Research
Built on published research
The memory engine is grounded in published research — not marketing. Here is the paper behind it and the benchmarks we run in the open.
Featured paper — SLoD
Semantic Level of Detail
Multi-scale knowledge representation via heat kernel diffusion on hyperbolic manifolds
SLoD borrows the Level of Detail idea from computer graphics — where 3D engines render geometry at variable resolution by distance — and builds an analogous operator for semantic memory. It is the core of how Mnemoverse stores, recalls, and consolidates what it knows.
- Author
- Edward Izgorodin
- Venue
- GRAAI — IEEE WCCI 2026
- Preprint
- arXiv:2603.08965
Multi-scale by design
A continuous zoom over knowledge: from individual facts to high-level themes, parameterized by a single scale. Memory returns the right granularity for each query.
Hyperbolic substrate
The Poincaré ball embeds tree-structured hierarchies with O(log n) distortion — provably near-optimal for the branching structure of real knowledge.
Proven coherence
Nearby scales produce semantically related representations (bounded error). Two theorems back the hierarchical-coherence and approximation guarantees.
Automatic boundaries
Natural abstraction levels are detected from spectral gaps — no manual tuning. The model finds where knowledge changes character on its own.
Evaluation
Evaluated in the open
We report on public memory benchmarks rather than internal scores you can't reproduce. Configurations are documented and raw results are committed to the repository.
Benchmarked, honestly
Mnemoverse is evaluated on public memory benchmarks — LoCoMo and LongMemEval — rather than internal scores you can't reproduce. On the public LoCoMo leaderboard, Mnemoverse currently sits at #2.
Live, verified numbers: see the benchmarks page.
Go deeper
Follow the work
Research library
The featured SLoD paper, plus written analysis of the AI memory landscape, the architecture decisions behind Mnemoverse, and where memory-augmented agents go wrong.
Benchmark results
Full per-category tables for LoCoMo, HotpotQA and LongMemEval — automated runs, documented configurations, and raw JSON committed to the repository.
The API reference, client integrations, and the methodology behind these results live in the documentation.