Add a custom providerto opencode

Where the file goes

• opencode.json at the root of a project applies to that project.
• .opencode/opencode.json inside a project does the same and wins over the file at the root.
• ~/.config/opencode/opencode.json applies to every project, and a project file overrides it.

Both .json and .jsonc are read, so comments are allowed if you name the file opencode.jsonc.

A minimal block

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "uncensored-opencode": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Uncensored Opencode",
      "options": {
        "baseURL": "https://uncensoredopencode.com/v1",
        "apiKey": "{env:UNCENSORED_OPENCODE_API_KEY}"
      },
      "models": {
        "glm-5.3-flash-uncensored": {
          "name": "GLM 5.3 Flash Uncensored",
          "tool_call": true,
          "reasoning": true,
          "attachment": true,
          "limit": { "context": 1048576, "output": 32000 }
        }
      }
    }
  },
  "model": "uncensored-opencode/glm-5.3-flash-uncensored"
}

Field by field

• The key under provider, here uncensored-opencode, is the provider id. It is the part before the slash whenever you name a model, as in uncensored-opencode/glm-5.3-flash-uncensored.
• npm picks the adapter. Use @ai-sdk/openai-compatible for an endpoint that serves /v1/chat/completions.
• name is what the provider is called in the /models list.
• options.baseURL is the endpoint up to and including /v1, with no trailing slash and without /chat/completions.
• options.apiKey is the key. Write {env:UNCENSORED_OPENCODE_API_KEY} to read it from an environment variable or {file:path} to read it from a file, so the key itself is never in the config.
• models is a map. Each key is the exact model id the endpoint expects, and opencode only offers the models you list.
• limit.context and limit.output are token limits. If you give a limit, give both.
• tool_call, reasoning and attachment say what the model can do: call tools, think before answering, and take images.

Check it loaded

opencode models uncensored-opencode

That prints one line per model, in the form uncensored-opencode/glm-5.3-flash-uncensored. If it prints nothing, opencode did not read your block.

Mistakes that hide a model

• The file is valid JSON but in the wrong place. Run opencode from the folder that holds opencode.json, or use the global file.
• The model id is close but not exact. Ids are case sensitive and the endpoint lists the real ones at https://uncensoredopencode.com/v1/models.
• baseURL ends in /chat/completions. opencode adds that itself, so the request goes to a path that does not exist.
• The environment variable is not set in the shell that started opencode. An unset variable becomes an empty key and the endpoint answers 401.
• The top-level model names a model that is not in the models map.
• limit has a context figure but no output figure.

Every model on this API

The opencode page has the block with all 7 models filled in.

Questions