A dynamic, self-constructing graph database
Spyndel combines a purpose-built graph store with an engine that writes the graph for you. You get an ontology of your documents that stays current as your library grows, without the modelling effort or the running AI costs.
The graph
- Statements: every requirement, specification and piece of information, with document, page and section.
- Topics and categories: discovered from your content and named in plain language, in a browsable hierarchy.
- Connections: verified links between statements across documents, each with a confidence score.
- Usage memory: which statements and links helped answer real questions.
The engine
- Builds itself: structure emerges from the documents themselves, with no schema to design.
- Grows incrementally: new documents are placed and linked without reprocessing the library.
- Separates substance from noise: citations, headers and boilerplate are recognized and kept out of the way.
- Runs locally: compact specialist models on CPU or GPU, with no per-document token costs.
Everything your team needs to use it
Browse
Move from categories to topics to individual statements, with the related statements, source documents and the text around each statement.
Concept & document maps
Interactive maps of how subjects and documents connect. Zoom from the whole library down to individual statements and their links.
Ask
Plain-language questions answered from the graph, with every claim tied to a checked citation. Bring your own AI provider.
Silk query language
A familiar graph query language for analysts who want counts, paths and custom questions, with a console built in.
AI assistant integration
Connect any MCP-capable assistant. It can search, traverse connections step by step, and get attribution ready to cite.
Document uploads
Drag in PDFs or send them through the API. Progress is shown live, and each document can be opened at the cited page.
Users & roles
Built-in accounts with role-based permissions. Define your own roles, such as reviewers who can read and ask, or editors who can add documents.
GPU acceleration
Detects your GPU and uses it automatically, so large libraries are processed hundreds of times faster than on CPU alone.
Private by design
Runs on your own servers. Your documents and graph stay in your environment; the only outside call is the AI provider you choose for Ask.