Best embedded key-value databases for Rust applications

Combined ranking

1. 1
   RredbGrade ↗—15 pts

   GPT #1Claude #1Gemini #1

   Best default for most Rust apps: pure Rust, stable format, full ACID durability, typed zero-copy API, MVCC readers, savepoints, and strong performance with little tuning; narrowly beats RocksDB for typical single-process workloads because integration and operation are much simpler.

   + model takes & fixes− hide details

   GPT Best default for most Rust apps: pure Rust, stable format, full ACID durability, typed zero-copy API, MVCC readers, savepoints, and strong performance with little tuning; narrowly beats RocksDB for typical single-process workloads because integration and operation are much simpler.

   Claude Pure-Rust, single-file embedded store with real ACID transactions, MVCC (concurrent readers with one writer), and a clean typed API; reached a stable 1.0/2.x with a documented on-disk format, zero C/C++ dependencies, and trivial cross-compilation — the safest default for a typical Rust app that wants a copy-of-LMDB-semantics store without FFI. Near-tie with RocksDB at the top; the choice splits on workload (redb wins for read-heavy/embedded simplicity).

   Gemini Pure Rust ACID-compliant copy-on-write B+tree database offering typed table schemas, MVCC for concurrent readers, zero unsafe code, and zero C FFI build overhead. Earns top rank assuming typical practitioners prioritize memory safety, developer ergonomics, and seamless Rust cross-compilation; near-tie with heed on read utility.

   Where it falls short

   per GPT Its single-writer B+tree architecture is not ideal for sustained write-heavy or multi-process workloads.

   per Claude Copy-on-write B-tree is not write-optimized — heavy sustained write/ingest workloads and very large datasets favor an LSM engine; ecosystem of tooling is thinner than RocksDB's.

   per Gemini Copy-on-write B+tree architecture makes write throughput significantly lower than LSM-tree alternatives under heavy append-only write workloads.

2. 2
   RRocksDBGrade ↗—10 pts

   GPT #2Claude #2Gemini #4

   Near-tie for first and the stronger choice at very large scale: exceptionally mature, fast under heavy writes, highly tunable, and rich in compaction, compression, column-family, snapshot, batch, and transaction features.

   + model takes & fixes− hide details

   GPT Near-tie for first and the stronger choice at very large scale: exceptionally mature, fast under heavy writes, highly tunable, and rich in compaction, compression, column-family, snapshot, batch, and transaction features.

   Claude The most battle-tested engine here — LSM-tree tuned over a decade at Meta scale, excellent for write-heavy and large-than-RAM workloads, with column families, tunable compaction, prefix seeks, and huge production mileage across the industry; the rust-rocksdb bindings are mature and widely deployed.

   Gemini Industry-standard C++ LSM-tree engine accessed via rust-rocksdb bindings, offering unmatched production battle-testing, extensive compaction tuning, block caching, and petabyte-scale write throughput. Assumes project requirements permit C++ FFI in exchange for high-scale performance.

   Where it falls short

   per GPT C++ build dependencies and substantial tuning and compaction complexity make it excessive for many ordinary Rust applications.

   per Claude It is C++ behind FFI — slow build times, large binary, painful cross-compilation, unsafe boundary, and a notoriously deep tuning surface; not for teams that want pure-Rust simplicity or minimal ops burden.

   per Gemini Heavy C++ build footprint that slows compilation, complicates cross-compilation setups, and risks FFI memory safety failures outside Rust runtime guarantees.

3. 3
   LLMDBGrade ↗—9 pts

   GPT #4Claude #3Gemini #2

   High-level, type-safe Rust wrapper around LMDB delivering unmatched zero-copy memory-mapped read performance, sub-millisecond query execution, and decades of proven ACID reliability. Near-tie with redb for read-heavy workloads, assuming maximum read throughput outweighs pure-Rust purity requirements.

   + model takes & fixes− hide details

   Gemini High-level, type-safe Rust wrapper around LMDB delivering unmatched zero-copy memory-mapped read performance, sub-millisecond query execution, and decades of proven ACID reliability. Near-tie with redb for read-heavy workloads, assuming maximum read throughput outweighs pure-Rust purity requirements.

   Claude Memory-mapped B+tree delivering the fastest reads in the category with zero-copy access, full ACID, crash-proof by design, and a tiny, extremely stable codebase; heed gives a well-maintained, ergonomic typed Rust wrapper. Ideal for read-dominated, latency-sensitive workloads.

   GPT Outstanding for read-heavy workloads: mature ACID copy-on-write storage, zero-copy memory-mapped reads, excellent concurrent-reader performance, multi-process access, and a good typed Rust interface through heed.

   Where it falls short

   per GPT Only one write transaction can proceed at a time, so write-intensive or highly concurrent mutation workloads are a poor fit.

   per Claude Single writer serialized globally, a fixed max map size you must pre-size, write amplification and the mmap model make it poor for write-heavy or unbounded-growth data; C dependency remains.

   per Gemini Bound to LMDB C-library mechanics, forcing fixed maximum file size pre-allocations, strict single-writer transaction limits, and C FFI dependency boundaries.

4. 4
   FFjallGrade ↗—8 pts

   GPT #3Claude #4Gemini #3

   Best Rust-native LSM option: safe Rust, range and prefix scans, compression, multiple keyspaces, background maintenance, optional value separation, and serializable single-writer or optimistic multi-writer transactions; it can outrank redb for write-heavy datasets.

   + model takes & fixes− hide details

   GPT Best Rust-native LSM option: safe Rust, range and prefix scans, compression, multiple keyspaces, background maintenance, optional value separation, and serializable single-writer or optimistic multi-writer transactions; it can outrank redb for write-heavy datasets.

   Gemini Modern, pure-Rust LSM-tree storage engine bringing partition keyspaces, customizable compaction strategies, and transactional features to high-write workloads without C++ toolchain friction. Earns its rank on real-world merit as the strongest native Rust alternative for write-heavy pipelines.

   Claude Pure-Rust LSM-tree store that fills redb's main gap — write-optimized, partition (column-family-like) support, actively and thoughtfully developed, and a good match for ingest-heavy pure-Rust services that don't want RocksDB's FFI.

   Where it falls short

   per GPT Durable-to-disk commits require an explicit persistence policy, making the default behavior easier to misuse than redb’s crash-safe default.

   per Claude Youngest and least battle-tested of the group with a smaller ecosystem and fewer large-scale production references; format/API stability guarantees are weaker than redb's or RocksDB's.

   per Gemini Newer codebase with less multi-year enterprise production battle-testing than legacy storage engines, alongside LSM background compaction I/O overhead under sustained load.

5. 5
   LlibmdbxGrade ↗—1 pts

   GPT —Claude #5Gemini —

   A hardened LMDB derivative that fixes several LMDB shortcomings (auto-sizing, better durability modes, larger limits) while keeping the fast mmap B-tree reads; proven at scale as the storage core of Reth, so it carries real high-throughput production credibility.

   + model takes & fixes− hide details

   Claude A hardened LMDB derivative that fixes several LMDB shortcomings (auto-sizing, better durability modes, larger limits) while keeping the fast mmap B-tree reads; proven at scale as the storage core of Reth, so it carries real high-throughput production credibility.

   Where it falls short

   per Claude Still C behind FFI, retains the single-writer model, and has sparse Rust-side documentation and a niche community — for specialists, not for a team wanting an approachable default.

6. 6
   SSurrealKVGrade ↗—1 pts

   GPT #5Claude —Gemini —

   Strongest specialist choice for versioned data: pure-Rust LSM storage with ACID transactions, MVCC, snapshot isolation, time-travel reads, checkpoint and restore, value separation, and support for datasets larger than memory.

   + model takes & fixes− hide details

   GPT Strongest specialist choice for versioned data: pure-Rust LSM storage with ACID transactions, MVCC, snapshot isolation, time-travel reads, checkpoint and restore, value separation, and support for datasets larger than memory.

   Where it falls short

   per GPT Its 0.x maturity and SurrealDB-first design make it a riskier general-purpose dependency than the four options above.

Just missed the top 5

By model

Common questions

Tracked by ModelsAgree · rank 1 = 5 pts … rank 5 = 1 pt · re-polled on demand