Custom commands fail when (even unused) provider is in config file #2205

Activity

1. github-actions commented on Aug 23, 2025

   

   github-actionsBot

   on Aug 23, 2025 – with GitHub Actions

   Contributor

   More actions

   This issue might be a duplicate of existing issues. Please check:

   • Executing commands using opencode run does not work #2200: Similar issue with executing commands using opencode run - although this was about commands not working, it deals with custom command execution failures
   • Error POSTing to endpoint (HTTP 400): jsonrpc: 2.0, error: code 3200, Session not found or expired #2061: Connection/session errors with HTTP 400 - deals with connection issues to endpoints
   • Agent seems unable to do anything using together.ai #2169: Agent unable to do anything with together.ai provider - similar connection/execution issues where commands appear to work but fail to connect
   • Error: Failed To Open Session #2046: Failed to open session errors - related to session connection failures
   • 500 Internal Server Error when sending tool_calls array in request body. (Open AI compatible api) #1908: 500 Internal Server Error with OpenAI compatible APIs - API connection issues

   Your specific error shows a ConnectionRefused to localhost:1234, which suggests you may be trying to connect to a local LLM server (like LM Studio, Ollama, etc.) that isn't running or accessible. Please check if your local server is running on that port.

   Feel free to ignore if none of these address your specific case.
2. rekram1-node

   assigned

   thdxr

   on Aug 23, 2025
3. aspiers commented on Aug 23, 2025

   

   aspiers

   on Aug 23, 2025

   ContributorAuthor

   More actions

   I do have LMStudio configured in my providers section, but this problem manifests even when I'm using Anthropic / sonic / GPT-5, so I don't think the localhost:1234 error is relevant.
4. aspiers commented on Aug 23, 2025

   

   aspiers

   on Aug 23, 2025

   ContributorAuthor

   More actions

   Ah! Sorry, looks like it's relevant after all. If I remove the providers section then custom commands start working again. But everything else was working even with the reference to a non-running LMStudio on port 1234, so this is a bug. It should not be necessary to add an LMStudio entry to providers every time you want to use it and remove it every time you shut LMStudio down.
5. aspiers

   changed the title [-]Running custom command causes "Error: Unable to connect. Is the computer able to access the url?"[/-] [+]Custom commands fail when (even unused) provider is in config file[/+] on Aug 23, 2025
6. aspiers commented on Aug 23, 2025

   

   aspiers

   on Aug 23, 2025

   ContributorAuthor

   More actions

   Issue title updated. Just to clarify, I had this fragment in opencode.json which comes directly from the docs:

     "provider": {
       "lmstudio": {
         "npm": "@ai-sdk/openai-compatible",
         "name": "LM Studio (local)",
         "options": {
           "baseURL": "http://127.0.0.1:1234/v1"
         },
         "models": {
           "google/gemma-3n-e4b": {
             "name": "Gemma 3n-e4b (local)"
           }
         }
       }
     }
   

   If I start up LM Studio and load the liquid/lfm2-1.2b model, even with opencode configured to use sonic, shockingly it attempts to use that local model (and not even Gemma which is the one in the config file!). So it looks like there is something really weird with how custom commands invoke LLM providers.
7. aspiers

   added a commit that references this issue on Aug 23, 2025

   opencode: work around anomalyco/opencode#2205

   ...

   cfc5c98
8. rekram1-node commented on Dec 28, 2025

   

   rekram1-node

   on Dec 28, 2025

   Contributor

   More actions

   [automated] Closing due to 90+ days of inactivity. Feel free to reopen if you still need this!
9. rekram1-node

   closed this as completedon Dec 28, 2025