Skip to main content
Hermes Agent can take actions on your behalf — reading and writing files in the workspace, running terminal commands, and calling tools exposed by MCP servers. Permissions control whether those actions happen immediately or only after you explicitly approve them. Choose a permission mode that matches how much autonomy you want to give the agent for the task at hand.

​
Permission modes

In manual mode, Hermes pauses before every action that requires a permission and displays a permission card in the chat. The card shows the action name, the specific parameters being requested (for example, the file path or command string), and a set of response buttons.You must respond before the agent continues. If the permission card is not answered within two minutes, the request is automatically denied and the agent receives a rejection response.This is the recommended mode when you are working in an unfamiliar codebase or running an agent task for the first time.

​
Changing the permission mode

​
Via settings

Open settings.json (or the Settings UI and search Hermes) and set hermes.permissionMode:
The change applies to the current session immediately — you do not need to restart.

​
Via the Control Center

Click the Control Center icon in the chat toolbar (or run Hermes: Open Control Center from the Command Palette) to open the embedded Hermes dashboard. The permissions section there mirrors the hermes.permissionMode setting and updates it in real time.

​
Approval options in manual mode

When a permission card appears in manual mode, you see up to four response buttons depending on what options Hermes sends: Permission cards remain visible in the chat history after you respond, so you can scroll back and review what the agent was approved to do during a session. Cards are also persisted in local session storage, which means they reappear correctly if you reload the extension or VS Code window mid-session.
autoApprove gives Hermes Agent permission to write files and run terminal commands in your workspace without prompting you first. Only use this mode in workspaces and environments you fully trust. Avoid it when working with sensitive data, production configurations, or codebases where an unreviewed write could cause harm.