Public formats

The generic import format

A documented schema that any conversation history can be converted into, so a provider or tool without a purpose-built importer still has a path in. Imported conversations become first-class OneAI conversations — searchable, reusable as context, with the original model, provider and timestamps preserved where the source includes them. Ready-made importers exist for ChatGPT and Claude exports.

The model connector contract

The interface a model endpoint must satisfy to appear in a OneAI catalog. In practice this is what lets any OpenAI-compatible HTTP endpoint be registered as a model — self-hosted runtimes today, and the same contract that third-party model publishers will serve behind later. See bring your own keys.

Why this page is a placeholder

Both specifications are written and are the authority for anyone building against OneAI, but they live in the specification repository rather than on this site, and publishing a document from one repo onto another is a decision we have not made yet. Copying them here by hand would create a second version that drifts from the real one — and a contract with two versions is worse than a contract with none.

So this page will not paraphrase them, and nothing that is not in the specifications will appear here as if it were. When the publishing route is settled, the actual documents show up at this address.

If you need either specification before then, ask for it directly — the answer is a document, not a wait.

What is not a public format

There is no public OneAI API, SDK or plugin platform in this release. The node and agent protocol — how a machine running OneAI serves models and executes tasks — is not designed yet; when it is, existing open agent-interoperability standards get evaluated before anything proprietary is invented. Neither has a specification to read today, and this site will not pretend otherwise.