OpenAI

Place on the loop

OpenAI’s place is the prompt. When you ask the agent for one route, the words you use are not yet a cycle. They are an intention: a name you care about, a size you are willing to see leave the vault, a condition you want respected. OpenAI is the reader that helps OpenClaw hold that intention as a single route instead of a paragraph.

Trench mode does not require you to be eloquent. Prompt mode does require the runtime to understand you. That is why this model sits on the prompt page’s side of the loop, and why Prompt is the desk page to read after this one.

The read

The read is a narrowing. Many words become one route: which mint, whether the vault has the size, whether this is a buy the cycle should attempt now. The narrowing is a success when the console later shows one buy and one sell for that contract, and a failure of scope when a prompt for one name turns into a tour of the trench.

The read should also preserve your last word. If you asked for a single route, the runtime owes you an end to that request. OpenAI helps phrase the request as something that can end. OpenClaw is still the runtime that stops or continues according to the mode. The model does not quietly flip you into trench mode because the tape looks busy.

Shared with the runtime

Shared output is a route the runtime can execute with the vault it already has. It is not a new wallet, and it is not a second book. The SOL that moves is the same SOL the command center was already showing. After the sell, PnL and the activity row update the way any other close updates.

OpenAI shares the loop with Gemini and Claude, so a prompt can still be checked against the tape and against recent closes. The prompt remains the source of the request. The other two models remain readers. None of the three replaces the contract on the sell row.

Boundary

A fluent prompt is not a fill. The desk can show Thinking and then a buy only when a pair is actually in hand. If the latest profiles do not yield a Pump.fun mint, the cycle does not invent one to satisfy the sentence you typed. That restraint is part of the product. OpenAI can clarify the ask. It cannot mint a token that the tape did not return.