OTO
oto

Knowledge shipped like code

The best-kept knowledge in any company is its code, because code has a repository, a review, a test suite and a deploy. OTO gives knowledge the same treatment, and oto has a mode for it.

Repository mode

oto init --name "Acme Claims" --ontology auto-claims --project claims --repo

The project is laid out to live in a GitHub repository: workflows for ingest, checks and deploy, a CLAUDE.md for the agent, an .mcp.json for the query tools, and a corpus that is committed.

A document is pushed to inbox/by a person, a connector, or the capture skill after a conversation
↓
The ingest workflow runsoto ingest claims the run, extracts the text, blocks personal data
↓
The agent authors on a branchone proposal per document, through the dry run and the check, and opens a pull request
↓
The pull request is the gatethe check workflow reruns the gates; a named person reads the diff and merges
↓
The merge deploysoto build, the store loaded into Neo4j and verified, the query repository pushed for readers

Every fact that reaches production went through a diff a person read. The ledger says who merged it and what became answerable. oto curate undo and git history agree.

What deploys where

Reader Store How it gets there
a developer on a laptop the SQLite build oto build
a team, and the agents that serve it Neo4j oto build --target neo4j --verify on merge
readers who need answers, not documents the query repository oto publish, then oto sync on their machine
anyone with a browser a static site oto build --target site, hosted anywhere

The two live stores answer alike, and an equivalence test in the engine holds every new query tool to it.

Corrections travel the same road

A correction said in conversation is written by the capture skill as a dated source in inbox/, committed and pushed. The pull request that follows carries the fact, names the speaker, dates it to the date given and not to today, and supersedes the fact it contradicts with a change note in the speaker's words. A correction made in conversation does this by hand.

Everything becomes a graph

Policy, infrastructure, the product portfolio: anything a team writes down and reviews can be shipped this way, as a graph in a repository with an ontology at the gate. The product lifecycle example is one; the report automation example is a factory built on it.