Memory Agent

A chat assistant that actually remembers you — across every conversation, even after you close the tab.

🔗 Live demo: https://mem0-memory-agent.vercel.app

Most chatbots forget everything the moment the session ends. Memory Agent pairs Claude with Mem0's long-term memory layer so it builds a persistent profile of each user — their name, goals, preferences, and past topics — and weaves that context naturally into every reply. It's a compact, production-shaped demo of the retrieve-augment-persist loop that gives an LLM durable memory.

Features

• Persistent long-term memory — facts you share are extracted and stored in Mem0, then recalled in future sessions (not just within one chat).
• Streaming responses — replies stream token-by-token over Server-Sent Events for a live, real-time feel.
• Per-user memory — each visitor gets their own userId (kept in localStorage), so memories are scoped to that person.
• Inspectable memory panel — a "Memories" drawer shows exactly what the agent currently remembers about you, with a one-click refresh.
• Clear all / switch user — wipe stored memories or start a fresh identity at any time.
• Conversation-starter suggestions — guided prompts on the empty state to demonstrate the memory loop quickly.
• Server-side keys — both the Anthropic and Mem0 API keys live only in server route handlers, never in the browser, with an origin allow-list on write endpoints.

How it works

The app is a Next.js project (App Router) with a React client and two server route handlers. On each turn, POST /api/chat first calls mem0.search() to pull the most relevant stored memories for the user's message, injects them into Claude's system prompt, then streams the completion from the @anthropic-ai/sdk (claude-sonnet-4-6) back to the browser as Server-Sent Events. After the response finishes, the full user/assistant exchange is handed to mem0.add() so new facts are extracted and persisted for next time. A second handler, /api/memories, exposes getAll (read) and deleteAll (clear) for the memory panel. Both the ANTHROPIC_API_KEY and MEM0_API_KEY are read from process.env inside the server routes, so secrets never reach the client.

Tech stack

• Framework: Next.js 16 (App Router) + React 19
• Language: TypeScript
• AI provider: Anthropic Claude (claude-sonnet-4-6) via @anthropic-ai/sdk
• Memory: Mem0 (mem0ai SDK) for long-term storage, retrieval, and extraction
• Styling/UI: Tailwind CSS v4 + lucide-react icons
• Persistence: Mem0 (server-side memory) + localStorage (client-side user identity)
• Hosting: Vercel

Running locally

npm install
npm run dev

Environment variables

Variable	Purpose
ANTHROPIC_API_KEY	Authenticates calls to the Claude API (used server-side in the chat route).
MEM0_API_KEY	Authenticates calls to Mem0 for storing, searching, and retrieving long-term memories.

---

Part of my AI engineering portfolio — built by Eduardo San Martin (github.com/anihc1978).