Security & Privacy

Agently is built with security at every layer: passwordless sign-in, workspace data isolation, OAuth for every integration, vaulted credentials, approvals before Jarvis acts, and Mac permissions you grant one at a time. Your workspace data is never used to train AI models. This page explains how your information is protected and what controls you have.

Signing in

Agently uses passwordless sign-in; there are no passwords stored in Agently.

• Email one-time code, enter your email, get a code, and you're in.
• Google sign-in, through Google's standard OAuth flow.

Sessions use signed tokens that are checked on every request and end when you sign out.

How your data is protected

Your data is yours. Everything in a workspace (conversations, documents, knowledge, cards) belongs to that workspace. Workspaces are isolated: one can never read another's data.

Private stays private. Knowledge and pages set to Just me, your private notes, and your personal mail and calendar connections are for you alone; other members, admins included, can't see them. Work your teammates can see, such as shared pages, Work cards and routines, can't read them either.

Encryption. All traffic travels over HTTPS, and database connections use TLS. Integration credentials are held in a secure credential vault; Agently's database stores only references to them.

How integrations are secured

Agently connects to more than 100 services through a secure connection layer:

• OAuth only. You approve access on the provider's own screen; Agently never sees your passwords.
• Vaulted credentials. Tokens are held in a secure vault, not in Agently's database.
• Minimal permissions. Only the access the tool needs is requested. Gmail, for example, is send-only.
• Personal where it should be. Gmail, Google Calendar, Outlook, LinkedIn and Superhuman only ever act for the member who connected them.
• Revocable. Disconnect any integration any time, from Agently or from the provider.

What Jarvis can and can't do

Jarvis can:

• Read from and act in your connected integrations, within the access you approved.
• Search your Brain.
• Create and manage cards, pages and routines.
• Search the public web.
• On your Mac, only if you turn it on: see your screen, use your apps, and work with files in your home folder.

Jarvis can't:

• Reach data outside your workspace.
• Use integrations you haven't connected.
• Act without a record: every step shows in the chat as it happens, and every app action is written to an audit log.
• Override workspace roles or permissions.
• Read another member's private data.

Approvals

In chats, voice and Slack requests, and on Work cards, Jarvis asks before it sends anything to people, deletes or archives anything, merges code, or uses a custom connector tool that isn't marked read-only. Reads, drafts and plain updates run straight away. Each routine asks before sending unless its creator set it to run unattended. See Approvals.

Jarvis also asks before saving a memory once it has read something someone else wrote, such as an email or a web page, so instructions hidden in that content can't act for you.

On your Mac

• Permissions only when needed. Agently asks macOS for the microphone, Accessibility, Screen Recording and the rest only when you first use a feature. See Grant permissions.
• "Hey, Jarvis" listens on your Mac. Nothing you say is sent anywhere until the phrase is heard.
• It is off until you turn it on. Jarvis can use this Mac and Allow scripts and commands are separate switches in Agently's settings on the Mac.
• Risky steps always ask. Steps that pay, send, delete or publish ask every time. Jarvis asks before scripts and most shell commands, and never reads credential folders or .env files.
• Nothing runs from shared work. Only requests from your own chat can reach your Mac; shared cards and routines can't.

Access control

Roles. Three workspace roles: Owner (everything, including billing and deleting the workspace), Admin (manage the workspace and members) and Member (use everything, no workspace management). See Workspace Management.

Permissions are checked on the server on every request, for workspaces, pages, cards, knowledge and settings.

Service to service. Agently's own services authenticate to each other with keys, so the agent service reaches your data only through authorized channels.

Rate limiting and monitoring

• Rate limiting. Requests are limited per user, with stricter limits on sign-in and other sensitive endpoints.
• Monitoring. Errors and requests are logged with IDs for debugging, without exposing sensitive data.
• Hardened responses. Standard security headers protect against common web attacks.

Controls you have

Export your data or delete your account in Settings → Privacy.

Security questions?

Choose Contact us in the app, or write to support@agently.dev.

Related Pages

• Integrations, how connections are secured
• Workspace Management, roles and access control
• Approvals, what Jarvis asks before doing
• FAQ, common questions and troubleshooting