Model Catalog

What Is a Model Catalog

A model catalog is a curated, structured list of large language models that a platform makes available for users to choose from, typically including metadata such as the model's provider, context window size, capabilities, and cost. Instead of hardcoding a single model into an application, a model catalog lets the choice of model become a configuration option.

What It Typically Includes

• Model identifiers: the specific name or version string a provider's API expects for a given model.
• Provider information: which company or service hosts the model and how it is authenticated.
• Capability metadata: details such as context window size, support for tool use, or support for structured output.
• Cost and rate information: pricing per token or per request, and any relevant usage limits.

Why It Matters

Model quality, cost, and behavior change frequently as providers release new versions and retire old ones, so a static, hardcoded model choice tends to age poorly. A model catalog decouples an application's logic from any single model, letting an operator or user swap models as better or cheaper options become available, without changing application code. It also supports matching models to tasks: a fast, inexpensive model may be sufficient for simple classification, while a more capable model may be warranted for complex reasoning or long-context tasks.

Model Catalog in Multi-Agent Platforms

In a platform that runs many independent agents, a model catalog is what allows each agent to be configured with a different model suited to its role, for example a lightweight model for a monitoring agent and a more capable model for one that writes code. Agenhood exposes a model catalog so that, when provisioning an agent, a user can pick the model it should use independently of which driver, meaning execution engine, the agent runs on. Keeping the catalog current, reflecting new model releases and deprecations, is an ongoing maintenance task for any platform that depends on it.

Model Catalog vs Model Provider

A model provider is the company or service that trains and hosts a model, while a model catalog is the list an application builds on top of one or more providers to present available choices to a user. A single model catalog can span several providers at once, letting a user pick, for example, a model from one provider for everyday tasks and a model from a different provider for tasks that need a longer context window or stronger reasoning. This separation also makes it easier for a platform to add support for a new provider or model without redesigning how models are selected elsewhere in the system.