Thanks for reporting this, and sorry for the slowdown.
A few things are worth checking first, since similar reports have come from a mix of service load, long-running chat context, network/proxy behavior, and local extension-host issues:
- Please update Cursor to the latest version, then fully quit and restart Cursor.
- Try starting a fresh Agent chat with Cmd+N on macOS or Ctrl+N on Windows/Linux. Long conversations can accumulate enough context to make responses slower.
- Test with extensions disabled by running this from a terminal:
cursor --disable-extensions
If performance improves, a third-party extension may be slowing down Cursor or the extension host. Re-enable extensions one by one to identify the culprit.
-
Run Cursor’s network diagnostics from Cursor Settings > Network > Run Diagnostics. If you are on a VPN, corporate proxy, Zscaler, or a managed network, also try a different network or temporarily disable the VPN/proxy if possible. Cursor uses streaming connections for AI features, and some network setups can stall those streams even when general internet access looks fine.
-
If you are on a proxy or corporate network, try Cursor Settings > Network > HTTP Compatibility Mode, set it to HTTP/1.1, then restart Cursor.
-
Try a different model to see whether the delay is specific to one model/provider or affects all Agent requests.
If it keeps happening, please share:
- Cursor version from Cursor > About Cursor
- OS and version
- Model selected
- Whether you are using a VPN/proxy/corporate network
- A Request ID from a slow Agent request (privacy mode disabled, if you can pleasE!)
You can copy a Request ID from the conversation’s “…” menu. That lets us look up the specific request path and distinguish model/provider delay, network stalls, extension-host issues, or Cursor-side infrastructure.