MengBotDocs
Troubleshooting

Troubleshooting

Find the failing layer first, then apply the smallest safe fix.

Most MengBot failures come from one of four layers: the desktop app, a local agent CLI, a connected service, or a computer backend.

Report a bug from any chat: @devs

In any chat with a bot, on the desktop app or on the Cloud, type @devs (it is the last entry in the @ list) and a few words about what went wrong, then send. Typing it is the decision to send: there is nothing else to press.

The bot answering the message adds a short summary: what you said went wrong, what it saw itself in the conversation, and steps to reproduce. MengBot adds its own facts: the app version, the computer's system, whether it runs on the desktop or the Cloud, the bot's engine and model, recent errors in that conversation, the server's latest warnings and errors, and the conversation's ids. Keys, passwords, tokens, cookies and the parameters of web addresses are masked first. The rest of the conversation is never sent, and neither is any file.

A line in the chat says where it went, for example Sent to the MengBot team (report …). our team get it by email and will reply to you@example.com. The reply address is the email you signed in with, or the one in your profile. If the report cannot go yet (no connection, or too many reports in the last hour), the line says so: the report is saved and sent on its own later, so there is no need to send it again.

Under your message you can follow what becomes of it: Got it, we're looking, then Fixed in 0.1.135, coming in your next update once the team has fixed it. When the app (or the Cloud) runs that version, you get one message in the same chat, Fixed in 0.1.135. Want to try again?, and one notification. The app asks the MengBot team about your reports about once an hour and each time it starts; it sends only the reports' ids.

Collect useful evidence

Before opening an issue, record:

  • MengBot version and operating system;
  • selected engine and model;
  • the exact visible error, not only “it failed”;
  • whether the same agent CLI works directly in a terminal;
  • whether restarting only the affected task changes the result;
  • the file from Settings → General → Diagnostics → Export Diagnostics…, which removes keys and passwords (see Where are my logs?).

Never post API keys, OAuth callback URLs, bearer tokens, or the contents of secure credential files.

On this page