I'm unhappy with Redis for storing embeddings

That’s a common feeling. Redis can work for embeddings, but it often becomes awkward or expensive once you need:

• large-scale vector search
• better filtering / metadata querying
• lower memory cost
• more predictable retrieval quality
• persistence and ops simplicity

A few better alternatives, depending on what you want:

If you want a dedicated vector database

• Qdrant — strong filtering, easy to use, good open-source option
• Milvus — very scalable, more operationally heavy
• Weaviate — good feature set, schema/metadata support
• Pinecone — managed, convenient, usually the fastest path to production

If you want to stay in Postgres

• pgvector — great if your dataset is moderate and you already use Postgres
• Works well when you need SQL + vectors + metadata in one place

If you want fast local / embedded search

• FAISS — excellent for in-process similarity search
• Good for prototypes or services where you manage persistence yourself

If you want search + vectors together

• Elasticsearch / OpenSearch — useful if hybrid keyword + vector search matters

If your main issue is Redis specifically

Redis is often not ideal when:

• embeddings are large
• the dataset grows a lot
• you need complex filtering
• you want cheaper storage per vector

It can still be fine for:

• smallish datasets
• caching recent vectors
• low-latency ephemeral workloads

If you want, I can help you choose a replacement based on:

1. dataset size
2. query pattern
3. whether you need metadata filters
4. budget / managed vs self-hosted
5. latency requirements

If you give me those, I can recommend the best option pretty directly.