paulmorphic

FastAPI Langgraph AI Agent — AI skill for Claude Code

AI community

E-commerce AI Agent - Agentic RAG, HITL, Skills.

How to install FastAPI Langgraph AI Agent

This entry records only its repository, not the path inside it, so there is no exact command to give. Open paulmorphic/fastapi-langgraph-ai-agent and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What FastAPI Langgraph AI Agent does

E-commerce AI Agent - Agentic RAG, HITL, Skills.

Alternatives in AI

  • Prismor — Self-hosted runtime control plane for AI agents 334 ★
  • GPT Image2 Ecommerce — GPT-Image-2 驱动的电商素材一键生成 Claude Code Skill E-commerce image generation skill powered by GPT-Image-2 via Codex C 332 ★
  • AI Trading Agent — Multi-symbol AI trading agent — Claude AI + FastAPI + Next.js + MetaTrader 5 72 ★

README

E-commerce AI Agent - Agentic RAG + HITL

[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)

A **multi-skill AI agent** built with **FastAPI**, **LangGraph**, **OpenAI ChatGPT**, and **ChromaDB**. The agent routes each request to the right skill - Q&A (RAG), Order (HITL), or Track - and orchestrates the full conversation flow. 7 nodes, 3 skills, 1 checkpointer, ChromaDB-backed semantic retrieval.

Skills

Each **skill** is a dedicated path through the LangGraph state graph. An LLM classifier picks the right one on every request:

Skill Trigger What happens
Q&A (RAG) "Tell me about laptops" Classify → search catalog → LLM composes answer from results
Order (HITL) "I want to buy a ProBook 15" LLM extracts product → draft → interrupt → confirm → place
Track "Where is my order?" Look up order registry by session → return status + ETA

Adding a new skill means adding one node + one edge - zero changes to existing skills.

Every request passes through **input and output guardrails** (LLM-based validation) before reaching the agent and before returning to the user.

Architecture highlights

  • Multi-skill agent - LLM-powered skill router classifies intent, conditional edges route to the right skill path
  • Agentic RAG - the agent decides when to run semantic retrieval (not every request goes through the vector store)
  • Real RAG with ChromaDB - offline indexing (embed the catalog once) + online retrieval (embed the query, top-k by cosine distance); vectors live in a separate ChromaDB container
  • Human-in-the-Loop - orders require explicit user confirmation via LangGraph's native interrupt() / Command(resume=...) pattern
  • Persistent state - graph execution state survives server restarts (SQLite checkpointer)
  • Package-by-feature - product/ and order/ domains each own their repository (ABC interface — ChromaDB for products, in-memo