A single-file memory layer for AI agents that packages data, embeddings, search structure, and metadata into one portable '.mv2' file — no server, database, or sidecar files. Organized as an append-only sequence of immutable 'Smart Frames' (content + timestamps + checksums), giving time-travel queries over past memory states. Core is a Rust crate (memvid-core) with Node.js, Python, and CLI SDKs on top.
- Storage
- One self-contained .mv2 file with a header, embedded write-ahead log for crash recovery, compressed data segments, a Tantivy full-text (lex) index, an HNSW vector index, a chronological time index, and a TOC footer. Optional password-based encryption (.mv2e). Multimodal ingestion via CLIP (images) and Whisper (audio); PDF text extraction in pure Rust.
- Retrieval
- Local search over the single file combining BM25 full-text (Tantivy) and vector similarity (HNSW), with sub-5ms 'Smart Recall' and predictive caching. Embeddings come from bundled local ONNX models (BGE/Nomic/GTE) or OpenAI API embeddings; an explicit model-binding guard prevents querying an index with a mismatched embedder.
- Self-host
- Self-host: trivial
- License
- Apache-2.0
- Pricing
- Open-source Apache-2.0 core (memvid-core crate + Node/Python/CLI SDKs), free to self-host (Free tier: $0, 50MB memory, unlimited local queries). Managed cloud at memvid.com/pricing: Starter $59/mo (25GB memory, 1k queries/mo, up to 5 memory files) and Pro $299/mo (125GB memory, 10k queries/mo, up to 10 files, session replay, 24/7 support, SLA); an Enterprise Platform tier is also offered with pricing on request. · Free + paid
- GitHub stars
- 16,199
- Last release
- 2026-05-27
- Last commit
- 2026-07-14
- First catalogued
- 2026-06-28
Strengths
- Infrastructure-free: an entire memory store (data + embeddings + lex/vector/time indexes) lives in one portable, append-only file with no server or external DB
- Hybrid retrieval (Tantivy BM25 + HNSW vectors) plus multimodal ingestion (CLIP images, Whisper audio, pure-Rust PDF) and optional file-level encryption
- Polyglot: a Rust core with Node.js, Python, and CLI SDKs; runs fully offline with bundled local ONNX embedders
Watch out
- README headlines self-reported benchmark gains (e.g. '+35% SOTA on LoCoMo', latency/throughput figures) — record via harvest-benchmarks with selfReported:true, not here
- The repo moved to the memvid/memvid org; Memvid v1 (QR-code-based memory) is deprecated — older write-ups describe a different design. The PyPI package 'memvid' (v0.1.3) still points to the deprecated v1 repo (olow304/memvid); v2 users need 'memvid-sdk' not 'memvid'.
Best for
- Agents or apps needing portable, serverless, single-file memory they can copy/version/share, with offline hybrid + multimodal retrieval
How it integrates
Benchmark results
No sourced results yet.
Sources
- Pricing page re-checked — unchanged: Free $0 (50MB memory, unlimited local queries); Starter $59/mo (25GB, 1k queries/mo); Pro $299/mo (125GB, 10k queries/mo) (vendor)
- Monthly card re-verification — GitHub API repo metadata: 16,088 stars, license Apache-2.0 confirmed, archived=false, last commit on the default branch 2026-07-14 (third-party)
- Memvid README (v2.0.140) (vendor)
- GitHub API repo metadata (stars 15699, Apache-2.0 license, pushed_at 2026-05-27) (third-party)
- PyPI memvid-sdk v2.0.160 — official Python SDK for Memvid v2 (vendor)
- npm memvid-cli v2.0.160 — official CLI package (vendor)
- npm @memvid/sdk v2.0.160 — official Node.js SDK (vendor)
- claude-brain — official Claude Code plugin using Memvid .mv2 memory (vendor)
- Memvid pricing page — Free ($0, 50MB), Starter ($59/mo, 25GB), Pro ($299/mo, 125GB), Enterprise Platform (custom) (vendor)
- GitHub license API — confirms Apache-2.0 (third-party)
Last verified 2026-08-01 · updated by refresh-framework-cards