Resumable Agent Sessions and Usage Protection After Network Interruptions

Ideas Feature Requests
,
1

Feature request for product/service

Chat

Describe the request

am a long-time Cursor power user and currently subscribe to the $200/month Ultra plan. I would like to request a more reliable recovery and usage-protection mechanism for Agent sessions interrupted by temporary network instability.

During long Agent tasks, a brief network fluctuation can sometimes disconnect the conversation while the Agent is planning, generating code, running tools, or applying changes. In some cases, the user receives no usable final result, but model usage may already have been counted.

The user then has to repeat the same prompt or restart the task, consuming additional usage for work that may already have been partially completed. This is especially frustrating for Ultra users running long and expensive Agent workflows.

I suggest implementing a combination of the following mechanisms:

  1. Server-side Agent checkpoints

Save the task state periodically, including the plan, conversation context, completed tool calls, generated changes, and pending actions.

  1. Automatic reconnect and resume

When connectivity returns, Cursor should reconnect to the original Agent run rather than starting a completely new run.

  1. Local message and event caching

Cache unsent user messages, streaming responses, tool-call events, and Agent status locally until the connection is restored.

  1. Idempotent retries

Retrying after a connection failure should not repeat completed tool calls, duplicate file changes, or charge the user again for the same completed work.

  1. Automatic usage protection

If a run fails because of a verified network or server interruption and produces no usable response or changes, the associated usage should be automatically restored or excluded from billing.

  1. Clear recovery status

Display statuses such as “Connection interrupted,” “Agent still running,” “Reconnecting,” “Resuming from checkpoint,” or “Run failed—usage restored,” instead of leaving the user uncertain about whether the Agent is still working.

  1. Manual resume option

Provide a “Resume interrupted run” button that continues from the most recent checkpoint without requiring the entire prompt and context to be processed again.

A resilient checkpoint-and-resume architecture would improve both reliability and fairness. It would also make users more comfortable assigning longer and more complex tasks to Cursor Agent.

Thank you for considering this improvement.

7

Hey ! Thanks for the detailed write-up!

Cursor already runs agent turns against server-side checkpoints, and the client will reconnect and resume the same run after a dropped connection rather than starting over, with retries designed not to repeat completed work.

Improving reliability under flaky networks and making the recovery status clearer are areas we’re actively working on.

On billing, if an interrupted run leaves you charged for work you didn’t get a usable result from, please reach out to [email protected] support with the details, and we can look into it.