Which Network Does the Local Agent Use, and Does It Mean Offline Operation?
Distinguish between the network where the command is located and the model processing path; choose task data carefully.
Tool Requests Are Issued from the Execution Host
Local commands directly use the network accessible to the target host, so the choice of computer affects which addresses and resources it can access. Opening a browser on your current computer does not mean the command is also issued from this computer; check the environment selector before sending tasks. System proxies, VPNs, or network policies may also alter access results and need to be checked on the machine actually executing the task.
Local Execution Still Requires Service Connection
Both Desktop and Local Agent need to maintain a connection with the service to receive tasks and return results. Being able to access a target website does not mean authentication and command channels are functioning properly; conversely, showing online connection does not guarantee the target service is accessible. Without network, local connection cannot be treated as offline model operation; after reconnecting, check the original task status to avoid duplicate submissions.
Manage Data According to Actual Processing Content
Task messages, files, and command outputs may be sent to the server or related service providers for processing. Local attachments, working directories, and local execution do not guarantee "no data leaves the computer." Only provide information necessary for the current task before starting to avoid unrelated credentials or private content entering outputs; when troubleshooting, also remove connection tokens and full connection commands. Current data descriptions can be viewed on the Security and Trust page.