How to set a local provider in opencode TUI?

[TangLaoya]

Question

Dear all,

I noticed that opencode can set a local provider:
https://opencode.ai/docs/providers/#lm-studio

where should I put the file opencode.json, is the file correct?
{ "$schema": "https://opencode.ai/config.json", "provider": { "lmstudio": { "npm": "@ai-sdk/openai-compatible", "name": "LM Studio (local)", "options": { "baseURL": "http://localhost:8080/v1" }, "models": { "GLM-4.7-Flash-UD-Q5_K_XL.gguf": { "name": "GLM-4.7-Flash-UD-Q5_K_XL.gguf" } } } } }

Is there any entry to set the local provider from opencode TUI?

I use the Windows version (Windows Server 2019, latest version 1.1.35, OpenCode Desktop Installer.exe).

Thanks,
Tang Laoya

[github-actions]

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

• Local LLM Support on MAC with LM Studio - Without an API Provider Setup #1480: Local LLM Support on MAC with LM Studio - direct question about local provider setup with LM Studio
• LM Studio Missing as Provider, custom provider for LM Studio non-functional. #1555: LM Studio Missing as Provider - custom provider configuration for LM Studio
• How to configure a local large model #1669: How to configure a local large model - general local model configuration
• opencode.json/config.json search path #3407: opencode.json/config.json search path - addresses where to put configuration files
• Config in windows? #1235: Config in windows? - Windows-specific configuration question
• My configured ~/.opencode.json is immediately overwritten and reset upon launching opencode #1228: My configured ~/.opencode.json is immediately overwritten - opencode.json configuration persistence
• [BUG]: "Other" option missing for Custom Provider configuration in v1.1.25 #9071: 'Other' option missing for Custom Provider configuration - TUI provider configuration in v1.1.25
• Running OpenCode Desktop application with local model & censorship #6783: Running OpenCode Desktop application with local model - local model configuration
• OpenCode displays models which aren't configured & don't exist in LM Studio #4232: OpenCode displays models which aren't configured - LM Studio integration

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

[TangLaoya]

Thanks. I found this issue
#6783
is useful. I tried to put config.json in %USERPROFILE%.config\opencode folder:
{ "$schema": "https://opencode.ai/config.json", "provider": { "lmstudio": { "npm": "@ai-sdk/openai-compatible", "name": "LM Studio (Local)", "options": { "baseURL": "localhost:15432/v1" }, "models": { "GLM-4.7-Flash-UD-Q5_K_XL.gguf": { "name": "GLM-4.7-Flash-UD-Q5_K_XL.gguf", } } } } }

and start Opencode again. I found in the bottom of Opencode TUI appears 'Big Pickle', and the local models are all list here.

But after I selected 'GLM-4.7-Flash-UD-Q5_K_XL.gguf' from the list (the llama-server also launched this one) and type a question in the window, the opencode replied:

Error: Unable to connect. Is the computer able to access the url?

What should I do furthermore?

Thanks,
Tang Laoya

[justfortheloveof]

Try adding http:// (or https://) to your baseURL, make sure the port is accurate as well.

[R44VC0RP]

Thanks @justfortheloveof for the assist!

@TangLaoya - the fix is to add http:// to your baseURL:

"baseURL": "http://localhost:15432/v1"

The LM Studio docs show the correct format. Let us know if you hit any other issues!

[Velkez]

I try this yesterday, and I had the same error. I solve this just activating the local server on developer section on LM Studio. Then I restart Opencode and this work for me. Be sure if the BaseURL is the same that LM Studio is show you.

[TangLaoya]

Thanks. After set the correct BaseURL, the problem is solved.

[TangLaoya]

However, currently although the opencode is connected with the llama_server, the same question, the web by llama_server can give result after several minutes, but in the opencode, I have waited for more than 1 hours, no any response...

Is there still something wrong?

Thanks,
Tang Laoya

[Velkez]

However, currently although the opencode is connected with the llama_server, the same question, the web by llama_server can give result after several minutes, but in the opencode, I have waited for more than 1 hours, no any response...

Is there still something wrong?

Thanks, Tang Laoya

Hello Tang Laoya, thats because Opencode send a lot of context in tokens. If you have MCPs the number of tokens increments. A partial solution is disable the MCPs servers. In my case, in LM Studio, I send a simple greeting, and I had a response in seconds, but then I try the same thing in Opencode and I had a response in 4 minute, for a simple greeting. That's because Opencode send to the local model 18k of tokens of context approx. While in LM Studio send a few 42 tokens. The model had a lot of tokens to process in comparison

[TangLaoya]

That's great, thank you very much.
But how to disable the MCPs servers in opencode?

Thanks,
Tang Laoya