Grok Bot stuck on “Reconnecting to your computer” — fully unusable — malformed listAgents gateway response

Support Bug Reports
,
1

Where does the bug appear (feature/product)?

Grok Bot

Describe the Bug

Grok Bot 0.43.0 on Ubuntu stuck “Reconnecting to your computer” — malformed listAgents gateway response

Steps to Reproduce

Grok Bot worked normally for about 30 minutes, then suddenly became permanently stuck on:

Reconnecting to your computer…

Retry does nothing.

I have already:

  • fully quit/restarted Grok Bot
  • signed out and back in
  • confirmed Privacy Mode/Data Sharing choice is explicitly saved
  • tried Recover Computer
  • verified normal internet connectivity
  • verified no VPN/proxy is active
  • killed all Grok Bot processes and relaunched
  • removed the local gateway-descriptor.json and allowed Grok Bot to regenerate it

The local daemon starts correctly, but Grok Bot consistently logs:

node-agent-coordinator: agents roster seed skipped:
SandGatewayMalformedReplyError:
gateway listAgents reply is malformed:
1: pushMessageContent: message: message:
secretRequest: target: required field

This exact error returns even after deleting the local gateway descriptor and relaunching, which suggests the malformed listAgents response is being returned fresh from the Grok Bot gateway/backend.

Could someone inspect my Grok Bot Agent Computer / gateway state from the backend and repair or reprovision it without deleting my existing Bots/files if possible?

Please also check whether there is a malformed queued secretRequest missing its required target field in my account/computer state.

Expected Behavior

The application should work and not be stuck on reconnecting.

Screenshots / Screen Recordings

Screenshot from 2026-09-05 17-34-59.png
Screenshot from 2026-09-05 17-34-59.png333×126 2.5 KB

Operating System

Linux

Version Information

  • Ubuntu 24.04
  • Grok Bot 0.43.0
  • no VPN
  • no proxy
  • cursor.com reachable over HTTPS
  • local local-exec-daemon starts normally

Additional Information

I reproduced this on two separate accounts.

A fresh account worked normally. Immediately after upgrading that account, Grok Bot became unable to reach its computer, exactly like my first upgraded account.

I fully ruled out local networking:

  • Ubuntu 24.04
  • VPN completely removed during test
  • normal DNS
  • https://api2.cursor.sh/ returns HTTP 200
  • Grok Bot local daemon starts

Grok Bot consistently receives a malformed gateway response:

SandGatewayMalformedReplyError:
gateway listAgents reply is malformed:
3: pushMessageContent: message: message:
secretRequest: target: required field

Recover does not fix it.

Please inspect the Agent Computer / gateway state for my upgraded accounts. This appears to be triggered by account upgrade/provisioning and likely requires backend repair.

Does this stop you from using Cursor

Yes - Cursor is unusable

5

Hey there @noname!

Your box should be back to normal now – please let us know if it isn’t!

7

Thanks, it did come back after the reset/backend fix, but unfortunately I’m now seeing several different Grok Bot failures again.

Current behavior:

  • The original “Reconnecting to your computer” / malformed `listAgents` issue was temporarily resolved.
  • I then started getting “Model provider is overloaded” on normal messages.
  • Background continuations also fail with “Background task follow-up failed — The model provider is under heavy load right now.”
  • After that, the error changed to “Bot failed to respond — Something went wrong while generating the reply. Try again.”
  • This is happening even while the UI still says the bot is “working.”
  • I already signed out/in and fully restarted Grok Bot.
  • I also used Reset Grok Bot’s Computer, which restored functionality temporarily, but the failures returned shortly afterward.

So the box is no longer permanently stuck on “Reconnecting,” but Grok Bot is still effectively unreliable/unusable because replies and background follow-ups repeatedly fail.

It does not seem specific to one task. The bot was able to work normally for a short period after recovery, then began failing again.

Could you please check whether there is still an issue with my Agent Computer/backend state or the model-provider routing for my account, and reprovision/recover it again if necessary?

I can provide request IDs / additional logs if there is a specific place you want me to pull them from.

17

Hey @noname, thanks for the detailed follow-up.

I just recreated your computer again a little while ago and it came up clean.

One question that would really help: did you ask any of your Bots to set up a VPN, proxy, Tailscale, or change DNS or network settings on the computer, or is there a routine or script that does something like that? The pattern we see is the computer losing its own connection to our servers a few minutes after a Bot runs some setup commands, and a VPN or DNS change inside the computer is the most common cause of exactly that

19

Yes, I tried to have the bot setup a VPN. Is there a more correct way to do this that I’m missing? Or did the bot put itself into a bad state by incorrectly setting up / using VPN?

23

Hey @noname, thanks for confirming!

Short answer to both: yes, the VPN caused it, and there’s no supported way to run a VPN on a personal Grok Bot computer right now.

The computer needs an always-on connection to our servers to run your Bots (that’s how messages, replies and credentials flow). A VPN client installed inside it takes over routing and DNS, so that connection fails. That’s also why a reset only helped briefly: resets keep your Bots, files, memory and routines, so the Bot re-ran the same VPN setup and knocked the fresh computer offline again.

What to do now:

  1. Ask the Bot to remove the VPN client and restore normal network settings. Delete any routine, script or saved instruction that sets up the VPN.
  2. Tell your Bots not to install VPN/proxy software or change DNS or network settings. Adding that line to the Bot’s instructions works well.

If you tell me what you needed the VPN for, I can try to point you to the right option!

  • Reaching a private or work network (internal apps, staging): supported on Enterprise via Team Setup, where an admin installs Tailscale or Cloudflare Tunnel on every team computer with an exit node inside your network.
  • Running something from your own machine or network: enable “Execution on Local Computer” in the desktop app’s Settings. The Bot runs those tasks on your machine (with per-action review), which can be on any VPN, while its own computer stays untouched.
  • Getting around a site block or appearing from another region: no supported path yet, the team is looking at it.
24

I have the exact same problem

28

Hey @noname, a quick update, since it changes part of my earlier answer.

There is now a supported way to do what you were after without installing a VPN on the Bot’s computer. In the Grok Bot desktop app, open Settings, go to Computer, and turn on “Route traffic through this computer”. While it is on, the Bot’s browser traffic leaves through your own desktop’s network and IP address instead of our cloud, so anything your machine can reach (including over a VPN running on your machine) the Bot’s browser can reach too.

A few things to know:

  1. Your current computer is from before this shipped, so the row will say “Update Grok Bot’s computer to enable local traffic routing”. Run that update from Settings, then Updates. It keeps your Bots, files and routines.
  2. The route only works while the desktop app is open and connected. When you close it, traffic goes back to the normal cloud path.
  3. Right now it covers the Bot’s browser. Commands the Bot runs in its terminal still go out through the cloud.
  4. Keep any VPN on your own machine, not on the Bot’s computer. Turning the route on does not touch the Bot’s own connection to us, so it will not cause the problems you hit before.

Let me know if the toggle does not show up or does not connect for you.