Permissions & Privacy

Assist is new in the IDA 9.5 beta, and its documentation is subject to change. Treat it as a work in progress rather than a finished reference.

Assist can read your database, run code, and (when allowed) modify the database, run Python, use a shell, or write files. Because an AI agent is deciding what to do, the permission model is designed so that nothing consequential happens without a decision you can see and make. You can edit this under Settings → Permissions.

The permission model

Each capability can be set to:

• On: the agent may use it freely.

• Off: it is removed from the advertised tool list and blocked from executing.

• Ask when needed: the agent may request it mid-turn, and you approve or deny it with a permission card.

Permission cards

When the agent requests a capability that requires approval (a shell command, Python, the debugger, or a file write), Assist shows a permission card for you to approve or deny.

Defaults

Out of the box, reading files is enabled. Capabilities that require approval are seeded to Ask when needed.

Capability ceilings and external clients. A capability set to Off applies to MCP clients and mesh peers as well, and they cannot use the interactive escalation path. A capability set to Ask is therefore effectively unavailable to an external agent; it will receive a refusal rather than a prompt. Choose On or Off for capabilities you want automation to use.