The tool approval card: the agent requests to run Email, you review the content, and choose Allow once — the agent proceeds and confirms the action.
Why tools need approval
Most tools require approval by default. This means every time the agent would call a tool, it asks first rather than running silently. You can see the tool name and a short preview of what it is about to do before you decide. Some tools — such as the image generation widget and certain internal context tools — have approval turned off by default and run without prompting. The tools you see in a conversation depend on what the agent’s administrator has enabled.Tool parameters configured by an administrator are shown in the approval card for transparency, but you cannot edit them. Only the agent’s administrator can change those settings.
Auto-decline when you move on
If you send a new message while an approval card is still waiting, IMP automatically declines that pending request before processing your message. You will not be stuck: the agent receives a denial response, and you can ask it to try again in your next message.Routine runs and approvals
When a scheduled or triggered routine reaches a tool that requires approval, the run pauses in a waiting for approval state. IMP opens the linked chat session so you can review the approval card and respond. Once you approve (or deny), the run resumes automatically — you do not need to restart it. The run banner in the chat links the session to the routine run, so you can navigate between the conversation and the run history on the Runs and schedules page.Tools that need an account or credentials additionally prompt for a connection via a credential card — see Connecting tools. Approval and authentication are independent: a tool can require both an approval step and a credential connection.
The approval card
When the agent requests to run a tool, a card appears inline in the conversation: Run{tool}?
“The agent wants to run this tool. Review the request and choose how to proceed.”
Below the description, a monospace block previews the key input the tool will use (for example, a file path, a command, a URL, or a query). Three buttons follow:
After you respond, the card collapses into a compact status line — “Approved:
{tool}” or “Denied: {tool}” — so the conversation stays readable.
Reviewing what you have already approved
Open Skills & tools from the ＋ menu in the composer. The Approved for this chat section lists every tool you have pre-approved in this conversation, with a Revoke button next to each one. Revoking a pre-approval means the agent will ask again the next time it wants to call that tool.Next steps
Skills and tools
Enable or disable individual tools and skills for a conversation.
Context compaction
Keep the conversation within the model’s context window.