Hybrid BM25 + vector + rerank → exact-token recall semantic search misses.
- Storage
- Cloud only
- Retrieval
- Hybrid BM25 + vector search with reranking for exact-token recall across 7 memory types.
- Self-host
- Managed only
- License
- OSS*
- Pricing
- ~$20/mo, paid only (no free tier) · Paid
- GitHub stars
- 34
- Last release
- —
- Last commit
- 2026-06-12
- First catalogued
- 2026-06-28
Strengths
- Great for codes/IDs/error strings
- 7 memory types
Watch out
- Only option with no free tier and no real self-host
Best for
- Agents needing lossless recall — full chronology, no semantic-search step
- Codes / IDs / error strings and preference-recall-heavy apps
Benchmark results
| Benchmark | Value | Backbone | Trust | Source |
|---|---|---|---|---|
| longmemeval | 79 accuracy | gpt-5.4 | Self-reported | RetainDB ↗ |
Sources
- https://github.com/RetainDB/RetainDB (vendor)
- RetainDB — chronological (non-lossy) recall; self-claimed preference-recall SOTA on LongMemEval (vendor)
Last verified 2026-06-28 · updated by refresh-framework-cards