Model Selection

To choose the right LLM model for each agent spawn automatically, apply a deterministic 5-layer resolution hierarchy. This evaluates per-agent config overrides, global defaults, session directives, charter preferences, and task-aware heuristics to select the optimal model.

The model selection hierarchy works by evaluating configuration layers in a fixed priority order: per-agent config first, then global default, session directive, charter preference, and task-aware auto heuristics, falling back to a default model if no preference is found.

To configure per-agent model overrides for different tasks, define specific models in your configuration. The resolution process checks these per-agent overrides before global defaults, ensuring coding, planning, and data analysis agents use specialized models automatically.

Resolved LLM model preferences are persisted to the .squad/config.json file. Storing the chosen model in this configuration file ensures cross-session consistency and provides traceable spawns for auditing and debugging agent behavior.

Yes, you can see which model was selected when an AI agent spawns. The resolved model is included in the agent spawn acknowledgment, providing traceable spawns that aid in auditing and debugging the deterministic model selection process.

If no per-agent config or global default is set for model selection, the resolution hierarchy applies task-aware heuristics to auto-select a model. If all higher-priority layers are absent, it ultimately falls back to a predefined fallback model.