routr-ai

name	routr-ai
description	Build AI features with Vercel AI SDK (chat, agents, RAG, tools). Use when: chatbot, LLM, streaming, embeddings, tool calling, AI SDK, RAG. Not for: multi-agent architecture or eval harness (→ routr-agents); chat UI styling only (→ routr-frontend).

routr-ai

Goal: production AI features — streaming, tools, providers.

When to activate

• Chatbot, LLM feature, streaming, embeddings, tool calling, RAG
• "AI SDK", provider setup, generateText/streamText/useChat

Do not activate

• Multi-agent architecture or eval harness → routr-agents
• Chat UI styling only, logic unchanged → routr-frontend

Iron law

Never put an API key in client-side code — server actions or route handlers only.

0. Bootstrap

See routr-catalog/references/skill-registry.md:

Order	Skill	Role
1	ai-sdk	generateText, streamText, useChat, tools
2	migrate-ai-sdk-v6-to-v7	SDK upgrades
3	find-docs	Provider API docs
4	routr-agents	Multi-agent architecture
5	routr-integrate	Provider setup

Chat UI → pair with routr-frontend.

1. Patterns

Pattern	Stack
Chat UI	useChat + streamText + routr-frontend
Agents + tools	ToolLoopAgent + tool-design
RAG	embeddings + vector store

2. Safety

API keys server-side; rate limits; prompt injection awareness.

Output format

## AI feature report
**Pattern:** chat UI / agent+tools / RAG
**Provider / model:** …
**Safety:** keys server-side, rate limits, injection awareness confirmed

Handoff

• SDK upgrade errors → migrate-ai-sdk-v6-to-v7
• Stream bug → routr-debug
• Multi-agent → routr-agents
• Chat UI polish → routr-frontend
• Deploy → routr-deploy
• Marketing → routr-marketing

References

• boundaries

Anti-patterns

• Client-side API keys
• Production agents without routr-agents