10. Tool Selection by Use Case

You are a Claude Code user (individual developer)

Problem	Tool
Command outputs flooding context	RTK
File reads consuming most of context budget	tilth, lean-ctx, or Token Savior
Monitoring token spend	ccusage (see Third-Party Tools)
Context growing too long in a session	/compact at 70% usage
Forgetting past session decisions	ICM memory system
Claude ignoring rules from long CLAUDE.md	Path-scoping (see context engineering guide)

You are building an AI application

Problem	Tool
Tool output JSON too verbose	Headroom
Prompts too long, need compression	LLMLingua
Routing across multiple LLM providers	Portkey
Compression + guardrails at the edge	Edgee
RAG chunks losing context	Anthropic Contextual Retrieval
Tracing agent execution	Langfuse or LangSmith
RAG quality measurement	Arize Phoenix
Continuous evaluation	Maxim AI

You are deploying a self-hosted LLM

Problem	Tool
GPU memory efficiency	vLLM (PagedAttention)
Shared-prefix caching (multi-agent, RAG)	SGLang (RadixAttention)
Caching repeated queries semantically	Redis with vector search