Redis: In-Memory Data Platform for Serverless and AI (2026)
Redis: In-Memory Data Platform for Serverless and AI (2026)
Quick Answer
Redis has evolved from a simple in-memory key-value store into a comprehensive multi-model data platform. Redis 8 delivers production-ready vector search with HNSW indexing, hybrid search combining semantic and keyword queries, and REST APIs for serverless compatibility. Important: In 2024, Redis licensing changed from BSD to source-available (RSALv2/SSPLv1). The Linux Foundation now maintains Valkey, an open-source fork.
What is Redis?
Redis (Remote Dictionary Server) is an in-memory data structure store that persists to disk. Unlike traditional databases that treat RAM as cache, Redis keeps the entire dataset in memory for microsecond latency. It supports strings, hashes, lists, sets, sorted sets, JSON, streams, time-series, and high-dimensional vectors.
The 2024 Licensing Fork
| Project | License | Maintainer | Status |
|---|---|---|---|
| Redis (official) | RSALv2 / SSPLv1 | Redis Inc. | Source-available |
| Valkey | BSD 3-Clause | Linux Foundation | Open source |
Implication: Cloud providers (AWS, Google, Oracle) cannot sell managed Redis based on versions 7.4+ without commercial agreements. Valkey remains fully open source.
Redis Stack: The Multi-Model Bundle
Redis Stack combines multiple modules into one distribution:
| Component | Purpose |
|---|---|
| RedisJSON | Native JSON type with atomic updates |
| RediSearch | Full-text search + secondary indexes |
| RedisTimeSeries | Compressed time-series data |
| RedisBloom | Probabilistic structures (Bloom, Cuckoo, Top-K) |
Key Features
Vector Search (Redis 8+)
Redis competes with dedicated vector databases through native indexing:
| Index Type | Use Case | Characteristics |
|---|---|---|
| HNSW | Production ANN | O(log N) search, high recall |
| FLAT | Exact search | O(N) brute force, 100% recall |
| SVS-VAMANA | Hardware-optimized | DiskANN-based algorithm |
Constraints:
- Max dimensions: 32,768
- Max indexes per DB: 10
- M parameter: Up to 512 (default: 16)
Hybrid Search with RRF
Redis 8.4 introduced FT.HYBRID for combining vector and text search:
FT.HYBRID my_index "semantic query" "keyword query" RRF 10
The Reciprocal Rank Fusion algorithm merges results from vector and text searches, normalizing scores across different scales.
Data Structures
| Structure | Use Case | Key Commands |
|---|---|---|
| String | Caching, counters | GET, SET, INCR |
| Hash | Objects | HGET, HSET, HMGET |
| List | Queues | LPUSH, RPOP, BLPOP |
| Set | Unique items, intersections | SADD, SINTER |
| Sorted Set | Leaderboards, ranking | ZADD, ZRANGE, ZRANK |
| Stream | Pub/sub, message log | XADD, XREAD, XGROUP |
Probabilistic Structures
Redis includes memory-efficient structures for massive datasets:
- HyperLogLog: Cardinality estimation with 12KB for 1B items (0.81% error)
- Bloom Filters: Set membership testing (no false negatives)
- Cuckoo Filters: Bloom filter with deletion support
- Count-Min Sketch: Frequency estimation in streams
- Top-K: Trending topics detection
Redis vs Competitors
Redis vs Memcached
| Feature | Redis | Memcached |
|---|---|---|
| Persistence | RDB + AOF | None (volatile) |
| Data Types | 8+ structures | Simple key-value |
| Replication | Async + Sentinel | None |
| Multithreading | I/O threads (v6+) | None |
Redis vs Upstash
| Feature | Redis Cloud | Upstash |
|---|---|---|
| Protocol | RESP (TCP) | HTTP/REST |
| Pricing | Fixed or Flex | Pay-per-request |
| Serverless | Via proxy | Native |
| Edge | Limited | Global regions |
Redis vs Dedicated Vector DBs
| Feature | Redis (RediSearch) | Pinecone | Weaviate |
|---|---|---|---|
| Primary Use | Multi-model + vectors | Vectors only | Vectors only |
| Indexing | HNSW / FLAT | HNSW only | HNSW / Flat / Dynamic |
| Hybrid | Native FT.HYBRID | Sparse + dense | BM25 + dense |
| Memory | Required | Disk-optimized | LSM + HNSW |
Deployment Options
Redis Cloud (Official)
| Tier | Cost | Features |
|---|---|---|
| Essentials (Free) | $0 | 30MB, 100 ops/sec, 30 connections |
| Fixed | $10+/month | Dedicated RAM, replication |
| Redis Flex | Usage-based | Decoupled storage + compute |
Hyperscaler Options
| Provider | Service | Notes |
|---|---|---|
| AWS | ElastiCache / MemoryDB | MemoryDB for durability |
| Google Cloud | Memorystore | Vector search limits |
| Azure | Azure Cache for Redis | Enterprise tier available |
Serverless: Upstash
For serverless and edge workloads:
- HTTP-based: No TCP connection management
- Global replication: Multi-region with read-your-writes
- Pricing: $0.20 per 100,000 commands
Client Libraries
Node.js
| Client | Type | Notes |
|---|---|---|
| node-redis | Official | Modern Promise API |
| ioredis | Community | Better cluster support |
Python
| Client | Type | Notes |
|---|---|---|
| redis-py | Standard | Thread-safe, connection pooling |
| RedisVL | AI-focused | Vector abstraction for RAG |
Java
| Client | Type | Notes |
|---|---|---|
| Jedis | Synchronous | Simple, blocking |
| Lettuce | Asynchronous | Netty-based, reactive |
Operational Considerations
Connection Limits
Redis uses a process-per-connection model:
- Each connection: ~10MB RAM
- Limit: Cannot handle tens of thousands of concurrent connections natively
- Solution: Connection poolers (PgBouncer equivalent) or serverless proxies
Persistence Strategies
| Method | Description | Trade-off |
|---|---|---|
| RDB | Point-in-time snapshots | Compact, but data loss since last snapshot |
| AOF | Append-only log | Durable, larger files |
| Hybrid | RDB + AOF | Balance of size and durability |
Auto Tiering (Enterprise)
Hot data in RAM, warm data on NVMe:
- Reduces cost for large datasets
- Transparent to application
- Maintains low latency for active data
Use Cases: Best For
- Caching: Session data, API responses, database queries
- Rate Limiting: Per-IP or per-user request limits
- Leaderboards: Real-time gaming with sorted sets
- Pub/Sub: Real-time messaging, chat systems
- Vector Search: RAG applications with hybrid search
- Streams: Event sourcing, message queues
Use Cases: Avoid For
- Primary Database Without Replication: Data loss risk without persistence
- Large Blob Storage: Store files in S3, not Redis
- Complex Multi-Row Transactions: Use PostgreSQL instead
Pricing Comparison
| Provider | Model | Entry Point |
|---|---|---|
| Redis Cloud Free | Fixed | 30MB, 100 ops/sec |
| Redis Cloud Fixed | Fixed | From $10/month |
| Redis Flex | Usage-based | Storage + throughput shards |
| Upstash | Per-request | $0.20 per 100K commands |
| AWS MemoryDB | Per-node | Depends on instance type |
FAQ
Is Redis free?
The Redis core is free, but licensing changed in 2024 to source-available (RSALv2/SSPLv1). The Valkey fork remains fully open-source (BSD). Managed services (Redis Cloud, AWS MemoryDB) charge for hosting.
Does Redis support vector search?
Yes, Redis 8+ supports HNSW and FLAT vector indexing with up to 32,768 dimensions. Redis 8.4 added hybrid search combining semantic and keyword queries using RRF.
What is the difference between Redis and Valkey?
Valkey is a Linux Foundation fork of Redis 7.2, maintained under the BSD license. Redis (official) moved to source-available licensing in 2024. Both are technically similar, but Valkey remains truly open source.
Can I use Redis with serverless?
Standard Redis requires persistent TCP connections, which doesn’t scale well with serverless. Use Upstash (HTTP-based) or Redis Cloud with connection pooling for serverless deployments.
Does Redis support ACID transactions?
Redis supports transactions via MULTI/EXEC, but they are not fully ACID compliant. Redis offers atomicity for command batches but lacks the isolation levels and rollback capabilities of relational databases.
Related Technologies: Upstash, PostgreSQL, Pinecone, Weaviate, Memcached