# Memvid

> A portable, single-file memory layer for AI agents with instant retrieval, long-term memory, and no database required.

Memvid is an open-source, portable AI memory system that packages data, embeddings, search structures, and metadata into a single `.mv2` file. Instead of running complex RAG pipelines or server-based vector databases, Memvid enables fast retrieval directly from the file, making it model-agnostic and infrastructure-free. It gives AI agents persistent, long-term memory they can carry anywhere, with sub-5ms local access and crash-safe, append-only writes.

- **Smart Frames** — *Memvid organizes memory as an append-only sequence of immutable Smart Frames, each storing content with timestamps, checksums, and metadata for efficient compression and parallel reads.*
- **Single-File Format (.mv2)** — *All data, indexes (full-text, vector, time), and metadata live in one portable file with no sidecar files, WAL files, or lock files.*
- **Time-Travel Debugging** — *Rewind, replay, or branch any memory state to inspect how knowledge evolves over time.*
- **Multi-Modal Search** — *Supports BM25 full-text search, HNSW vector similarity search, CLIP visual embeddings for image search, and Whisper audio transcription.*
- **Local & Cloud Embeddings** — *Use local ONNX models (BGE-small, BGE-base, Nomic, GTE-large) or cloud API embeddings via OpenAI, with model-binding to prevent accidental mixing.*
- **Multi-Language SDKs** — *Available as a Rust crate (`memvid-core`), Node.js SDK (`@memvid/sdk`), Python SDK (`memvid-sdk`), and a CLI (`memvid-cli` via npm).*
- **Offline-First** — *Works fully offline with no server dependencies, making it ideal for edge deployments, air-gapped systems, and resource-constrained environments.*
- **Capsule Context** — *Self-contained, shareable `.mv2` memory capsules support rules, expiry, and encryption for secure knowledge sharing.*
- **Benchmark Performance** — *Achieves +35% SOTA on LoCoMo long-horizon recall, 0.025ms P50 latency, and 1,372× higher throughput than standard memory systems.*
- **Broad Use Cases** — *Supports long-running AI agents, enterprise knowledge bases, codebase understanding, customer support agents, and auditable AI workflows.*

## Features
- Single-file portable memory (.mv2 format)
- Smart Frames append-only memory architecture
- Sub-5ms local memory retrieval
- BM25 full-text search with Tantivy
- HNSW vector similarity search
- CLIP visual embeddings for image search
- Whisper audio transcription
- Local ONNX text embedding models
- OpenAI cloud API embeddings
- Time-travel debugging and memory rewind
- Crash-safe immutable frame commits
- Encryption support for memory capsules
- Model-agnostic and infrastructure-free
- Offline-first operation
- Multi-threaded ingestion
- PDF text extraction
- Temporal/natural language date parsing

## Integrations
OpenAI, Rust (cargo), Node.js (npm), Python (pip), HuggingFace ONNX models, BGE embeddings, Nomic embeddings, Whisper

## Platforms
API, CLI, DEVELOPER_SDK

## Pricing
Open Source

## Version
v2.0.139

## Links
- Website: https://www.memvid.com
- Documentation: https://docs.memvid.com/
- Repository: https://github.com/memvid/memvid
- EveryDev.ai: https://www.everydev.ai/tools/memvid