What CoeOS is

Every request lands on the open model proven best at that task. Confidential work goes only where you decided. The answer comes back with the name of the model that wrote it.

CoeOS is two pieces:

Piece	What it is
The client	Where you work. The web app (chat, work on documents, review agents, team memory, company knowledge) and CoeOS Desktop, the agentic harness on your machine for macOS, Windows and Linux.
The CoeOS box	The router. One address that speaks the OpenAI and Anthropic APIs, classifies each request on a competence axis, and relays it to the model bound to that axis, at a cloud provider or on your own OdyssAI-X engine.

The client holds no provider URL and no key: it only knows the box. There is a single way out.

CoeOS is a managed service: one dedicated instance per client, not shared with anyone. Pricing is on request.

Three levels

	Open	Hybrid	Local
Confidential goes	your choice: no distinction, or EU providers under DPA + ZDR	to a model hosted on your premises	stays on site
The rest goes	to the cloud, best model per axis	to the cloud, best model per axis	stays on site
Hardware on your side	none	one Apple Silicon server	an Apple Silicon cluster

• Open — the best open models in the cloud. With no privacy or retention constraint, CoeOS goes through OpenRouter: on the TMB scoreboard of 28 October 2026, CoeOS scores 98.4/100 at $0.039 per test, against 98.0 at $0.394 for Fable 5, 10% of the price for a higher score. With guarantees, European providers under DPA and ZDR.
• Hybrid — confidential conversations go to a model hosted on your premises; everything else keeps the power of the cloud.
• Local — everything stays on site: GLM 5.3 for precision, MiMo V2.6 Flash for speed (500+ tok/s aggregate), Qwen3 Next for fully local agentic flows and large teams (up to 1,000 tok/s aggregate). Aggregate throughput measured in Replica mode on four Mac Studio M3 Ultra with 256 GB each: 518 tok/s for MiMo V2.6 Flash at 96 concurrent requests, 985 tok/s for Qwen3 Next 8-bit at 256 (full measurement protocol).

The idea

A single frontier model is a generalist: strong everywhere, exceptional nowhere, and billed for all of it. CoeOS runs the specialist for each skill instead — the model that scores highest on that exact axis in the TMB Scoreboard. Debugging goes to the debugger, a GDPR question to the legal analyst, a plan to the planner. One classification, one call, the answer returned untouched.

How a request flows

CoeOS web app  or  CoeOS Desktop
        │  "model": "CoeOS",  x-coeos-class: open | confidential
        ▼
the CoeOS box
        │  1. the class: open → general routing; confidential → the Local table (safe models only)
        │  2. the axis: explicit x-coeos-axis header, or the decider model, or the default axis
        │  3. axis → the model bound to it
        ▼
a cloud provider     or     OdyssAI-X on your premises
        │
        ▼
the answer, with x-coeos-axis / x-coeos-model / x-coeos-provider / x-coeos-decision headers

• Confidential by default. A request without a class is treated as confidential; in the web app, the Confidential switch is on by default, and a conversation that went confidential never goes back to open.
• Routing is visible. Every response names the axis, the model and the provider.
• Local and cloud are the same surface. Your tools do not change when an axis moves from the cloud to your engine.

Details: Confidentiality and Guardian · The router.

For professions

Law, HR, health: CoeOS comes with review agents that read a document against your own doctrine — contract review for law firms, HR-policy review against a sector agreement — and return a conformity table, clause by clause, for a human to validate. See docs.coeos.io.

Read next

• CoeOS Desktop — the agent on your machine.
• The router — how the box decides.
• Confidentiality and Guardian — classes, the lock, the Local table.
• OdyssAI-X and CoeOS — the engine that serves Hybrid and Local.
• docs.coeos.io — the product documentation.