Hey @rasmusson,

We have created an internal ticket to look into this which we will be tracking as "GHC-244"

Hey @rasmusson,

I suspect this is not specific to the Langchain nodes and is likely linked to another general issue that we have open where Axios the http package we use doesn't pick up system defined proxies correctly if it is an HTTP proxy due to a failure on the CONNECT method.

Ok. As I understand the axios issue resulted in bad gateway error. In my case in the langchain nodes I dont get any error, the request just go straight to internet without going through the proxy. The node works just fine if I allow direct internet access but keep the system defined proxy.

Hey @rasmusson,

That sounds a bit different, How are you setting the system proxy? Assuming http_proxy and https_proxy are being set it should still work unless Langchain itself uses an http library that doesn't respect system set proxies.

yes, http_proxy and https_proxy env vars. Im using information extractor together with open AI model

Perfect, I have moved this to our AI team to investigate. It looks like Langchain may have an issue with proxies.

Same issue here. Disabling http_proxy and https_proxy on an internet connected host proves the AI nodes fail when it's set. Unfortunately not an option for hosts that require a proxy

This seems even more broken in recent versions:
In 1.50.2 I can at least run AI model nodes with a firewall rule bypassing the proxy 😢 (the rest of the traffic correctly goes via the proxy)

However in 1.69.2 the AI Agent never seems to call the chat model, it will sit there indefinitely. Even debug logging gives no indication why. I've tried adding the model base url to no_proxy but it still fails. The only way it will work is by removing http_proxy and https_proxy entirely... which means we can have AI and nothing else, or everything else but not AI 😖

Debug logging from 1.69.2

I confirm. Even HTTP Request node doesn't work with globally set proxy:

    environment:
...
      - HTTP_PROXY=http://v2raya:20171
      - HTTPS_PROXY=http://v2raya:20171

I get:

Service unavailable - try again later or consider setting this node to retry automatically (in the node settings)
503 - ""

n8n version: 1.74.3

@alparo that doesn't seem like an error reaching the proxy. Confirming I'm on 1.74.3 with these in the environment

http_proxy=http://proxy.example.com:3128
https_proxy=http://proxy.example.com:3128

And my normal HTTP requests are going via the proxy successfully... albeit not using HTTP CONNECT (another axios issue)

Currently v1.80.5 still repro this issue.
I use Charles as Proxy to inspect the payload, google mail node can work via Charles proxy, but ai agent node can't.
Both nodes can get data correctly, just the later one cannot using the proxy.
Through Charles, I can see that google mail api is called through User Agent AXIOS, it hinted that ai agent may not use the same http connection tunnel.

I investigated through the source code, confirm that open ai node does NOT respect any http_proxy setting, when it call langchain's openai client.
The core reason happens, because HTTP "fetch" api does not respect http_proxy. It is required to set the Proxy Agent explicitly. Other HTTP nodes works because they are using AXIOS for HTTP invoking instead.

I have fixed this issue in this commit, which is appled for a pull.