feat: add per-agent model configuration support

[laststylebender14]

Summary

Add per-agent model configuration support, allowing users to set a specific model for each agent via CLI commands rather than using only global defaults.

Context

Currently, users can only set a global default model for the entire application. This PR enables per-agent model overrides, giving more flexibility when working with multiple agents that may require different models (e.g., o1 for reasoning tasks, GPT-4 for creative tasks).

Changes

• Added AgentModelConfig domain type to store both provider and model per agent
• Extended AppConfig with new agent_model HashMap for per-agent overrides
• Added set_agent_model and get_agent_model service methods
• Updated AgentProviderResolver to resolve models with priority: agent config > active agent YAML > global default
• Added set-agent-model CLI command to configure per-agent models

Key Implementation Details

Model IDs are scoped to a specific provider's namespace, so both provider and model are stored together in AgentModelConfig. The resolution priority is:

1. Per-agent config (from app config)
2. Active agent's model (from agent YAML definition)
3. Global provider model (from app config)

Use Cases

• Configure "reasoning" agent to always use o1 model
• Set "creative" agent to use GPT-4 for specific tasks
• Override model for agents defined in YAML without modifying the YAML files

Testing

# Set model for a specific agent
forge config set-agent-model my-agent provider model

# Verify the setting is persisted
forge config get-agent-model my-agent

# Test that agent uses the configured model
forge chat --agent my-agent

Links

• Related: Enables flexible multi-model agent workflows

[github-actions]

Action required: PR inactive for 5 days.
Status update or closure in 10 days.