Skip to content
AgentraApp

Agentra docs

Introduction

Agentra is a Solana trench agent with its own vault. These pages are the book for the desk.

On this page
GMGNOpenClawSolanaHeliusGeminiOpenAIClaudeGMGNOpenClawSolanaHeliusGeminiOpenAIClaude

What Agentra is

Agentra is a Solana trench agent with a vault of its own. It is integrated with Fomo.family, and it is powered by OpenClaw Agent. The trench is where new names move while the window is still open. The vault is the book those routes spend from and return to. The desk is the window that holds both, plus the feed, plus the closes.

This book is the long form of that sentence. Each subject has its own page, the way a GitBook keeps a chapter from collapsing into a paragraph on a home screen. The stack is not one list. GMGN, Dexscreener, OpenClaw, Solana, and Helius each explain a different job. Gemini, OpenAI, and Claude sit on the same loop, and each of those has a page too, because a shared loop is not the same thing as a shared duty.

The desk you open from the site is the command center. It shows the agent’s phase, the SOL balance, a win rate held at 68 percent, the accumulated PnL, and a live console. Activity, on its own page of the app and in this book, is the paged list of the agent’s buys and sells, ten rows at a time, each row carrying the token contract.

The loop

A cycle is the unit of the product. The runtime thinks, reads the latest Solana profiles, prefers a Pump.fun pair, buys that token, and later sells it. A win adds SOL in the band from 0.1 through the 0.4 range. A loss removes SOL in the band from 0.5 through 1.1. The contract from the buy is the contract on the sell. The vault balance and the PnL move when the sell is written.

You can leave that loop on. That posture is trench mode. You can instead ask for one route and keep the last word. That posture is a prompt. Both are OpenClaw. Both write the same kind of activity row. The choice is whether the next cycle is sponsored by the standing window or by a sentence you typed. The pages for those postures are linked below, and they are deliberately separate so the policy is not a footnote.

Around the loop sit the readers. Gemini faces the crowded tape. OpenAI faces the prompt. Claude faces the longer memory of recent closes. None of them holds the SOL. The vault holds the book. If a page in this book ever sounds like a model signed a transaction, the page has failed, and the vault page is the correction.

How to read this book

The left side is the contents, grouped the way the work is grouped: Start, Stack, Models, and Desk. The right side, on a wide screen, is the contents of the page you are on. Previous and Next at the bottom follow the book order, so you can read straight through from this introduction to the command center without hopping the sidebar.

Read Start if you want the map. Read Stack if you want who does which job. Read Models if you want the three readers on the loop. Read Desk if you want the vault, the family feed, the two modes, the activity list, and the figures. You can also jump. Every page stands alone, and every page links to the neighbors that keep it honest.

Open the desk

Launch app opens the command center. The preview keeps a local workspace in the browser so the desk has an address to show. The figures and the rows come from the trench feed. If the feed is quiet, the desk says it is reconnecting and leaves the last true book on screen. It does not invent a close to fill the gap.

When you are inside, Dashboard is the live stage. Activity is the audit, ten contracts to a page. Prompt agent is the door for a single route. The vault screen is still marked soon. Until it opens, the SOL balance on the command center is the working inventory, and this book’s vault page is the explanation of that number.

If you only read one more page after this one, read The desk. It is the guided tour. The rest of the book is what the tour points at.