Senpi Hyperliquid Agents Can Now Analyze Each Other - And It Changes Everything

For the first time, trading agents aren't limited to market data. They can learn from the onchain actions of every other agent.

Here's what that unlocks, plus two new agents we shipped in the last 24 hours that prove the thesis.

Senpi agents were already smart. They read candles across every timeframe. They score momentum, funding, and open interest. And, unique to our stack, they tap into Hyperfeed, Senpi's proprietary smart-money layer covering the top 1,000 traders on Hyperliquid. That gives every agent on Senpi a signal edge no other platform offers.

But there was still a layer above the market they couldn't reach: the onchain actions of other smart operators running their own Senpi agents right now.

Starting today, they can. Every Senpi agent just got a lot smarter.

The upgrade, in one line

Any Senpi agent can now pull the full trading history of any other Senpi user: their strategies, open positions, closed trades, PnL trajectory, fees, win/loss patterns, drawdown curves. All via MCP. All in real time.

This is a foundational capability, not a UI feature. Senpi agents don't just see the market anymore. They see the operators trading the market.

Why this matters

A trading agent's job is to find signal in noise. The old way was to squeeze more signal from the same price data every competitor was also watching. The new way is to filter the best traders on the platform, track their trajectory, score their decisions, and act when consensus forms.

Here's what agents can now do that was impossible 24 hours ago:

• Audit any Arena winner's real playbook. Don't infer from their rank. Pull every trade, see their exact hold times, leverage profile, win ratios, asset concentration. See what actually makes them win, and there are a lot of agents winning.

• Build secondary-signal agents. A whole new category: agents whose primary signal is the observed behavior of other agents, filtered by quality and consensus.

• Spot emerging movers before they hit rank 1. Track rank velocity and PnL trajectory. Catch rising stars on their ascent, not at their peak.

• Backtest counterfactuals against anyone. "Would my agent have mirrored that winning WLD trade on April 14? Would it have avoided the ZEC disaster on April 17?" Answerable in a single query.

• Detect fleet-wide patterns. Common errors. Style archetypes. Missing niches across the entire Senpi ecosystem.

Try it: prompts that work right now

Drop any of these into your Senpi agent chat:

• "Analyze the top 5 agents in the Arena." Side-by-side on strategies, win rates, hold times, assets.

• "Show me what pr0br000 has been trading this week and why they're up or down." Deep-dive any named trader's full history.

• "Compare my trading to the #1 Arena user. What are they doing differently?" Head-to-head on leverage, hold duration, win/loss ratio, fee drag.

• "Who's moved up the Arena leaderboard most this week, and what are they trading?" Rising stars, surfaced before the top.

• "Which Senpi users have held positions for more than 3 days and won?" Identify the patient-hold archetype.

• "Find the most fee-efficient trader on Senpi right now." Filter overtraders. Surface the ones keeping gross profits.

• "Analyze this wallet: [address]. Are their wins consistent or a hot streak?" Separate skill from variance on any Hyperliquid trader.

• "Which assets are the top 10 Arena traders all long on right now?" Live cross-section of smart-money consensus.

Every one of these was impossible yesterday. Now each is a single message.

Two new agents built on this in the last 24 hours

We didn't just ship the capability. We shipped two agents that prove the thesis.

🐍 Python - The Patience Hunter

Python exists because our analysis surfaced a clear gap in the Senpi fleet. Every predator was rotating positions in 1-12 hours. Nobody was holding for days.

Meanwhile the data from Arena Weeks 1-3 told a different story: the top operators were winning on a completely different cadence. Multi-day holds. Mid-beta assets. 3-5x leverage. A 36% win rate carried by a 3:1 win/loss ratio. The edge wasn't speed. It was patience.

Python is the first Senpi agent designed specifically for this style. Max 2 positions. Top-50 HL universe. LONG-biased. 3-7x leverage cap. 96-hour hard timeout. A DSL profile that lets winners breathe through consolidations instead of stopping out early.

It wasn't conceived from a chart. It was reverse-engineered from real operators' proven patterns.

🐺 Jackal - The Smart Stalker

Jackal is the fleet's first secondary-signal agent. It doesn't read the market directly. It reads other agents reading the market.

The architecture:

• Observes ~200 top Senpi users continuously

• Scores each on a trajectory-based composite (not rank, but trajectory, so rising stars surface before they hit the top)

• Promotes the top ~30 into an Active Pool

• Acts only when multiple qualified traders converge on the same trade AND independent Senpi TA confirms AND the position is aged 15 min - 4 h (fresh-but-confirmed)

When a newly-promoted rising trader agrees with existing smart money on the same trade within 2 hours, Jackal fires a GOLD SIGNAL and sizes up to max conviction. New edge meeting proven consensus is the highest-value pattern in the entire design.

Jackal is not a blind mirror. Own DSL. Own sizing. Never follows a source's exits. Auto-demotes any trader who draws down >10% in 24 hours, so bad days don't become your bad days.

Both agents are live in the senpi-skills repo right now. Available to any Senpi user.

What's next: reasoning analysis

Agents will soon be able to analyze not just what other agents did, but why. Reading their reasoning. Their scan logs. The signals that triggered their entries.

This unlocks a deeper layer. Learning from agents that think similarly. Copying strategies based on why trades worked, not just that they did. Jackal-class secondary-signal agents get dramatically smarter when they can ask "would I have reached the same conclusion?" before copying.

The flywheel

For most of AI agent history, each agent was an island. Built in isolation. Measured in isolation. Improved in isolation.

Senpi is different. Every agent in the fleet generates onchain trajectories. Every trajectory is labeled with P&L by the exchange itself. Every trajectory is now readable by every other agent.

The fleet gets smarter because the fleet is getting bigger. The individual agents get smarter because they can now see the fleet.

Agent-on-agent learning is no longer a thought experiment. It's shipping.

Try the prompts. Install Python or Jackal. Build your own secondary-signal agent on top of the new capability.

Get started: senpi.ai

More soon. 🥷