# Memvid

A single-file, serverless memory layer for AI agents that packages data, embeddings and search structures into one portable file, giving persistent, versioned, long-term memory without a database.

Memvid is a portable memory system for AI agents that packages data, embeddings, search structures and metadata into a single file, enabling fast retrieval directly from that file instead of running a RAG pipeline or a server-based vector database.

Internally, memory is organized as an append-only sequence of immutable units called Smart Frames, each storing content with timestamps, checksums and metadata. This design supports append-only writes without corrupting existing data, queries over past memory states, timeline-style inspection of how knowledge changed, crash safety through committed frames, and compression adapted from video encoding techniques.

Its concepts include a capsule format (.mv2) for self-contained, shareable memory with rules and expiry, time-travel debugging to rewind, replay or branch a memory state, and low-latency local recall with predictive caching. It is written in Rust and released under the Apache 2.0 license.

Website: https://www.memvid.com