Feature Request: Per-Mode and Per-Agent LLM Routing

Feature Description

Mode-Specific Model Routing: Users can define a default model for each of Command Code's core operational modes.

• Planning Mode (planning): Maps to high-parameter, reasoning-heavy models (e.g., Deepseek V4 Pro, GLM 5.1) for complex architectural design, system breakdown, and step-by-step logic generation.
• Edit Mode (edit): Maps to fast, highly capable coding models optimized for syntax and refactoring to execute the plan quickly.
• Auto-Accept Mode (auto-accept): Maps to ultra-fast, low-latency models for rapid, low-risk changes (e.g., fixing linting errors, adding docstrings) where developer oversight is bypassed.
• YOLO Mode (yolo): Maps to high-context, highly capable models authorized to make sweeping, project-wide changes with deep contextual understanding.
• Custom Agent Model Binding: Users creating specialized, domain-specific agents within Command Code can lock a specific model to that agent.
• CLI Flag Overrides: The ability to temporarily override the default model for a specific session or command using a flag (e.g., command-code --plan --model deepseek-v4-pro).

Use Case

Explore-Plan-Execute Workflow

• Launching off multiple agents to explore legacy repo, each using lower cost and faster agents with big context windows
• Planning with SOTA OSS models with high reasoning capabilities
• Execute plans proposed by the Planning model using a cost effective model

Additional Context

No response

How important is this to you?

Important for my workflow

Commencez par retracer comment Command Code sélectionne actuellement les modèles pour la planification, l’édition, auto-accept, YOLO et les agents personnalisés, puis examinez l’analyse des options de CLI pour une surcharge temporaire du modèle. Le travail est considéré comme terminé lorsque les valeurs par défaut des modes, les associations par agent et la surcharge d’exemple avec --model fonctionnent sans perturber le workflow Explore-Plan-Execute.