arya-model-router
bởi staratheris
AI & LLMsTóm tắt
Token-saver router: elige modelo (cheap/default/pro) y usa sub-agentes para tareas pesadas.
Mô tả
Kỹ năng này làm gì
Routes each task to one of three OpenAI model tiers based on complexity, keeping routine interactions on cheaper models and escalating to stronger ones only when needed. Before passing heavy tasks to a pro model, it can compress the context into a brief to avoid large token transfers. Model tiers and routing rules are configurable via a JSON file.
Eliminates manual model switching by applying routing rules automatically, so you pay for stronger models only when the task justifies it.
Khi nào nên dùng
- Keeping daily assistant costs low by defaulting to gpt-4o-mini
- Auto-routing a complex debugging session to gpt-4.1
- Running a long document analysis in a sub-agent without bloating main context
- Forcing cheap mode explicitly for quick lookup tasks
- Pre-briefing a pro model before handing off a large codebase review
Quy trình ví dụ
User asks: analyze this codebase for security vulnerabilities
- 1
router.py classifies the task using rules.json and assigns it to the pro tier
- 2
main agent compresses relevant context into a brief
- 3
sub-agent is spawned with gpt-4.1 and receives the brief instead of full context
- 4
sub-agent performs the security analysis
- 5
results are returned to the main agent
Security vulnerability report generated by gpt-4.1, with minimal token cost to the main agent
Yêu cầu
- OpenAI API key
Cài đặt
clawhub install staratheris/arya-model-routerChạy lệnh CLI, hoặc dán prompt vào Claude Code, Cursor, Codex hay coding agent bạn đang dùng.