gaurav0807/travel-planning-agent

This is a Python project that demonstrates how to build a multi-agent AI system using LangGraph and AWS Bedrock, with travel planning as the practical scenario. The core purpose is educational: it shows how to implement a memory architecture called CoALA (Cognitive Architectures for Language Agents) with all four of its memory types working together. When you run it, you have a conversation with an AI that helps you plan a trip. You describe your destination, dates, budget, and interests, and the system runs four specialized agents in sequence: a trip analyzer that asks clarifying questions and extracts your requirements, a flight searcher that suggests options, a hotel searcher that finds accommodation within your budget, and an itinerary planner that assembles a day-by-day schedule. The four memory types from the CoALA framework each serve a different role. Working memory holds the current session state, your conversation, trip details, and the options found so far. Episodic memory persists your past trips and preferences across sessions so the system can personalize future recommendations. Semantic memory is a knowledge base of facts about destinations, hotels, and flights. Procedural memory is the set of system prompts that define how each agent should behave and format its output. Sessions are saved with a SQLite database as a checkpointer, and you can resume any previous session by passing its ID on the command line. The underlying AI model is Claude Haiku 4.5, accessed through the AWS Bedrock Converse API. LangGraph is used to connect the agents and define the routing logic between them. The project is released under the MIT license.