opencode.json connecting to Azure AI Foundry models

Azure AI and opencode.md

Adding Azure AI Foundry Models to opencode

This took me a while to get going to pass the "api-version": "2025-01-01-preview" and then I kept hitting MAX TOKEN LENGTH issues when using a smaller model.

1. Configure the provider in your opencode.json file:

   "I did this is my WSL instance"

   mkdir -p ~/.config/opencode/
   touch ~/.config/opencode/opencode.json
   code ~/.config/opencode/opencode.json

   {
       "$schema": "https://opencode.ai/config.json",
       "provider": {
           "azure-foundry": {
               "npm": "@ai-sdk/openai-compatible",
               "name": "Azure Foundry",
               "options": {
                   "baseURL": "https://<MY_AI_FOUNDRY_INSTANCE>.cognitiveservices.azure.com/openai/deployments/<MY_DEPLOYMENT_NAME>/",
                   "queryParams": {
                       "api-version": "2025-01-01-preview"
                   }
               },
               "models": {
                   "MY_MODEL": {
                       "name": "<CUSTOM_MODEL_DISPLAY_NAME_IN_OPENCODE>"
                   }
               }
           }
       }
   }

2. To use this configuration in my devcontainer.json I add a mount to map this file.

   {
      ....
      ,
      "mounts": [
         "source=${localEnv:HOME}/.config/opencode,target=/home/node/.config/opencode,type=bind",
         ......
      ],
      ......
   }

3. Assuming you have opencode installed npm install -g opencode-ai in your devcontainer run opencode auth login
   • Arrow key up to other
   • enter azure-foundry
   • enter YOUR_SECURE_API_TOKEN
   • Run /models in opencode to select your Azure AI Foundry model

🎉 Happy coding!

If anyone can figure out how to limit the max tokens sent to a model in opencode please let me know.
I want to play with some of the smaller models locally.

possibly after you download OLLAMA on your machine and get a MCP client (Claude) seems to be good to start with and dont have to buy a subscription, add the following and then you have something to try your models on locally

"ollama": {
"command": "uv",
"args": [
"--directory",
"C:/Source/MCP/PythonServer/weather",
"run",
"ollama.py"
]
},