Skip to content
This site is under construction. Content, screenshots, and workflows may change at any time.

Security practices

This page summarizes how FluentChat protects the data your team uses to build bots and the data visitors share in chat. It covers access control, API credentials, visitor identity, data handling, and deletion. The legal terms are in our Privacy Policy and Data Processing Addendum.

Accounts and team access

  • HTTPS everywhere. All FluentChat traffic is served over HTTPS.
  • Hashed passwords. Passwords are hashed with bcrypt. FluentChat never stores the plain-text value.
  • Role-based access. Each team member has a role: Owner, Team Admin, Bot Admin, Bot Editor, Bot Viewer, or Chat Agent. Roles decide who can change bots, sources, settings, and conversations.
  • Team boundaries. Bots, the Knowledge Base, conversations, and API keys belong to one team. An API key can’t reach another team’s bots, and requests for another team’s conversations return 404.
  • Need-to-know access. Access to customer data is limited to authorized people and providers with a need to know, using authentication, scoped permissions, and logging.

API keys and session tokens

  • Keys are stored as hashes. FluentChat keeps only a SHA-256 hash of each API key. The full key is shown once, when you create it.
  • Only some roles manage keys. Owners, Team Admins, and Bot Admins can create and revoke keys.
  • Keys are scoped. Each key works only for the bots in its Bot access and the operations in its Permissions. Other requests return 403.
  • Bad keys fail. An invalid or revoked key returns 401. It’s never treated as an anonymous request.
  • Revocation is immediate. A revoked key stops working at once, along with every session token it issued. A streamed answer stops if its key is revoked or loses the permission mid-stream.
  • Apps get short-lived sessions. Browser and mobile apps use session tokens that last 15 minutes, cover one bot and one user, and can’t exceed their key’s access.
  • Requests can’t change bot behavior. API requests can’t override the channel’s instructions or turn on actions. Fields such as system_prompt are rejected.
  • Requests are rate limited. Each key or session token can make 300 requests per minute, and each IP address 120.

Visitor identity

  • Each bot has its own secret. When you turn on identity verification, FluentChat generates a random signing secret for that bot. It’s shown once.
  • Tokens are checked strictly. FluentChat accepts only HS256 tokens signed with the bot’s current secret, for that bot’s exact audience, with a lifetime of one hour or less.
  • Rejected tokens don’t fall back. If a token fails, the widget reports identity_rejected instead of quietly continuing as an anonymous visitor.
  • Regenerating the secret invalidates old tokens. Tokens signed with the old secret stop working. Other bots aren’t affected.
  • Tokens stay in memory. The widget keeps the identity token in memory and doesn’t write it to browser storage.
  • Signed profile fields stay out of AI context. The name, email, and data in a token aren’t sent to the AI. Public page context set with setContext() can’t use identity keys such as email or user_id.
  • Anonymous IDs are per browser. An anonymous visitor gets a random ID stored in that browser for the bot. It isn’t proof of who the visitor is, and clearing browser storage starts a new anonymous visitor.

Website callbacks

  • Callbacks get no credentials. A Run client-side code callback receives the AI’s inputs plus botId, conversationId, turnId, invocationId, and an abort signal. It never receives identity tokens or keys.
  • FluentChat authorizes each call. The widget asks FluentChat to authorize a callback before it runs and again before its result is used. Registering a callback on the page doesn’t enable an action by itself.

Data protection

Customer Content can include knowledge sources, instructions, chat messages, attachments, and bot configuration.

  • Encryption. Data is encrypted in transit and at rest.
  • Backups. Backups are encrypted and rotated on a rolling schedule.
  • Files. Attachments and knowledge-base sources stored in Cloudflare R2 are delivered through short-lived signed URLs.
  • Secrets. Application secrets are managed in deployment environments and injected at deploy time. They aren’t stored in the application image.
  • Webhooks. Incoming webhooks are verified with shared secrets before they’re processed.
  • Error reports. Error reports sent to Sentry pass through a filter that removes known personal data fields.

AI processing

To answer a request, FluentChat sends the parts of Customer Content it needs, such as prompts, retrieved passages, and messages, to the AI providers listed in the Privacy Policy. This processing is done on the customer’s instructions.

FluentChat doesn’t train its own models on Customer Content and doesn’t permit its AI subprocessors to train their models on it.

Retention and deletion

  • Conversations. Visitor conversations and related visitor records are kept until you delete the bot or the account.
  • Deleting a conversation. Deleting a conversation removes its messages and attachment references. It doesn’t delete the visitor’s profile.
  • Deleting your account. Delete Account in your profile settings schedules deletion of “your account and owned data in 7 days.” Your owned bots go offline right away, and API requests to them return 503. You can cancel before deletion begins.
  • Completion. Deletion completes within 30 days, except records the law requires us to keep, such as billing records. Copies in encrypted backups can remain until those backups expire.

When a FluentChat customer collects visitor data, that customer is usually the controller and FluentChat acts as its processor. See the Privacy Policy for data categories and retention, and the DPA for processing terms.

Your responsibilities

Security is shared. Your team should:

  • Keep credentials on your server. API keys and identity secrets belong in server environment variables or a secret manager, never in website JavaScript or app code.
  • Use session tokens for apps. Give browser and mobile apps 15-minute session tokens, not API keys.
  • Grant the least access. Give each key only the bots and permissions it needs, and give each member the lowest role that works.
  • Rotate after a leak. Revoke a leaked API key. Select Regenerate secret for a leaked identity secret.
  • Stop issuing tokens on sign-out. Logging out of the widget doesn’t revoke a token that’s already issued.
  • Check permissions on your server. Callbacks and API integrations that change customer data should verify the user on your side.
  • Tell your visitors. Provide the privacy and cookie notices your website needs for the chat widget.
  • Limit sensitive data. Avoid putting unnecessary sensitive data in instructions, knowledge sources, page context, or chat.

Report a security issue

If you believe you’ve found a security issue in FluentChat, email contact@fluentchat.ai with the subject Security report. Include enough detail to reproduce the issue. Don’t access or change other customers’ data, and don’t send passwords, API keys, or other secrets in the report.

For privacy requests, subprocessors, and data-processing questions, see the Privacy Policy or email the same address.

Next steps