SurrealDB launches Agent Memory, scoring 77.2% on BEAM (Strict)
tobiemh · reddit · 2026-10-15
SurrealDB founder Tobie announced Agent Memory, built on the company's multi-model database. It turns conversations and documents into structured entities, relationships and facts, tracking provenance and temporal history.
- Single-store design: SurrealDB natively supports graphs, vectors and documents, so agent context layers and knowledge graphs don't require stitching multiple storage systems.
- Benchmark philosophy: The team argues memory benchmark scores vary wildly with evaluation protocol, so instead of chasing the highest number they targeted BEAM (Strict), hitting 77.2% accuracy.
- Alternatives: They cite Mem0, Honcho and Cognee as interesting approaches and position their work as a different path.
The author is soliciting feedback from builders on long-term memory and mutable-fact management.
Related event: SurrealDB Launches Agent Memory, Scoring 77.2% on BEAM(4 posts)→
More from coding & agent
- Indie dev builds a MOBA in one week with Opus 5.5: 24 heroes, 3 battlegrounds — TAbrodi · 2026-10-15
- AI-built MOBA hits 13.1K plays and 11.3K players in its first week — TAbrodi · 2026-10-15
- Paige Bailey: Ship an Eval People Care About and You Own What Labs Hill-Climb On — DynamicWebPaige · 2026-10-15
- Voice-braindump to PRs: one dev's four-step workflow routing tasks across ten cloud agent projects — vinvan · 2026-10-15
- AI-rewritten libtiff closes gap to ~10x of C version, fueling hopes of retiring buggy C libs — jd_pressman · 2026-10-15
- DigiPet: open-source framework gives an LLM agent a cozy explorable desktop world — Nekileo · 2026-10-15