Give your AI agents and assistants long term memory: store facts from conversations and recall the right ones later, from a private local model on your own server.
Mem0 is an open source, Apache 2.0 licensed memory layer for AI applications. You send it the messages of a conversation and it uses a language model to extract the facts worth keeping, such as preferences, names, dates and decisions, then stores them with embeddings so later searches return the most relevant memories for a user, agent or session. The self hosted server provides a REST API for adding, searching, updating and deleting memories and a web dashboard for browsing memories, entities, API keys and request logs.
It suits developers building chatbots, copilots and agents that should remember each user across sessions without keeping every past message in the prompt.
The image remembers and recalls the moment you sign in: a small local language model (Qwen2.5 1.5B on the CPU through Ollama) extracts memories and a local embedding model indexes them in PostgreSQL with pgvector, so nothing leaves the VM and no API key is needed. Mem0's anonymous telemetry is switched off, and the guide explains how to connect OpenAI or another provider when you want faster extraction. First boot creates the administrator account, the admin API key, the database password, the signing secret, the model key and the TLS certificate uniquely for each VM, and the setup wizard is closed so nobody else can claim the server. The dashboard and API sit behind HTTPS, the containers cannot reach the Azure metadata service or your private network, and a step by step deployment guide and 24/7 support come with it.
Real screenshots taken while testing this image against its deployment guide.