OTO, the product
OTO, Ontology-To-Operations, turns what a company already has, its documents and its conversations, into knowledge graphs it can trust: one per domain, every fact dated and cited, compiled like code and served to people and to agents. From enterprise meaning to running agents: it is a system of four parts, published by Cynergis, and you use all four.
The four parts
A controlled vocabulary. The ontology is a controlled vocabulary: the agent understands, reasons and acts in terms the enterprise declared, with a reason and a confirmer on each. That is what makes it accurate, a fact is only a fact in a declared class; aligned, it cites the enterprise's own terms; and bounded, it can only name what is declared and only act through what the graph declares, leaving evidence.
Ontology to operations. A graph that only sits there is a library; OTO's is built to act. The ontology is the contract between what the enterprise means and what its agents do, and the same graph declares what can be done, on which entity, through which tool: an agent carries it out and the response is written back as evidence. Next comes an ontology that carries flows as well as actions, so a process is a graph that runs. Actions says how far that goes today.
The engine is deterministic and needs no model. The Agent brings its own. The Verifier is partly the engine's gates and partly agents. The Studio reads the graph the engine serves. Take one away and the others lose their point: a graph nobody reads, an agent nobody checks, a check on nothing, a face on nothing.
How a team uses it
From a folder of documents to answers, in one pass. Each step names the part that does it and who is accountable.
| Step | What happens | Part | Who |
|---|---|---|---|
| 1. Install | the plugin from the registry, with a pack for the domain | the Agent | you, once, in Claude Code or your harness |
| 2. Start | a project from the pack's ontology; the session hook says where it stands | the Knowledge Graph | the pack's start skill |
| 3. Ingest | documents into the corpus, extracted, hashed, scanned for personal data | the Knowledge Graph, the Verifier | the engine |
| 4. Interview | the vocabulary, from the questions the graph must answer; every class with a reason and a name | the Agent, the Verifier | the agent asks, you confirm |
| 5. Propose | one proposal per document: the facts, their classes and relations, the sentence for each | the Agent | the agent |
| 6. Gate | the dry run, the merge, the check, the diff; apply by name | the Verifier, the Knowledge Graph | the engine refuses, you decide |
| 7. Build | the graph compiled into the store, refused if it says what the vocabulary does not declare | the Knowledge Graph | the engine |
| 8. See | the explorer, the reader, or the domain's own view | the Studio | you |
| 9. Ask | a question in your own words, answered with the document and the date | the Agent | the agent, through the query tools |
| 10. Correct | a change said in conversation becomes a dated source and supersedes the old fact | the Agent, the Verifier | the agent captures, you confirm |
| 11. Measure | a gold set, the bench, feedback promoted into tests | the Verifier | the evaluate skill |
| 12. Ship | the project in a repository: a push ingests, the pull request is the gate, the merge deploys | all four | the team, like code |
Getting started walks this path. The tutorials do each piece by hand, with the output the engine printed.
Many graphs
A graph per domain, owned by the people who know it, joined by shared vocabulary: that is a graph per domain, and the mesh. Shipping each graph from a git repository with the pull request as the gate is knowledge shipped like code.
Where the pages are
| Part | Its pages |
|---|---|
| The Knowledge Graph | Concepts, Deployment, Inside the engine, the command line, the changelog |
| The Agent | Install in Claude Code, the skills, packs, from a document to the graph |
| The Verifier | the gates, bench, feedback, vet-provenance, evaluate |
| The Studio | views, the pending lane and the preview, the lifecycle screens |