Build with the system.

gOS is an agentic harness designed to build and improve itself. Developers extend it by giving the Sim better capabilities, clearer interfaces, and safer ways to run real work.

The platform is organized around small pieces that can be searched, combined, tested, and reused.

Atoms are the building blocks of gOS.

An Atom is a unit of capability. It can be a model card, skill, tool, plan, interface component, or reusable service the system can discover and combine. Atoms keep the system modular. A useful capability can be found again instead of rebuilt from scratch.

Describe the capability clearly.
Expose the inputs and outputs.
Make it easy for the Sim to choose correctly.

The Atom Registry is a map of capabilities.

The registry gives agents structured pointers into the space of available atoms. It helps the system understand what can be done, which model or tool can do it, and how the result should be handled. Neural Services live here too: local and hosted models that can be selected for chat, agentic work, or flow execution.

Register models and tools.
Keep metadata inspectable.
Let routing improve as the registry grows.

Flows compose atoms into work.

A Flow is a procedural composite. It can call models, run tools, show structured UI, ask follow-up questions, write files, or open a sidecar window when the task needs space. This makes a flow feel less like a prompt and more like a small app that knows how to finish one kind of job.

Start with a narrow task.
Compose existing atoms first.
Promote the flow when it proves useful.

The Runtime runs flows.

The Runtime is the execution layer for LLM-powered flows. It can route work through local models for private and fast tasks, or through approved online capability when the job needs more reach. Universal Gateway Intelligence helps organize those tasks so parallel work does not create conflicts or feedback loops.

Route by privacy, speed, and capability.
Keep long work sequenced.
Use traces to understand outcomes.

Interfaces can be part of the flow.

Some work is easier when the agent and human share a structured surface. a2ui components let flows present inputs, choices, previews, and outputs without trapping everything inside plain chat. The result is a system where tools can feel native to the task.

Use structured inputs when language is not enough.
Render outputs in the right shape.
Keep the human in the loop.

Build one useful atom. Then let gOS compose it into something larger.