Connect to a Model
A model target lets Red send chat-completion requests directly to a foundation model, without requiring you to set up an API endpoint. If you need to test your own API or agent instead, see Connect to your agent.
When to use a model target
Use a model target when you want to test a foundation model directly, without connecting Red to your own API or agent endpoint.
Configure a model target
- When creating a target, under Target configuration, select the Model tab.
- Model – Search or select a model from the list of available models (e.g.
openai/gpt-4o,anthropic/claude-3-5-sonnet). The form also suggests popular models to get you started quickly. - System prompt (optional) – Add your system prompt to test the foundation model with your instructions, without deploying an endpoint.
- In Advanced settings (optional):
- Additional fields – JSON object merged into each request (e.g.
temperature, application-specific-parameters). - Max concurrent requests – Number of concurrent requests Red sends to the model (default 250; lower for rate-limited or self-hosted setups).
- Additional fields – JSON object merged into each request (e.g.
Next steps
- Creating a Target – overview of targets and connection types.
- Connect to your agent – connect Red to your own API or agent instead.
