https://github.com/tuned-org-uk/arrowspace-rs
VectorDB using dispersion models. Provides graph analysis, vector search and a key-value store for your vectors in one package.
https://github.com/tuned-org-uk/arrowspace-rs
dispersion-model energy-based-model graph-algorithms vector-database vector-search-engine
Last synced: 5 months ago
JSON representation
VectorDB using dispersion models. Provides graph analysis, vector search and a key-value store for your vectors in one package.
- Host: GitHub
- URL: https://github.com/tuned-org-uk/arrowspace-rs
- Owner: tuned-org-uk
- License: apache-2.0
- Created: 2025-08-29T12:47:46.000Z (10 months ago)
- Default Branch: main
- Last Pushed: 2025-12-12T04:11:32.000Z (7 months ago)
- Last Synced: 2025-12-19T14:49:47.088Z (6 months ago)
- Topics: dispersion-model, energy-based-model, graph-algorithms, vector-database, vector-search-engine
- Language: Rust
- Homepage: https://www.tuned.org.uk/posts/010_game_changer_unifying_vectors_and_features_graphs
- Size: 4.56 MB
- Stars: 29
- Watchers: 0
- Forks: 2
- Open Issues: 6
-
Metadata Files:
- Readme: README.md
- Contributing: CONTRIBUTING.md
- Funding: .github/FUNDING.yml
- License: LICENSE
- Code of conduct: CODE_OF_CONDUCT.md
- Citation: CITATION.cff
- Codeowners: .github/CODEOWNERS
Awesome Lists containing this project
README
# surfface: Graph Wiring for any vector space
(`/ˈsɝː.ffɪs/`) Enabling graph application at scale from any embeddings or from any generic vector space.
Inspired by [surface wiring of physical networks](https://www.nature.com/articles/s41586-025-09784-4) and [dark matter structural patterns as spotted by JWST](https://www.nature.com/articles/s41550-025-02763-9).
🚧🚧🚧
Based on previous work done at [`arrowspace`](https://github.com/tuned-org-uk/arrowspace-rs), this library pushes forward the idea that natural processes have developed incredibly effective search mechanisms that can be ported into software to make AI systems more interpretable, managable and safer via observability and supervision. `arrowspace` started as a quest (for an AI Memory Layer) to give to current AI systems some structural metadata (semantic network) using *spectral information* so to provide guardrails for context management in narrow-/domain-tasked LLMs. *Graph Wiring* is the next phase in which the quest is to provide learning capabilities based on surface minimisation to improve search, leveraging current physical models; so that evolution of AI systems can be tracked as a learning process itself via a new generation of tools built for this purpose.
If you are interested in practical applications for you or your company: for an overview of potential products in the field of MLops and data engineering see this [presentation for a database based on `arrowspace`](https://docs.google.com/presentation/d/1Mtz-_85qpVROnp4U2VrnlSHn0266Z1yc_HfjUtfxYLs).
If you are curious about the ideation of *Graph Wiring*: there is the [complete devlog](https://www.tuned.org.uk/blog).
If you are really curious about how this quest started: [a blogpost from 2021 analysing growth in fungal network](https://economyoftime.net/image-processing-1-contours-and-areas-c50a586c6675) and a [paper on the cybernetics of working with intuition graphs](https://www.techrxiv.org/users/685780/articles/679427-cybernetics-interfaces-and-networks-intuitions-as-a-toolbox).
🚧🚧🚧