Custom LLM providers

Customize providers to connect another LLM service, use separate credentials, or choose the models and token limits available to your agents.

Add these definitions to ~/.skiller/settings/providers.json or the file selected by SKILLER_AGENT_PROVIDERS_FILE. Each new provider has its own name and declares a supported adapter.

Custom OpenAI-compatible provider

Connect a service that supports the OpenAI Chat Completions API:

{
  "providers": {
    "my-provider": {
      "adapter": "openai",
      "enabled": true,
      "base_url": "https://llm.example.com/v1",
      "api_key_env": "MY_PROVIDER_API_KEY",
      "timeout_seconds": 60,
      "temperature": 1,
      "top_p": 1,
      "parallel_tool_calls": true,
      "tool_choice": "auto",
      "options": {},
      "models": [
        {
          "model": "my-model",
          "context_window_tokens": 128000,
          "max_output_tokens": 4096
        }
      ]
    }
  }
}

See OpenAI adapter for fields, credentials, and provider options.

Custom Codex provider

Define a separate Codex entry with its own model list. Use an existing OAuth credentials file created through authentication:

{
  "providers": {
    "my-codex": {
      "adapter": "codex",
      "enabled": true,
      "credentials_file": "~/.skiller/secrets/openai-codex.json",
      "timeout_seconds": 120,
      "parallel_tool_calls": true,
      "models": [
        {
          "model": "gpt-5.4",
          "context_window_tokens": 1050000
        }
      ]
    }
  }
}

See Codex adapter for fields, OAuth credentials, and model limitations.

Custom Bedrock provider

Use a named AWS profile and a Bedrock model or inference profile available to that profile:

{
  "providers": {
    "my-bedrock": {
      "adapter": "bedrock",
      "enabled": true,
      "profile": "my-aws-profile",
      "timeout_seconds": 45,
      "models": [
        {
          "model": "us.anthropic.claude-opus-4-6-v1",
          "context_window_tokens": 1000000,
          "max_output_tokens": 4096
        }
      ]
    }
  }
}

See Bedrock adapter for fields, AWS profiles, and model access.

Select the provider

Select the provider and model in your agent configuration:

{
  "llm": {
    "provider": "my-provider",
    "model": "my-model"
  }
}