FluctlightDB

Preprint

FluctlightDB: A Memory Model of Data for AI Agents

A preprint by Ganesh S, an independent researcher, submitted on 10 July 2026 and updated on 14 September 2026. It is on arXiv as 2608.12365. It is not a journal article.

The argument

For fifty years, data systems have answered two questions. The relational model asked which records match a predicate. The vector model asked which vectors lie nearest a query. Neither was built for cue-driven, provenance-weighted recall across long sessions.

The paper proposes treating long-term agent memory as its own model of data, with write semantics (encoding, separation, consolidation, provenance) and read semantics (a cue spreading through a linked memory graph), and presents FluctlightDB as an embedded engine that implements that contract with experience() and activate(). Embedded, here, means the same packaging as SQLite: the library runs in-process, one brain directory on disk, no server to provision.

It makes that case narrowly. It does not claim to be a new kind of memory layer above Mem0, Zep or HippoRAG-style systems. It claims an engine contract that could sit beneath such a layer. It claims no new neuroscience and no new transformer. Retrieval and question-answering are different measurements, and the paper types its figures by which one it ran.

What it measured

The figures this site will repeat are the ones with a file path beside them, on the benchmarks page. They are maintainer-reported, frozen in July 2026, with open harnesses, and nobody outside the project has reproduced them yet. An earlier headline was withdrawn; the reason is on that page, next to the figure, and it is not repeated as a result.

The engine, the harnesses and the frozen results are in the GitHub repository. The one-minute path the paper names is the same install line as the rest of this site.

Cite this

Ganesh S, “FluctlightDB: A Memory Model of Data for AI Agents”, 2026. Preprint, arXiv:2608.12365. Archived at doi:10.5281/zenodo.20949890.

The citation file in the repository lists the family name as S and the given name as Ganesh, so the BibTeX inverts them. The byline on arXiv reads Ganesh S. ORCID 0009-0006-7758-4114. 

@article{ganesh2026fluctlightdb,
  title         = {FluctlightDB: A Memory Model of Data for AI Agents},
  author        = {S, Ganesh},
  year          = {2026},
  month         = jul,
  eprint        = {2608.12365},
  archiveprefix = {arXiv},
  primaryclass  = {cs.DB},
  doi           = {10.5281/zenodo.20949890},
  url           = {https://arxiv.org/abs/2608.12365}
}

Read it, and the thing it describes

On this site: install it and open a brain, the words the paper uses, the figures with their conditions, and the API the code actually has.