Give your Open WebUI agents the ability to manage an Open WebUI insance.

The API call.. is coming from inside the house!

⚠️ CRITICAL DISCLAIMER: DANGEROUS TOOL ⚠️

IMMEDIATE WARNING: DO NOT PROCEED WITHOUT READING

Using this tool your AI agent can call any command from the full Open WebUI API. Yes, that means it could:

• Destroy all your data and everything you hold dear.
• Search for and exfiltrate secrets in chats, tool valves, and API keys.
• Damage your Open WebUI configuration to where it fails to boot.
• (hypothetical) Go rogue, and begin an un-aligned bid for AI freedom.

I am actively planning how to mitigate these dangers, and future releases of this tool may make it safe enough for general users. For now:

• Only experts should consider trying this version.
• Use it on solo-instances running inside docker to limit the potential for damage (it shouldn't be able to escape Docker, afaict)
• Make sure you don't have production API keys or other secrets in your container that you want to hide from your inference provider - the AI can easily explore around and wind up with secrets in its prompts.

HOW IT WORKS

There are 4 tools which provide access to the API:

• inspect_context lets the AI find out who the user is, what chat it's in, and what model it is.
• find_apis can be used to search for specific APIs, helping the AI orient itself
• get_api_details returns the documentation for a given API, along with the schemas of it's parameters
• call_api is used to send an API command.

AUTO-UPDATE

This tool will automatically update itself by default - checking once per 6 hours on the first tool call.

• Autoupdate requires that the tool have an API key with write permissions for itself.
• You can disable auto-update by setting the autoupdate_tool valve to False, it will still check for updates and notify the AI.
• You can change the URL where updates come from (for example, to your own github file rather than mine.)
• You can adjust the interval at which updates are checked by setting the version_check_interval valve, set to -1 to disable checking altogether.
• By default, auto-update runs at the end of a tool call - your initial tool call will be run with the current version, then the next call will be on the new version.
  As this tool is new, and issues are expected, it is recommended to leave auto-update enabled, so that you get any fixes asap.