Memvid

Memvid gives AI agents persistent memory in a portable file. It's for developers who need agents to recall conversations or search documents without running a separate database server.

• Portable knowledge: Keep content, embeddings, search indexes and metadata together in a .mv2 file, and connect it to models or agents through MCP, SDKs or a direct API.
• Hybrid retrieval: Search with both BM25 keyword matching and semantic vectors, using local ONNX embedding models such as BGE, nomic-embed-text and GTE.
• Memory history: Query by time, inspect earlier memory states, and rewind, replay or branch memory when investigating an agent's behavior.
• Document and media support: Extract PDF text, search images with CLIP embeddings and transcribe audio with Whisper through optional features.
• Recovery and encryption: Recover committed data after crashes and optionally protect memory capsules with password-based encryption.

Memvid is open source under Apache 2.0. The same memory file can live on your machine, an on-premises server, or in a private or public cloud. Retrieval works fully offline with local embeddings once the model files are available. Optional OpenAI API embeddings send content to OpenAI for embedding generation.