Orchestra Mode (Multi-Model Orchestration)
Concept
Multiple AI models collaborate as a team:
- Chief Model analyzes user request, breaks into subtasks
- Expert Roles execute tasks in parallel (frontend, backend, bug_fixer, etc.)
- Chief Model synthesizes results into a single coherent response
Built-in Roles
| Role | Default Model | Purpose |
|---|---|---|
| Planner | Claude | Software architecture, task decomposition |
| Frontend | Grok | UI development |
| Backend | GPT-4o | API/server logic |
| Bug Fixer | Gemini | Debugging, root cause analysis |
| Reviewer | Claude | Code quality review |
| Security | GPT-4o | Security auditing |
| DevOps | Grok | Infrastructure/deploy |
| General | GPT-4o | General-purpose fallback |
Execution Model
- Parallel Tasks: Independent tasks run concurrently (goroutines + WaitGroup)
- Sequential Tasks: Dependency resolution via
depends_onfield - Retry: Rate-limit aware retry with exponential backoff (up to 3 retries)
- Streaming: Progress updates streamed per phase (plan → execute → synthesize)
Frontend Controls
- Settings Tab: Enable/disable, configure chief model, assign models to roles
- Config Dialog: Role editor with model selection, system prompt editing, custom role support
- Slash Command:
/orchestra on,/orchestra off,/orchestra config,/orchestra status