Your openclaw service logs show it cannot reach its configured Ollama provider (qwen2.5-coder:32b), with every request failing with a "fetch failed" error. Your Ollama service is in a separate project from your openclaw service, which means they cannot communicate over the private network. You would need to ensure your Ollama service has a public domain and that your openclaw service's Ollama URL is configured to use that public domain.

Hi, no that is a separate project. I have not joined my ollama service to my openclaw yet as there was still connection issues between them. I have been using openrouter/ollama and all was fine, then poof this happened by itself.

This openclaw instance is connected directly to openrouter.

Hello simplyict,

the prblm is most likely on the openrouter side not railway , check two things ; your openrouter credit balance at openrouter.ai/settings/credits, and whether your api key is still valid at openrouter.ai/keys. if both look fine , check openrouter's status page for any ongoing issues , nothing changed on your end so the problem came from outside your setup

Hope this help you :)

Hey, did you by any chance run out of openrouter credits? or maybe the provider went down? i recommend using groq for llama models, its extremly fast and very cheap

ok so yes plenety of creddit there. thats not the normal message that comes up if I run out of credit. It tells me I ran out of credit in red. 🙂 and operouter API is still valid. last used 8 hours ago when the first issue happened. I never changed anything at all I agree. one minute it was replying and the next minute it wsn't with the LLM message.

I was actually working with it on setting up Ollama and openclaw inhouse on 2 servers and trying to get openclaw around this totally weird control ui requires device identity issue when the system flaked out. Running in Railway and using openrouter has not been without its drawbacks for sure. Crashes, rebuilds, more crashes and more rebuilds. burning tokens. sO I wanted to establish a local lan setup inhouse. and thats when the railway claw went dark

use the railway cli to ssh into the service then try and curl openrouter api

check your railway environment variables , something may have accidentally switched from your openrouter endpoint to a local ollama address while you were setting up the inhouse servers , that would cause this error

yes I am guessing that he took it upon himself to think he needed to switch himself, not give examples of how to do this. I do find that half of hsi replies are nonsense most of the time. ok I went through every variable I could find. nothing seems to be wrong. all seem to be as it was. it doesn't show in the railway variables where it is getting the settings for the openrouter etc. if it was changed, there is no setting there.

i have cli entered the railway app and looked at the data/workspace folder for the json file. it is still showing it is pointing to openrouter. question here , why is it even with nano installed, I can't seem to move inside the json file so I couldn't edit it even if I tried lol

The openclaw dockerfile says its debian so you should be able to just run `apt install nano`

can you tell me what does the config file actually show for the provider url and api key?

i was saying that you should try and curl openrouter api, not access any config

run `curl -H 'Authorization: Bearer ' -X POST https://openrouter.ai/api/v1`

curl: (6) Could not resolve host: Bearer

curl: (3) URL rejected: Bad hostname

using my openrouter api key

I already installed nano, and ran nano openclaw.json. the file loads up but the cusror does not move. no matter how many times I ext and re open, close railway-cli and rep-open. i can not adjust anything. the only thing that works is control X to exit

dns issue maybe?