Providers: Abacus connectivity error #6836

Activity

1. lucaschallamel

   added

   bugSomething isn't workingSomething isn't working

   on Jan 4, 2026
2. github-actions

   assigned

   rekram1-node

   on Jan 4, 2026
3. github-actions commented on Jan 4, 2026

   

   github-actionsBot

   on Jan 4, 2026 – with GitHub Actions

   Contributor

   More actions

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

   • Attempting to use LM Studio with OpenCode - TypeError [ERR_INVALID_URL]: "undefined/chat/completions" cannot be parsed as URL. #195: Attempting to use custom provider with OpenAI-compatible endpoint - similar error pattern with undefined endpoint in URL
   • Custom OpenAI-compatible provider options not being passed to API calls #5674: Custom OpenAI-compatible provider options not being passed to API calls - configuration options not reaching API calls
   • Custom OpenAI-compatible provider returns no text content #5210: Custom OpenAI-compatible provider returns no text content - issues with custom endpoint configurations

   The "Method chat/completions/chat/completions not found" error suggests the endpoint path may be duplicated in the request, which could indicate a configuration issue with how the Abacus provider URL is being constructed.

   Feel free to ignore if none of these address your specific case.

   👍React with 👍1Reacted by Ofir
4. dantoac commented on Jan 10, 2026

   

   dantoac

   on Jan 10, 2026

   Last edited by dantoac

   More actions

   Seems the BaseURL for the provider is wrong. You can overload it in .config/opencode.json

   Example:

   { "provider": {
         "Abacus": {
   	    "npm": "@ai-sdk/openai-compatible",
   	    "name": "Abacus AI",
   	    "options": {
   		"baseURL": "https://routellm.abacus.ai/v1/"
   	    },
   	    "models": {
   		"gpt-5": {
   		    "name": "GPT-5"
   		},
   	    }
   	}
   }
   

   EDIT: The expected structure seems to be different, as some models seems to work only a few times after some retries. I just tested GPT-5, Claude and route-llm in Plan mode.

   I can confirm that GLM-4.7 worked at first try, so it should be someting regarding the structure accepted by abacus.

   FWIW, in https://abacus.ai/app/route-llm-apis says:

   Parameters
   
   Our API follows the [OpenAI Chat Completions API spec](https://platform.openai.com/docs/api-reference/chat/create), with a few limitations outlined below:
   
   • We only support the following parameters:
   
       model
       messages
       max_tokens
       stream
       temperature
       stop
       n
       presence_penalty
       frequency_penalty
       logprobs (not supported for gpt-5)
       response_format (only {"type": "json"})
   
5. dantoac commented on Jan 10, 2026

   

   dantoac

   on Jan 10, 2026

   More actions

   GLM-4.7 responded with "Bad Request: Invalid role "tool" in messages. Must be one of user, assistant, or system.", so confirming the missing parameter.
6. meirm commented on Jan 10, 2026

   

   meirm

   on Jan 10, 2026

   Last edited by meirm

   More actions

   This worked for me

   cat .config/opencode/opencode.json

   {
     "provider": {
       "abacus": {
         "name": "Abacus AI",
         "npm": "@ai-sdk/openai-compatible",
         "models": {
           "gpt-5": {
             "name": "GPT-5"
           }
         },
         "options": {
           "baseURL": "https://routellm.abacus.ai/v1/"
         }
       }
     },
     "$schema": "https://opencode.ai/config.json"
   }
   

   ❤️React with ❤️1Reacted by EAH9TL
7. InSelfControll commented on Jan 13, 2026

   

   InSelfControll

   on Jan 13, 2026

   More actions

   how can I make it work with all the models of abacus ai ?
   Currently the main issue that it sends this twice: Not Found: Method chat/completions/chat/completions not found.
   Looks like someone messed up in the app code and wrote twice /chat/compilations in the request and that's why it fails.
8. dantoac commented on Jan 15, 2026

   

   dantoac

   on Jan 15, 2026

   More actions

   how can I make it work with all the models of abacus ai ? Currently the main issue that it sends this twice: Not Found: Method chat/completions/chat/completions not found. Looks like someone messed up in the app code and wrote twice /chat/compilations in the request and that's why it fails.

   Hi, you need to use the "baseURL": "https://routellm.abacus.ai/v1/" solution proposed above, in your ~/.config/opencode/opencode.json
9. rekram1-node commented on Jan 15, 2026

   

   rekram1-node

   on Jan 15, 2026

   Collaborator

   More actions

   Btw models.dev is the models api we source from and maintain, PRs to fix or update or add are always welcome.

   👍React with 👍1Reacted by Daniel Aguayo
10. rekram1-node commented on Jan 15, 2026

    

    rekram1-node

    on Jan 15, 2026

    Collaborator

    More actions

    Thanks for providing work around tho @dantoac ill update models.dev

    👍React with 👍1Reacted by Daniel Aguayo
11. rekram1-node commented on Jan 15, 2026

    

    rekram1-node

    on Jan 15, 2026

    Collaborator

    More actions

    you should be able to remove it from your config if you run:
    opencode models --refresh

    And then rm from cfg

    And then try again

    Caches refresh automatically but this will force it

    🚀React with 🚀1Reacted by Daniel Aguayo
12. rekram1-node

    closed this as completedon Jan 15, 2026