OTO

The engine in ten minutes

This is the engine on its own, for a person at a terminal who wants to see what oto does before involving an agent. Getting started is the path with the agent: install the plugin, build a graph from a folder, see it, ask it.

Ten minutes, nothing to install but uv. The engine runs through uvx, which fetches it from its repository on first use and caches it. Every command below was run on a clean machine before this page was published.

1. Run the engine

uvx --from "oto-kg @ git+https://github.com/Cynergis/oto" oto version

If you would rather type oto than the uvx line, install it once:

uv tool install "oto-kg @ git+https://github.com/Cynergis/oto"
oto version

The rest of this page writes oto; prefix with the uvx line if you did not install.

2. Register the catalogue and fetch a pack

A pack is a domain ready to use: an ontology, the skills that know how to use it, and its views. The catalogue at cynergis.github.io/oto-registry lists them.

oto registry add https://github.com/Cynergis/oto-registry
oto pack list
oto pack add software-architecture

3. Start a project from it

oto init --name "Acme Platform" --pack software-architecture --project acme

The project holds the ontology, its recorded reasoning, its rules, and a small sample graph so the whole loop works before you add a single document. The project remembers where the ontology came from, so oto status says when a newer release exists.

4. Build, and ask

oto build --project acme
oto query --project acme count
oto query --project acme search "payment"
oto query --project acme explain component.payment-api

The build is a compiler: it refuses a graph that uses a class or relation the ontology does not declare. Every answer cites the document a fact came from and says when the fact was true.

5. See it

oto serve --project acme --http 8765

Open http://127.0.0.1:8765/: the explorer, with every entity's status, dates, source and connections. The same server answers agents over JSON-RPC with the kg_* tools, and oto serve with no --http speaks over stdio for an MCP client.

6. Your own documents

cp ~/docs/platform/* acme/inbox/
oto ingest --project acme
oto survey --project acme
oto status --project acme

From here the work needs judgment, and the skills carry it: the build-knowledge-base skill takes a folder of documents all the way to a served graph, through the four gates every fact passes. Read Install in Claude Code to have those skills and the query tools in your agent, or follow the journeys at a terminal.

What next

  • The tutorials: the complete loop on one real document, with the output the engine printed; then a correction made in conversation, every question the graph answers, and your vocabulary as a pack.
  • The examples: where OTO is used and how, one page per domain.
  • The concepts: what "what is true now" means, and why nothing is overwritten.
  • The command line: every command as the engine describes it.
  • The gates: what each step refuses, and why that is the point.