Show HN: 100-Line LLM Framework (github.com)

Pocket Flow is a new minimalist LLM framework that fits its core abstraction—the “Graph”—into ~100 lines of code (≈56KB) with zero dependencies and no vendor lock‑in. It’s available via pip or as a one‑file copy and now has ports in TypeScript, Java, C++, Go, Rust and PHP. The project emphasizes agentic coding (agents building agents) for rapid LLM app development and ships many example flows and tutorials (chat, RAG, streaming, memory, map‑reduce, multi‑agent, Model Context Protocol, A2A, etc.) to show common patterns in a tiny API surface. Significance: Pocket Flow is pitched as an educational and rapid‑prototyping alternative to heavyweight frameworks like LangChain, CrewAI and AutoGen, arguing that most production use‑cases can be expressed with a small graph abstraction instead of large feature sets and vendor wrappers. That makes it easy to audit, extend, and reason about behavior, and lowers the barrier for experimentation and agentic workflows. Tradeoffs are implicit: the minimal core lacks built‑in integrations and tooling present in larger frameworks, so production deployments will require wiring to stores, tools and providers manually. For researchers and developers, Pocket Flow is a compact template to prototype LLM patterns, study agent architectures, and iterate quickly without heavy dependencies.