Own · Run · Connect
The system of record for your AI work.
Your conversations, context and provenance belong to you and your team — independently of any provider. Run them on any model: commercial, open-source, or local. OneAI is the permanent layer; models are interchangeable engines underneath.
OneAI is in its v1.0 testing phase. The workspace is real and usable with your own API keys; billing is not switched on and no one is charged yet. Signup is invite-only for beta users.
Own
Your AI work is the asset. It should belong to you.
Conversations, prompts, context, history and provenance live in OneAI — not inside whichever provider you happened to use that month. Every answer records which model produced it, from which context, at which version. Export everything, any time.
Run
One workspace, any model underneath.
Frontier models, open-source models, your company's own models, and models served from your own hardware. Switch model mid-conversation without restarting it: the history stays, and OneAI prepares the context the new model needs.
Connect
Models never run bare.
They run connected to your context and history, working alongside other models and other people — shared team context, saved prompts, and later agent flows on your own machines. Everything lands back in the same record.
The problem
Every conversation deepens someone else’s lock-in.
The durable asset in AI usage is not the model. It is the accumulated conversations, context and history around the models — and today that asset sits inside whichever provider you happen to use.
Switching models means abandoning history, re-teaching context, and losing any record of what was sent where. Models get better every few months; your history cannot follow them.
OneAI inverts the arrangement. The workspace, the context and the history are the permanent layer. The models plug in underneath.
What that buys you
Transparency and versioning, not just a chat window.
See what was actually sent
The inspector shows what you typed next to what OneAI actually sent to the model, and why: which instructions, which context items, which versions. It reads the stored record — it does not re-derive what today's code would send.
History that never silently rewrites
Context items are versioned and every interaction records the versions it used. Change a project's context tomorrow and last week's conversation still shows the version it actually ran on.
Bring years of history with you
Import your ChatGPT and Claude exports and they become first-class OneAI conversations — searchable, and reusable as context. A documented generic import format covers everything else.
Deterministic control at the input
Mark context and conversations sensitive, and restrict which models they may reach. Teams set a model allowlist. What may be sent is enforced — and we do not pretend that filtering what a model answers is the same guarantee.
Teams
Shared context, with control over what leaves the room.
A team’s working knowledge — standards, product context, terminology, reusable prompts — lives as versioned OneAI context that every member’s conversations can draw on, on any model.
Four roles (Owner, Admin, Member, Viewer), a per-team model allowlist, and sensitivity rules that keep marked content away from models it should not reach. An activity log records who did what — structure and actor, not prompt bodies.
Because the record already stores what was sent where, under which policy and which context version, auditing is a reading problem rather than a new system.
To be clear
What OneAI is not.
Not ChatGPT with a model dropdown
Multi-model chat is a commodity. The product is the owned, versioned, inspectable record around the models — the part that compounds.
Not a gateway or a router
Moving requests to providers is one verb of three, and we do not compete on automatic model selection. OneAI can sit in front of a gateway; it is not one.
Not a coding agent
Developers are welcome and Git-linked history is on the roadmap, but OneAI does not become an IDE or fight coding agents on their home turf.
Start with your own keys.
Bring your own provider keys and use the full core workspace on the Free plan. Nothing is billed during the testing phase. Signup is invite-only for beta users.