Decide first whether you want a general vector database or a ready-made product search layer, then whether you'll run it yourself. Open-source engines give you control and more work, managed services take the hosting off your hands, and search engines with vector support suit apps where keyword and semantic search need to live together.
Marqo's homepage now talks about AI-native search, recommendations and merchandising trained on a product catalogue. That's a clear pitch for online stores, and it's a different thing from the general-purpose vector store many developers went looking for. If you need somewhere to store embeddings and query them for your own app, here are seven alternatives in no particular order.
Product search or a database?
Once a tool moves towards a finished product for one industry, developers building something else have to ask whether it still fits. 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.
Storage and querying come down to your own workload: how many vectors you hold, how often they change, whether you filter on metadata as well as similarity, and whether you mix keyword and semantic search. Hosting comes down to your team: whether you've got people to run, patch and scale a cluster, or whether you'd rather someone else did. Run the same real task through each option, with your own embeddings and your own queries, before you decide.
Run it yourself or let someone host it
Open-source engines you can host
Qdrant is an open-source vector search engine written in Rust, and Chroma offers open-source search infrastructure for AI apps. Both suit developers who want control and don't mind running their own infrastructure, with hosted options mentioned too.
Managed vector services
Zilliz offers a fully managed service built on Milvus across the major clouds, which suits teams who'd rather not run clusters. LanceDB takes a different angle, describing itself as a multimodal lakehouse for curating training data at scale.
Search engines with vectors built in
Typesense and Meilisearch started as fast site and app search engines and now add vector search, with self-hosted and cloud options. Vespa is a broader AI search platform covering search, recommendations and personalisation. These suit apps where keyword and semantic search need to work together, much like the catalogue search pitch you may be moving away from.
Seven vector search options compared
| Tool | Engine type | Good for |
|---|---|---|
| Chroma | Open-source search infrastructure | Developers prototyping AI retrieval on open source |
| LanceDB | Multimodal lakehouse | Curating large multimodal datasets for model training |
| Meilisearch | Search engine, cloud or self-hosted | Site search plus AI retrieval in one engine |
| Qdrant | Open-source vector search engine | Fast vector similarity search through an API |
| Typesense | Open-source search engine | Typo-tolerant site and app search with vectors |
| Vespa | AI search platform | Search, recommendations and personalisation at scale |
| Zilliz | Managed vector service on Milvus | Fully managed vector search on the big clouds |
The alternatives from A to Z
Alphabetical and in no particular order. Each line reflects how the vendor describes its own product.
Chroma
Best for: Developers who want open-source search infrastructure for AI apps, with a free plan named alongside the open-source project.
LanceDB
Best for: AI teams curating large multimodal datasets and engineering features, where the vendor's pitch is training on the right data faster.
Meilisearch
Best for: Product teams wanting one engine for search and AI retrieval, run on the vendor's managed cloud or self-hosted from the open-source code.
Qdrant
Best for: Engineers who want an open-source vector search engine written in Rust, with a convenient API and a focus on speed at scale, per the vendor.
Typesense
Best for: Teams adding fast, typo-tolerant search to a site or app, with vector search, simple APIs and self-hosting, according to the vendor.
Vespa
Best for: Companies building AI search, agents, personalisation and recommendations on one platform; the vendor names a free trial.
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.
Zilliz
Best for: Organisations that want a fully managed vector service built on Milvus, running in regions across AWS, Google Cloud and Azure, per the vendor.
Advice for developers choosing a vector store
Don't choose on a vendor's benchmark chart. Every chart is run on data and queries that suit the vendor, and yours won't look the same. Load a realistic slice of your own embeddings, run the queries your app will run, including any metadata filters, and measure what you get on the hardware or plan you'd use in production.
Think about the next two years, not the next sprint. Ask how each option handles growing data, re-indexing and backups, and how hard it'd be to move away later. If you're building agents on top of retrieval, my guide to AI agents for business covers where they help, and my piece on how AI search is changing marketing explains why good retrieval matters beyond your own app.
An afternoon benchmark with your own embeddings
- Pick a representative sample of your real data and generate embeddings with the model you'll use in production.
- Write down twenty or so real queries, with the answers you'd expect, so you can judge relevance as well as speed.
- Load the same data into each shortlisted option, using the hosting setup you'd really choose.
- Run every query with and without your usual metadata filters, and note speed and result quality.
- Update and delete a batch of records, then check how quickly the changes show up in results.
- Keep the option that gave the most relevant answers at a speed your app can live with.
Do you make vector databases? 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 stores embeddings, which are lists of numbers that represent the meaning of text, images or other data, and finds the items closest to a query. It's the retrieval layer behind many AI search and chat apps.
Do I need a dedicated vector database?
Not always. For small projects, a search engine or database with vector support may be enough. Dedicated engines help when data grows large or queries need to be very fast.
Should I self-host or use a managed vector service?
Self-hosting gives you control but means running, patching and scaling it yourself. A managed service takes that work away. Choose based on your team's time and skills.
What's hybrid search?
Hybrid search combines keyword matching with vector similarity, so exact terms and meaning both count. It helps when users search for product names, codes or specific phrases.
Can I use a vector database for retrieval-augmented generation?
Yes, that's one of the most common uses. The database finds the passages closest to a question, and a language model uses them to write its answer, so retrieval quality shapes answer quality.
How hard is it to move embeddings to a new vector store?
The vectors themselves move easily as data, but indexes, filters and query code need rebuilding. Plan time for re-indexing and testing before you switch traffic.
