Reusable subagent routing and execution helpers for TanStack AI apps.

This package is built around one production opinion: finite subagent routing should be deterministic by default. If the possible outputs are known — answer directly, use tools, write a plan, spawn one specialist, spawn multiple specialists, or refuse/escalate — routing is classification, not a task that needs another LLM call.

• Score-based intent routing with routeSubagentRequest(...)
• App-configurable routing policies with createSubagentRouter(...)
• TanStack AI tool factories for route_subagents and run_subagents
• Bounded worker validation and fanout
• App-defined tools, profiles, models, prompts, tracing, persistence, and UI
• Lifecycle callbacks and background run handles

npm install @5queezer/tanstack-ai-subagents @tanstack/ai zod

@tanstack/ai and zod are peer dependencies.

Verified on commit 7cf49a3:

npm test
# 12 tests, 12 pass
npm pack --dry-run
# includes dist/, README.md, LICENSE, and demo/routing-in-action.cast

• Routing Policy
• TanStack AI Integration
• Testing and Proof
• Design Boundaries

• GitHub: https://github.com/5queezer/tanstack-ai-subagents
• npm: https://www.npmjs.com/package/@5queezer/tanstack-ai-subagents