Redis 8.4 Brings Hybrid Search, Atomic Multi-Key Operations, and Auto-Repair AOF

Redis 8.4 has reached general availability on Redis Cloud Pro, delivering a feature set focused on search capabilities, atomicity guarantees, and operational resilience. The headline addition is hybrid search — the ability to combine full-text and vector queries in a single operation — alongside new atomic string commands and automatic repair for corrupted append-only files.

Hybrid Search

The new FT.HYBRID command combines full-text search and vector similarity search using Reciprocal Rank Fusion (RRF) or linear combination scoring. In practical terms, this means an application can search for documents that are both semantically similar to an embedding vector and contain specific keywords, with a single command that returns results ranked by both criteria. Previously, this required two separate queries and application-level result merging.

Hybrid search is particularly relevant for retrieval-augmented generation (RAG) pipelines, where combining keyword precision with semantic recall produces significantly better results than either approach alone. The FT.HYBRID command supports pre-filtering, post-filtering, and score boosting, giving developers fine-grained control over ranking behavior.

New String Commands

Redis 8.4 introduces several commands that address long-standing atomicity gaps. MSETEX performs atomic multi-key SET operations with per-key expiration — previously, setting multiple keys with TTLs required a MULTI/EXEC transaction. DELEX atomically deletes a key and returns its value, useful for "consume and remove" patterns. The SET command gains atomic compare-and-set and compare-and-delete extensions, enabling optimistic concurrency control without Lua scripts.

AOF Auto-Repair

Append-only file (AOF) persistence is Redis's primary durability mechanism, but corrupted AOF files — typically caused by power failures or disk errors — have historically required manual intervention with the redis-check-aof tool. Redis 8.4 automatically detects and repairs corrupted AOF tails during startup, truncating the file to the last valid entry and logging the repair. This reduces the mean time to recovery for Redis instances that experience unexpected shutdowns.

Performance

Lookahead prefetching parses multiple commands from the input buffer in advance, allowing the CPU to prefetch relevant data structures before each command executes. The result is measurably lower latency under high-throughput workloads. JSON memory usage has also been optimized with homogeneous array compaction and short string inlining, reducing the memory footprint of JSON documents by up to 30% in typical workloads.