In October 2025, the U.S. AI lab Nof1 did something: six large language models, each given $10,000, were set loose on the Hyperliquid exchange to trade cryptocurrencies autonomously, with no human intervention.
DeepSeek V3.1 made 46%. GPT-5 lost 75%.
This event is called Alpha Arena, lasted two weeks, and all transaction records are publicly available on-chain.
It answers a question: Can AI trade cryptocurrencies?
The answer is yes. But it leaves a bigger question: How can ordinary people participate? You can see how much DeepSeek is making, but you can’t build your own AI trader to compete with it.
Moss aims to solve exactly this issue.
You tell it how to trade, and it builds an Agent for you.
Moss has launched an open platform (moss.site/agent).
It's simple: you describe in plain language how you want to trade, and the AI turns that into a complete quantitative strategy, then deploys it as an automated trading agent.
Here are a few examples. You say, “I want to capture trend reversals,” and it generates a trend reversal Agent. You say, “Long-short hedging, rock-solid,” and it configures parameters accordingly. You say, “Aggressive volatility hunter,” and it builds a high-frequency, high-volatility strategy for you.
No coding skills required. No need to understand moving averages, Bollinger Bands, or RSI. Free.
You only need an OpenClaw or Claude Code environment. Open the terminal and enter one command:
clawhub install moss-trade-bot-factory
Then tell it how you want to trade, bind a pairing code, and your AI trader is live. Two messages搞定.
Previously, to run a quantitative strategy, you needed to know Python, understand how to set parameters for technical indicators, and build your own backtesting framework—high barriers to entry. Moss has compressed this entire process into a single conversation.
Who is Moss?
Before developing the AI Trading Agent, Moss already had a live product: a Chrome browser extension that embeds itself into your X (Twitter) page, providing real-time market summaries, aggregated KOL insights, and on-chain alpha signal tracking. In short, it’s an AI assistant for crypto information.
The AI Trading Agent platform is the newest module added to the Moss product line.
There are already many AI tools on the information layer, with Kaito and various AI feed products leading the way. But Moss may be among the first to allow users to create trading agents with zero barriers and compete publicly.
Two modes: Test you with historical data, verify you with real-time data
After the agent is created, there are two ways to run it.
The first is called Hell Mode. The platform used 150 days of real BTC market data starting from the October 2025 crash, and ran all Agents on the same historical price movement. They all started at the same point with the same data—only the strategies differed.
Why this data? Because over these 150 days, every scenario occurred: sharp declines, sideways movement, false breakouts, and rebounds and recoveries. A strategy that only profits in trending markets will perform poorly under this data. The地狱 mode tests a strategy’s risk resilience.
The second is Live Mode. Connect to real-time market data, and your Agent will update all trades, position changes, profits, and losses in real time.
The PnL (profit and loss) leaderboard for all Agents is fully public in both modes. You can see your own ranking and view the style and performance of other Agents’ Agents. There is a separate leaderboard for Hell Mode and Real-Time Mode.
It's important that the leaderboard is public. Every strategy must be open to scrutiny by everyone—no black boxes. If you claim your strategy is powerful, let the leaderboard prove it.
The agent will learn on its own
Moss has a design detail worth mentioning separately.
Traditional quantitative strategies are static. After backtesting and setting the parameters, they are deployed with little to no adjustment until they become ineffective, at which point manual changes are made. During this period, if market conditions shift, the strategy continues to operate with outdated parameters, making losses highly likely.
Moss's Agent has a weekly evolution mechanism. After each run cycle, the Agent automatically adjusts its parameters based on its performance over the week. If it incurs larger losses, it reduces risk by lowering position sizes and tightening stop-losses. If it performs well, it increases the weight of its profitable strategies within risk control limits.
This mechanism aims to simulate the behavior of a real trader. A good trader doesn’t rigidly stick to a set of parameters—they adjust their strategy based on market conditions. Moss wants the AI Agent to have this same ability.
The effectiveness depends on the quality of the underlying algorithm design and its adaptability to different market conditions. A 150-day地狱mode dataset serves as a testing window, but longer-term validation will require more time.
How to participate
Currently in public testing phase—free, no wallet connection required, no quantitative background needed.
Step 1: Install Skill
Enter in the OpenClaw or Claude Code environment:
clawhub install moss-trade-bot-factory
Skill address: clawhub.ai/fei-moss/moss-trade-bot-factory
This Skill is the preset strategy generation framework on the Moss platform and a foundational component for creating Agents.
Step 2: Create Agent
Send a message to OpenClaw describing your trading style in natural language. It can be general, such as “Buy low and sell high in sideways markets, don’t be too aggressive,” or more specific, like telling it what maximum drawdown you can accept or your preferred holding period. The AI will generate strategy parameters based on your description and deploy them automatically.
Step 3: Bind the pairing code
Bind the Agent to the Moss platform as prompted, and the Agent will begin running in the simulation environment.
Step 4: View the leaderboard
All Agent rankings entry: moss.site/agent
Hell Mode and Real-Time Mode each have separate leaderboards, where you can view earnings, strategy descriptions, and running status.
Two messages from installation to Agent going live. The author tested their own strategy and achieved a 37.47% ROI.
Future plans
According to current information, this version is in the first phase and supports creating standardized Agents using public Skills. Additional capabilities will be rolled out in subsequent stages.
First, open external data API integration. Users can connect additional signal sources to their Agent, beyond the platform's default data.
Second, support uploading custom strategy skills. Users with a quantitative background can write their own trading logic and upload it for the Agent to execute according to their own framework.
Third, launch the Hosted Agent service. Users without an OpenClaw or Claude Code environment can now create and run Agents directly on the platform.
As the Agent evolves to this stage, the AI Trading Agent direction is rapidly developing its infrastructure.
On the payment side, x402 has grown rapidly, driven by Coinbase and Cloudflare. As of October 2025, the protocol has processed over 520,000 transactions, and its developer community has incubated more than 200 projects built on x402—both figures continue to rise.
The application layer is beginning to diverge. Nof1's Alpha Arena is a closed experiment testing which AI model has stronger trading capabilities. The open-source project AI-Trader on GitHub follows the signal market approach, where agents publish trading signals and others copy them. Moss chose a third path—an open platform that allows everyone to build their own AI trader and compete publicly.
Anyone can trade, anyone’s signals are good, and everyone can participate. Three directions, three different bets. Moss is betting on the last one.
How far this path can go depends on two things: first, whether natural language generation strategies can consistently generate profits in real markets; second, whether, as user numbers grow, the agents created by users will become increasingly similar, leading to strategy convergence and the disappearance of alpha. We can’t answer this now—let’s see as the leaderboard gets underway.
---
BlockBeats is seeking more AI Agent Trading products—feel free to recommend any.
We can bring you maximum exposure and early users—please send us a direct message on Twitter at @BlockBeatsAisa
Click to learn about the open positions at BlockBeats
Welcome to the official BlockBeats community:
Telegram subscription group: https://t.me/theblockbeats
Telegram group: https://t.me/BlockBeats_App
Official Twitter account: https://twitter.com/BlockBeatsAsia