What AI providers see

What an agent sends

As it works, an agent sends its provider what it needs, which can include:

• Your messages, and the files and images you attach.
• Project files it opens, and the output of commands it runs.
• Your project instructions, working folder, and essential application instructions.
• The website open beside a conversation, background processes, and saved questions and answers, when the agent asks Enjoy for that information.
• Docs, workflows, and other conversations in the same project, when it reads them with Enjoy’s tools.
• Pages and screenshots from Enjoy’s browser, when it checks a website.
• Earlier messages in the conversation. If you switch agents partway through, Enjoy sends the new agent the messages it missed.

Enjoy’s tools give an agent access to the current project only. What else it can open on your computer depends on the agent and its permission mode. See Staying safe with agent permissions.

The provider’s own rules apply

Each provider handles what it receives under its own privacy policy and your account settings there, such as how long it keeps your data and how it may use it. Check those settings with each provider you use. For example, Muse Code subscriptions start in Contributor mode, where Meta may use your content to train its models until you turn it off in Muse Code’s settings. See Meta’s subscription guide. How Enjoy itself handles information is in the privacy policy.

Keep secrets out of reach

Anything in a message, or in a file an agent reads, can reach the provider. Don’t paste passwords, API keys, or card numbers into messages. See Keep passwords and API keys safe.

Photos you attach in the iPhone app have their location data removed before they’re sent to your computer.

Agents keep their own history

Agent programs keep their own record of their sessions on your computer, separate from Enjoy’s conversations. You may see Enjoy conversations in the agent’s own app.

Your messages reach every agent as you wrote them. The one exception is a message that starts with “/”, such as a file path: Claude Code, Grok Build, and OpenCode get it with “User message:” in front, so they don’t run it as one of their own commands. Claude Code doesn’t run a message with an attached image as a command, so that one reaches it unchanged. When you choose one of Claude Code’s, Codex’s, or Muse Code’s skills for a message, the agent gets the skill’s name with your message, so it runs that skill. See Run one of your agent’s skills. Enjoy doesn’t attach a snapshot of previews, background processes, or saved decisions. Agents look those up when they need them. Enjoy’s own instructions, such as how to reply and which folder to work in, are sent separately:

• Claude Code and Codex receive them outside your message, so they don’t appear in it in the agent’s app. Earlier messages an agent missed, such as after you switch agents, are sent the same way. When there are too many to fit, the oldest are left out, and the agent can read them with Enjoy’s tools. Enjoy adds a labeled note after your message instead when an older version of the agent can’t take them separately, or when Enjoy’s instructions for that message are too long to send separately.
• Grok Build, OpenCode, and Antigravity have no separate place for them, so Enjoy adds them as a labeled note after your message. The full setup note comes again each time Enjoy restarts the agent, which is usually once per request, so you’ll often see it in the agent’s own app.
• Muse Code has no separate place for them either, so Enjoy adds them as a labeled note after your message. The full setup note comes again each time Enjoy restarts Muse Code, which is usually once per request.

When a message comes from another device, such as your phone, or the conversation includes content Enjoy can’t trace to this computer, the instructions also say so, so the agent takes extra care. Older messages may still show an Enjoy context block.

If one of your own Claude Code hooks blocks a message, Claude Code doesn’t send it to the model. A follow-up blocked this way shows Not delivered, and the agent still finishes the rest of its work.

Enjoy reads past Claude Code and Codex sessions when you import conversations, skipping sessions already owned by Enjoy. See Import past Claude Code and Codex conversations.