Choose a LanceDB replacement by how much infrastructure you want to run. If you'd rather not manage servers, a managed vector service is simplest. If you want control, an open source engine you host yourself fits better. If search is the product, a full search platform with vectors built in may suit. Test each on your own data at your real scale.
LanceDB now calls itself a multimodal lakehouse for AI, aimed at curating large datasets and engineering features so teams can train better models. That's a broader pitch than a plain vector store, and it's worth being clear about which part you're replacing. If you mainly store embeddings and search them, the seven options below, in no particular order, are the ones to compare.
Lakehouse, vector store or search engine?
Developers usually reach for LanceDB for one of two reasons: an embedded vector store that runs next to their code, or a place to keep and prepare large multimodal datasets. The tools here cover the first job in different ways, and none of them is a drop-in replacement for the second.
Dedicated vector databases
Qdrant, Weaviate and Chroma are built around storing embeddings and finding the nearest matches fast. All three talk about open source, and they suit teams building retrieval for chatbots, agents and recommendation features.
Managed vector service
Zilliz runs Milvus as a fully managed service across the big cloud providers, for teams that want vector search without operating the database themselves.
Search engines with vectors built in
Vespa and Typesense are full search engines that handle vectors alongside keyword search, which helps when you need both in the same query. Marqo is narrower, aimed at ecommerce search, recommendations and merchandising trained on a product catalogue.
Run it yourself or let someone host it
The thing to test in any replacement is how data is stored and queried at scale, and whether you run it yourself or use a managed service. A vector search that's instant on a test set of a few thousand embeddings can slow down a lot once you load your real collection, add filters and start writing new data while people are querying.
Self-hosting gives you control over where data lives and how the system is tuned, but someone on your team has to handle upgrades, backups and scaling. A managed service takes that work away and adds a dependency on the vendor's uptime and regions. Neither is better in general. It depends on whether you have people who want to run databases, and on where your data is allowed to sit.
Seven vector search options compared
| Tool | Kind of store | Good for |
|---|---|---|
| Chroma | Open source search infrastructure | Developers starting AI search on open source |
| Marqo | AI product search | Ecommerce search and recommendations on a catalogue |
| Qdrant | Open source vector engine | Fast similarity search with a simple API |
| Typesense | Open source search engine | Typo-tolerant site and app search plus vectors |
| Vespa | AI search platform | Search, recommendations and retrieval together |
| Weaviate | AI-native vector database | Building complete AI apps with less lock-in |
| Zilliz | Managed Milvus service | Fully managed vector search on major clouds |
Databases and engines, A to Z
Sorted alphabetically and in no particular order. Each line describes what the vendor says about its own product.
Chroma
Best for: Developers who want open source search infrastructure for AI applications, with a free plan named on the vendor's site.
Marqo
Best for: Online retailers wanting search, recommendations and merchandising trained on their own catalogue rather than a general database, per the vendor.
Qdrant
Best for: Engineers wanting an open source vector search engine written in Rust, with the vendor stressing speed, scale and a convenient API.
Typesense
Best for: Teams building site or app search that forgives typos, with vector search and self-hosting available, as the vendor describes.
Vespa
Best for: Groups running search, personalisation, recommendations and retrieval for AI agents on one platform, according to the vendor.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
Weaviate
Best for: Builders of AI applications worried about hallucination, data leakage and vendor lock-in, which the vendor says it helps reduce; a free plan is named.
Zilliz
Best for: Companies wanting a fully managed service built on Milvus, running across AWS, Google Cloud and Azure regions, per the vendor.
Engineering advice before you migrate
Keep your embedding model the same during the move. If you change the database and the model at once, you won't know which one caused a change in result quality. Export your vectors and metadata, load them into two shortlisted options, and run the same set of real queries against both, with the same filters your application uses.
Also think about what sits around the database. If you're building agents or retrieval for a chatbot, the store is only one part of the system. My guide to AI agents for business covers where retrieval fits in, and if you're starting from scratch, my walkthrough on how to build your first AI agent for marketing shows a simpler path before you worry about scale.
Benchmark on your own embeddings
- Export a realistic sample of your vectors and metadata, not a toy dataset.
- Write a list of real queries, including the filters your application applies.
- Load the sample into two shortlisted options using the same index settings where possible.
- Measure query speed and result quality while writing new data at the same time.
- Repeat at your expected full size before you commit, then plan the cut-over.
Do you make vector database software? Tool makers can ask to be considered for a featured place on this list.
Want your tool on this list? Featured spots and reviews here are paid partnerships. Here's how to partner with me.
Frequently asked questions
What is a vector database?
It's a database built to store embeddings, which are lists of numbers that represent text, images or other data, and to find the closest matches quickly. It's the core of most AI retrieval features.
Do I need a vector database or a search engine?
If you only need similarity search over embeddings, a vector database is enough. If you need keyword search, filters and ranking as well, a search engine with vector support may suit better.
Should I self-host or use a managed vector service?
Self-hosting gives control over data and tuning but needs people to run it. A managed service removes that work and adds a dependency on the vendor. Choose based on your team and data rules.
Can I move my embeddings from LanceDB to another database?
Yes. Export the vectors with their IDs and metadata, then load them into the new store. Keep the same embedding model so the results stay comparable.
How do I test vector search quality?
Use real queries with known good answers, then check how many of the right results come back near the top. Run the same test on each option with the same data.
Where can my vector data be stored?
It depends on what the data contains and where your users are. Data protection rules vary by country, so check each vendor's hosting regions and ask your regulator or data protection lead.
