include:

• OpenAIProcesses the content needed for AgentHub’s AI-assisted features.
• SupabaseHosts authentication, application data, and private file storage.
• VercelHosts and runs the application, so submitted data passes through its infrastructure.
• ResendDelivers public-site contact inquiries, private new-account alerts, the short account-owner onboarding sequence, team invitations, trial notices, client request notifications, and published client updates when an authorized workspace member chooses Send. Resend receives the recipient address and message content. Onboarding and client update emails are text-only, and their links are not routed through email open or click tracking. AgentHub does not add open or click tracking.
• LovableProvides the public website platform, which carries website chats between your browser and xAI, and the contact-form connector that routes inquiries to Resend.
• GoogleProvides public-website analytics, tag management, and web fonts. If you connect it, Google Sheets, Google Drive and Google’s file picker provide the files you choose to read or create.
• PipedreamConnects Google accounts you choose to connect, holds their sign-in, and passes the reads and file creation you ask for.
• CloudflareProvides Turnstile protection on authentication forms and the bot check before a website chat.
• xAIRuns website chats. xAI receives the visitor’s typed messages, their voice if they talk, and the conversation, and may record and transcribe them.
• StripeProcesses paid-plan checkout, subscription billing, and payment account management. AgentHub does not store full card details.

If you connect Slack, Google, HubSpot, n8n, Make, or Zapier, data also passes between AgentHub and that provider according to the connection you authorize and that provider’s terms. AgentHub’s connections read data or receive it, with two exceptions: the Slack app replies to a linked member in their private messages with the app, and a Google connection can create a new Google Sheet when a member with permission confirms it. The application does not run or edit a connected client workflow.

We update this list before a new provider receives workspace content.

AgentHub may send a short sequence to help a new account owner test the product and share feedback. Every message in that sequence includes a way to stop the remaining onboarding emails. Stopping the sequence does not stop account confirmation, security, billing, client update, or other messages you explicitly request.

Connected-service tokens are encrypted before storage. Access to workspace data is limited by workspace membership and role. AgentHub staff see workspace activity, plan and connection health to run the service; that view does not show client messages or evidence. Staff open workspace content only to help when you ask, to protect the service, or where the law requires it. You are responsible for making sure you have permission to connect a service, select a Slack channel, upload evidence, and process the client data involved.

Do not paste passwords, private keys, API keys, or other secrets into request or evidence fields. AgentHub does not need them there.

Workspace records and confirmed request evidence remain until they are deleted or the workspace is closed. Some integration records may be retained long enough to preserve an operating history or meet security and legal obligations. Disconnecting a provider stops new access. For n8n, Make, HubSpot and Zapier, AgentHub also deletes its copy of the credential, and for HubSpot it deletes the imported contacts and deals. For n8n it also deletes the setup summaries, as turning setup reading off does. Imported workflows and received alerts stay in the workspace.

The workspace owner can download a copy of the workspace’s clients, contacts, workflows, requests with their approvals and recorded checks, client updates, incidents and request preparation history from the app, including while the workspace is read-only.

A workspace owner can close the workspace in Settings. Access ends at once for everyone. After AgentHub confirms that billing has stopped, it removes the workspace’s connections and deletes its clients, requests, client updates, files and conversations, usually within minutes.

Anyone can delete their own account in Settings (an owner closes their workspace first, or at the same time). This deletes the login, the profile, the person’s private conversations and files, and their own connections and API keys. Requests, approvals, checks and updates they worked on stay with their team, shown as “Former member”.

After closing or deleting, we keep payment records, AI usage amounts and a record of the closing or deletion itself, without the deleted content. Supabase, which provides sign-in, keeps its own audit entry of an account deletion, including the email address. Copies already sent by email, downloaded or posted in Slack are outside AgentHub.

Some deletions are handled by hand: an account with a set-aside workspace that still holds work, with marketplace sales or purchases, or with older records. For those, or to request deletion of anything else, email info@agenthub.ai from the email address you use for AgentHub. We may need to verify the request and preserve limited records where the law requires it.

AgentHub’s database is hosted in the United States. We and our providers process information in the United States and may process it in other countries where those providers operate.

AgentHub is for business use and is not directed to children. We do not knowingly collect information from anyone under 18. If you believe a child has given us information, email info@agenthub.ai and we will delete it.

Questions or requests to access, correct, or delete your information: info@agenthub.ai.