model-selection

Model selection for AI agents uses a 4-layer hierarchy—User Override, Charter Preference, Task-Aware Auto-Selection, and Default—to match model capability and cost to specific agent task requirements.

A robust fallback strategy defines alternative models in a chain, so if a preferred model like claude-sonnet-4.5 is unavailable, the agent automatically degrades to an alternative to ensure task completion.

Task complexity adjustment maps specific agent roles to appropriate models, automatically selecting high-quality models for coding tasks while defaulting to cost-effective models for simple logging operations.

Yes, the User Override layer sits at the top of the 4-layer hierarchy, allowing you to manually specify a model and bypass Charter Preference, Task-Aware Auto-Selection, and Default selections for specific agent tasks.

Configure model tiers by categorizing available AI models into premium, standard, and fast/cheap tiers, then mapping agent tasks to the appropriate tier based on complexity and cost-effectiveness requirements.

When a selected AI model is unavailable, defined fallback chains engage to intelligently route the agent task to an alternative model, ensuring task completion without manual intervention or failure.