Agent With Memory

Agent With Memory (AWM 2.0)

AWM 2.0 is a MongoDB-first starter for building AI applications with persistent memory, retrieval, and agent runtime infrastructure.

It is designed for developers who want a serious reference implementation, not a toy chatbot and not a one-command production platform.

What You Get

• A real FastAPI API surface for agents, chat, memories, evaluation, HITL, time-travel, and WebSocket chat
• A 7-type memory system backed by MongoDB
• Hybrid retrieval with vector search, text search, and fallbacks when cluster capabilities are limited
• A LangGraph-first runtime with persisted checkpoints and long-term memory
• Seeded validation scripts for realistic data, Atlas cloud validation, and Atlas Local Preview validation
• Test-only external LLM smoke tooling that stays out of runtime code paths

What This Repo Is

This repo is a strong starting point if you are building:

• an AI assistant with cross-session memory
• a memory-rich RAG application
• a LangGraph-based agent on MongoDB
• a reference architecture for MongoDB-powered AI systems

This repo is not:

• a promise that one git clone gives you a production app
• a hosted SaaS template with auth, billing, frontend, and deployment all finished
• a benchmark-backed claim of being the single best boilerplate on earth

Why Developers Actually Use It

The point is not “look how many features fit in one README.”

The point is that you can start with a codebase that already solves the annoying parts teams usually rebuild badly:

• memory types with clear ownership and isolation rules
• retrieval that can work across cluster capability differences
• long-running chat state and replayable checkpoints
• seeded validation with real data instead of fake “hello world” strings
• API routes that match the runtime instead of demo-only placeholders

If you are building your own agent app, you should be able to copy patterns from here without first reverse-engineering a bunch of undocumented dec