Olow304

memvid: Give AI Agents Portable, Single-File Memory

Memvid is a Rust-based, single-file memory layer for AI agents, designed to replace complex RAG pipelines and server-based vector databases. It packages data and search indexes together, using Tantivy for full-text search and HNSW for vector search.

QuietRustApache-2.0AI AgentsRAGAgent Memory
Stars
17.3k
Forks
1.5k
Stars, last 30 days
+820
Last commit
4 months ago
Contributors
23

Our take

Quietno commits in the last 90 days

Health score

44/100

How it is scored
Activity
41
Community
70
Issues
46
Pull requests
20

+820 stars in the last 30 days

Memvid is a distinctive portable-memory approach with several integration paths, but its recent development has been quiet despite releases in the past year.

Good fit if

  • You want agent memory stored and retrieved from a self-contained file rather than a server-based vector database.
  • Your application is Rust-based, or you can use the listed Python or Node.js SDKs.
  • You value an Apache-2.0 project with contributions from multiple people rather than a project concentrated around one contributor.

Look elsewhere if

  • Your team needs evidence of ongoing commit activity before adopting a core dependency.
  • You need a managed database service rather than a local file-based memory layer.
  • You need embeddings without downloading local model files or configuring an external API provider.
All health signals
Last commit2026-07-14 (4 months ago)
Commits, last 90 days0
Releases, last 12 months10 (latest v2.0.140, 2026-05-27)
Contributors23 (top contributor: 26% of commits)
Issues closed, last 90 days1
Pull requests merged, last 90 days0
Project age17 months

Checked on 2026-10-30 with the GitHub API.

Project history

Daily GitHub snapshots since Oct 7, 2026. History covers stars and forks; health signals are not tracked historically.

Selected snapshot

Stars

17,422

Forks

1,467

Change since Nov 16, 2026: +6 stars·0 forks

Overview

Memvid is a Rust-based memory layer for AI agents that puts data, metadata, and search structures in a portable .mv2 file. It aims to replace complex RAG pipelines and server-based vector databases with retrieval directly from a self-contained file. Its format organizes data into append-only Smart Frames, with Tantivy for optional full-text search and HNSW for optional vector search.

The project is for developers building agents that need persistent memory across sessions, especially when they want to avoid operating a separate database service. Rust is the core implementation, with CLI, Node.js, and Python SDKs also listed in the README.

Key Features

  • Store memory data and indexes together in a single .mv2 file, without sidecar files.
  • Append immutable Smart Frames and inspect or query earlier memory states.
  • Enable Tantivy BM25 full-text search and HNSW vector search through Rust feature flags.
  • Support local text embeddings through ONNX, with model files downloaded separately.
  • Add optional PDF extraction, CLIP image embeddings, and Whisper transcription through feature flags.
  • Use the CLI, Node.js SDK, Python SDK, or Rust crate to work with memory files.

Use Cases

  • Agent developers can keep conversation or task memory between runs without running a separate vector database.
  • Teams working offline or on constrained infrastructure can package knowledge and search indexes as portable files.
  • Knowledge-base builders can combine full-text and vector retrieval for documents, with optional PDF ingestion.
  • Developers building multimodal workflows can use the optional CLIP and Whisper features to search images or transcribe audio.

What you need

Detected in the repository

  • Rust 1.85.0 or newer (from Cargo.toml)
  • A test suite and automated checks on GitHub Actions

License in plain words

Apache-2.0permissive

  • Commercial use: yes
  • Modify and redistribute: yes
  • You must keep: the license, the NOTICE file and a note of your changes
  • Share your changes: no
  • Includes an explicit patent grant from the contributors:

A summary, not legal advice: the LICENSE file is what applies.

Getting Started

Install the Rust crate:

cargo add memvid-core

See the README for feature flags, SDKs, setup details, and examples.

Alternatives

  • RuVector: RuVector provides a broader vector and agent-memory toolkit with multiple runtime paths, while Memvid packages full-text and vector indexes in a single file.
  • xerj: XERJ is a local-file search engine and MCP server with an Elasticsearch-compatible API, while Memvid is a single-file memory layer.
ProjectLanguageLicenseStarsStatus
memvidRustApache-2.017.3kQuiet
RuVectorRustMIT4.6kActive
xerjRustApache-2.013.2kNew

Considerations

The health data marks the project as quiet, with no commits in the last 90 days, although it has continued to publish releases over the past year. The repository has 23 contributors, and contributions are not concentrated in one person. Issue handling has also been limited recently, with only one issue closed in the last 90 days.

The Rust crate requires Rust 1.85 or newer. Some capabilities require optional features and additional setup: local text embeddings require manually downloaded ONNX model files, and OpenAI embeddings require an API key. The repository is licensed under Apache-2.0.

Found this useful?

Share it with someone who would like memvid.

Comparisons