Managed Vector Database with Hybrid Search
A managed vector database on AWS, Google Cloud and Azure: HNSW and exact search, filters, hybrid text-and-vector ranking, in a Redis-compatible database. 59.3K searches a second at recall 0.943 in...
Store embeddings beside the rest of your application's data and search them by meaning, keyword or both, in one Redis-compatible database that KiviDB Cloud runs for you. No second system to keep in step.
Vectors where the data already is
- 59.3K: Searches a second At recall 0.943 on glove-25-angular, against 31.8K for Redis 8.10 (query engine) at recall 0.912, in the run published on the vector page.
- Filters and hybrid ranking Tag, numeric and text filters in the same query, and FT.HYBRID to rank by text and vector scores together.
- Searchable as soon as written A vector is in the index when its write returns. No batch job, no refresh interval.
- One system, not two Sessions, cache, queues and embeddings in one database, with one client and one bill.
- Redis Stack commands FT.CREATE and FT.SEARCH work as they do in Redis Stack, the commands that LangChain, LlamaIndex and other Redis vector integrations are built on.
How the managed options differ
| KiviDB Cloud | Dedicated vector databases | Postgres with pgvector | |
|---|---|---|---|
| What it is | An in-memory database with vector search built in | A service for vectors and their metadata | An extension to a relational database |
| Latency | In memory | Varies by service and tier | On disk, with caching |
| Also holds your cache, sessions and queues | yes | no | no |
| Hybrid text and vector ranking | yes (FT.HYBRID) | Varies by service | With full-text search added |
| Client | Any Redis client | The service's own SDK | Any Postgres client |
Dedicated vector databases include Pinecone, Qdrant Cloud, Weaviate Cloud and Zilliz Cloud. Each differs; check its documentation. When the dataset is billions of vectors and too large for memory, a disk-based vector database is usually the better fit.
For AI apps, ready to use
- Semantic cache: Serve an LLM's answer again for a question that means the same.
- Agent memory: Conversation history and lasting memories, found by meaning.
- Context retriever: Search tools for AI agents over your records.
- Vector benchmark: Throughput and recall at every ef setting, against Redis and Dragonfly.
From a trial to production
| Free | Essentials | Pro | |
|---|---|---|---|
| Size | 100 MB | 1 to 256 GB | 4 to 256 GB, or sharded to 512 GB in some regions |
| Nodes | A shared sandbox | One node | A primary and one or more replicas |
| Automatic failover | no | no | yes |
| Committed availability | no | 99.5% | 99.95% |
| From, on AWS | $0 for seven days | $14.52 a month for 1 GB | $95.54 a month for 5 GB |
Prices are live quotes from the rate card, at standard throughput. Google Cloud and Azure are priced on the pricing page. The free tier is a 100 MB shared sandbox with a seven-day lifetime budget, one per owner.
Questions
Is KiviDB a vector database?
Yes. KiviDB stores vectors in hashes or JSON documents and indexes them with HNSW or exact (FLAT) search, with L2, cosine and inner-product distance, filters, and hybrid text-and-vector ranking. It is also a general in-memory database, so the same instance can hold your cache and sessions.
How fast is KiviDB's vector search?
In the published run on glove-25-angular (1.18 million vectors), KiviDB served 59,300 searches a second at recall 0.943 with ef 64, against 31,800 for Redis 8.10 (query engine). Every setting is on the vector page.
Does it work with LangChain and LlamaIndex?
Their Redis vector store integrations are built on FT.CREATE and FT.SEARCH, which KiviDB implements. Test the integration you use against a free database before you rely on it.
How much does a managed vector database cost on KiviDB Cloud?
The same as any KiviDB database: from $14.52 a month for 1 GB on one node, or $95.54 for 5 GB with a replica, billed by the hour. There is no per-query or per-vector charge.
Search by meaning in minutes
Create a free database, index a few thousand embeddings, and query them from the console's Search tab.