LEANN vs turbovec
Which should you use? A side-by-side look at what each one does, who it suits, what it costs, how hard it is to set up, and where each falls short.
Open source · MIT
Local semantic search over your own files and chats, using far less storage than typical vector databases.
Open source · MIT
A compressed vector search index for Python and Rust that runs inside your own process.
The short answer
Choose LEANN if you need…
- Developers building a private RAG prototype on a personal machine
- People who want to semantically search their own files or notes locally
- Users of coding agents who want semantic code search through an MCP service
- Anyone with a large document collection who is short on disk space
Choose turbovec if you need…
- Python or Rust developers building semantic search over a few million or fewer embeddings on a single machine
- RAG pipelines that need a local, offline vector store
- Projects where memory use for embeddings is a constraint
- Apps that frequently add and remove vectors and need fast updates
Side-by-side
| LEANN | turbovec | |
|---|---|---|
| Type | Library | Library |
| Price | Free and open source | Free and open source |
| Licence | MIT | MIT |
| GitHub stars | 13.8k | 17.5k |
| Setup difficulty | Intermediate | Intermediate |
| How you get it | Install with a package manager | Install with a package manager |
| Main language | Python | Rust |
| Last code update | 4 days ago | 30 days ago |
What each one is
LEANN
LEANN is a Python tool that builds a searchable index of your personal documents, such as files, emails, and chat logs, so an AI assistant can find relevant passages before answering. It is a vector database that recomputes embeddings on demand instead of storing them all, which the project says cuts storage by about 97% without accuracy loss. People use it to run private retrieval-augmented generation (RAG) on their own laptop without sending data to a cloud service.
turbovec
turbovec stores embedding vectors (lists of numbers that represent text or other content) in a compressed form using 2 or 4 bits per dimension, based on Google Research's TurboQuant method. It then finds the vectors closest to a query, which is the core step in semantic search and retrieval-augmented generation (RAG). It runs as a library inside your program rather than as a separate server.
Best for
LEANN
- Developers building a private RAG prototype on a personal machine
- People who want to semantically search their own files or notes locally
- Users of coding agents who want semantic code search through an MCP service
- Anyone with a large document collection who is short on disk space
turbovec
- Python or Rust developers building semantic search over a few million or fewer embeddings on a single machine
- RAG pipelines that need a local, offline vector store
- Projects where memory use for embeddings is a constraint
- Apps that frequently add and remove vectors and need fast updates
- Teams already using LangChain, LlamaIndex, Haystack or Agno who want a drop-in store
Strengths
LEANN
- Keeps data on the local machine, with no cloud dependency described in the README
- Claims large storage savings, with a benchmark example of 60M chunks in 6GB versus 201GB
- Works with common tooling including Ollama, LangChain, and LlamaIndex, per the repository topics
turbovec
- Much lower memory use than float32 storage: the README reports 7.8× smaller at 4-bit and 15.5× at 2-bit
- Benchmarks in the README show faster search than FAISS at matched recall, on the stated hardware and corpora
- Runs locally with no network service, and supports incremental, crash-safe saves
Not ideal for
LEANN
- Teams needing a hosted, multi-user database, since the README centres on a personal, single-device setup
- Users who need GPU acceleration now, as the README says it is only being considered based on survey votes
- People who want a polished end-user app with no command line, since setup uses uv and Python tooling
turbovec
- Very large corpora with strict latency needs: the README says turbovec scans every vector and suggests a graph index such as HNSW for hundreds of millions of vectors
- Cases needing exact float results, since quantization is lossy and recall depends on your data
- Teams wanting a hosted or networked vector database, because it has no server, replication or multi-tenant API
- Vectors whose dimension is not a multiple of 8 or exceeds 16,384, since those are stated limits
Trade-offs
LEANN
- Setup involves Python version requirements, uv, and optional build steps for DiskANN that can be fiddly on Linux and macOS
- Some README sections are truncated here, so the full feature set and configuration are not verified
- Accuracy and storage claims come from the project and its paper; independent verification was not part of this review
turbovec
- Search cost grows linearly with corpus size because it scans every vector, per the README's own benchmarks
- Quantization is lossy, so top results can differ from exact float search; the README says to check recall on your own data
- Benchmarks are the author's, run on specific GCP machines, so results on your hardware may differ
Getting started
LEANN
Install uv
curl -LsSf https://astral.sh/uv/install.sh | shClone the repository
git clone https://github.com/StarTrail-org/LEANN.git leannturbovec
Install the Python package
pip install turbovecAdd the Rust crate
cargo add turbovecFrequently asked questions
What is the difference between LEANN and turbovec?
LEANN: Local semantic search over your own files and chats, using far less storage than typical vector databases. turbovec: A compressed vector search index for Python and Rust that runs inside your own process. LEANN suits developers building a private rag prototype on a personal machine, while turbovec suits python or rust developers building semantic search over a few million or fewer embeddings on a single machine.
Are LEANN and turbovec free?
LEANN: Free and open source. turbovec: Free and open source.
Is LEANN or turbovec more popular?
turbovec has more GitHub stars (17.5k) than LEANN (13.8k). Popularity isn't the same as fit — check which one matches your use case above.
What are other alternatives to LEANN and turbovec?
Other options people compare them with include LlamaIndex, Faiss, PageIndex, LightRAG, LangGraph.
Keep comparing
Built from our catalogue entries for both products: open-source details come from each project's README (AI-assisted, refreshed weekly) and GitHub data; AI-tool details are editorial. Spotted something wrong? Let us know from either product's page.