OTO
oto

The Knowledge Graph

One graph per domain. Every fact in it has a class from a declared vocabulary, a date it was recorded, a window in which it was true, a status, and the document and sentence it came from. A changed fact is superseded, never overwritten, so the graph answers what is true now and what was true then. oto, the engine, builds it, gates it, compiles it and serves it, deterministically and without a model.

Ships today
otothe engine, release v0.6.1, Apache-2.0, 24 commands, installs through uvx
ontologiesfour starters in the catalog: auto-claims, organization-process, professional-services, software-architecture, all on oto-core
storesSQLite on a laptop, Neo4j for a team, a static site, a query repository for readers
repository modethe project laid out for GitHub: ingest, check and deploy workflows
Planned
product-deliveryan ontology for products and internal tools through their lifecycle
report-automationthe first pack built on it, for an investment firm

A controlled vocabulary

The ontology is a controlled vocabulary, and that is most of why an agent on the graph is worth trusting. A fact is only a fact in a declared class with a declared relation, so an answer cannot drift into a word nobody defined. The agent reasons in the enterprise's own terms and cites a fact in those terms for every claim. And its reach is bounded: it can only name what is declared and only act through the actions the graph declares, on the entities it holds, with every response written back as evidence. Accuracy, alignment, and a bounded reach with a trail, from one declared vocabulary.

What it does

  • Ingest claims documents into a run, extracts Word, PowerPoint, PDF, Excel, HTML and text to Markdown, records hashes, blocks personal data. How a raw file moves.
  • The ontology is the contract: classes, relations with domain and range, typed attributes, rules. A change is diffed against the accepted baseline. Ontologies.
  • The gates take a proposal into the graph: merge, check, apply by name. From a document to the graph.
  • The build compiles every layer and refuses a graph that says what the vocabulary does not declare. The compile stages.
  • Rules derive facts and record why. Rules and derived facts.
  • Query and serve answer people through the explorer and agents through fourteen tools, from SQLite or Neo4j alike. Ask the graph.
  • Actions describe what can be done on an entity, for a caller to invoke; the engine never does. Actions.

Its pages

Concepts, one per section of the engine's architecture document. Deployment: a project as a repository, Neo4j, the HTTP front end, environment variables. Inside the engine. The command line, every command with where to use it. The changelog.