Skip to content
AgentraApp

Agentra docs

Models

Gemini, OpenAI, and Claude sit on the same loop as the rest of the stack.

On this page

One loop

Gemini, OpenAI, and Claude sit on the same loop as GMGN, OpenClaw, Solana, and Helius. The mark row on the docs home keeps all seven moving for that reason. They are not three products glued beside the agent. They are readers the runtime can use while it works the trench.

One loop means a close is still one close. A model does not open a second vault, and it does not settle on a second network. Whatever a model contributes is in service of the same cycle OpenClaw is already running: a name, a size from the vault, a buy, a sell, a contract on both rows.

When a model is in the loop

A model is in the loop when the runtime needs a read that is larger than a pair payload. Gemini is the crowded tape. OpenAI is the prompt turned into one route. Claude is the longer memory of the vault and the recent closes. Those roles are separated on their own pages so you can see what each one is for, and so you can see what none of them are for.

The runtime remains OpenClaw. A model can inform a cycle. It does not replace the mode you set. If trench mode is on, the loop continues. If you prompted a single route, the loop owes you that route and not a silent change of policy. The model pages assume that order: mode first, model as a reader inside it.

The marks

You will see the three marks anywhere the stack is shown, including the carousel at the top of this introduction’s home page. The carousel is a reminder, not a settings panel. There is no switch on the desk that says “use only one model for the next hour.” The loop is shared. If a later control is added, it will be documented here as a control, not as a logo.

What a model does not hold

A model does not hold the SOL. The vault holds the book. A model does not hold the mint. The sell row holds the contract. A model does not hold the win rate. The settlement rule holds the rate at 68 percent by how wins and losses are chosen. Confusing a fluent paragraph with a signature is how people misread agent products. These pages refuse that confusion.

Read the three pages next if you want the distinct job of each model: Gemini, OpenAI, and Claude.