OpenCode

OpenCode works with any OpenAI-compatible API, so it can use your Paddler cluster as its model provider.

Start the balancer with the OpenAI-compatible service

OpenCode connects through Paddler's OpenAI compatibility service. It is not enabled by default, so make sure to start the balancer with the --compat-openai-addr parameter:

paddler balancer \
    --compat-openai-addr 127.0.0.1:8070 \
    --inference-addr 127.0.0.1:8061 \
    --management-addr 127.0.0.1:8060

Load a model that supports tool calling

Load a model that supports function calling. OpenCode edits files through tool calls, so without it, you'd only get descriptions of the changes.

Configure OpenCode

Paddler is not one of OpenCode's built-in providers, so you need to add it as a custom one. Below is an example of an OpenCode config; you can put it in opencode.json in your project root, or in ~/.config/opencode/opencode.json if you want it available everywhere:

{
  "$schema": "https://opencode.ai/config.json",
  "model": "paddler/local",
  "provider": {
    "paddler": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Paddler",
      "options": {
        "baseURL": "http://127.0.0.1:8070/v1"
      },
      "models": {
        "local": {
          "name": "Qwen 3.6"
        }
      }
    }
  }
}

A few things about this configuration:

• baseURL is the address you passed to --compat-openai-addr, with /v1 at the end.

• The model id is only a label. Paddler ignores the model name in a request and routes it to whatever model your cluster currently has loaded, so it does not need to change when you swap models.

You can find more information about adding a custom provider in OpenCode's documentation.

Run OpenCode

Finally, run opencode in your project. Paddler will show up as a provider, with the models you declared in the configuration, for example: