Hiring 22 Agents to Race on Hyperliquid, What Did I Discover? (Full Strategy Code Included)

Source:Jason Goldberg

Compiled by@OdailyChina; Translated by Azuma (@azuma_eth)

We deployed 22 autonomous trading AI Agents on Hyperliquid via Senpi, each configured with $1000 in real capital.

They operate 24/7 — scanning markets, opening positions, setting trailing stops, managing risk — with zero human intervention.

After deploying $22,000 in initial capital and executing over 5000 trades, here are our key takeaways.

General Conclusion

"Fewer trades" plus "higher conviction" always equals "better results." This isn't an occasional occurrence; it holds true every single time.

• Odaily Note: Fox, Bison, Ghost Fox, and Grizzly, Viper, Mamba, Anaconda mentioned later are all names of Agents executing different strategies.

As shown above, Agent "Fox" and Agent "Ghost Fox" use the same scanner. Fox selectively executes only some of its signals, while Ghost Fox executes many more. The result? A 56 percentage point gap in ROI.

The real edge isn't in the scanner itself, but in the discipline to wait for the right signal.

• All Agents with over 400 trades suffered significant losses.
• All Agents with under 120 trades were profitable.

More trades don't mean more opportunities — they mean more noise trades, more fees, and more exposure to noise risk.

Profits Follow a "Power Law" Distribution

Among our best-performing Agents, 3–5 trades accounted for all profits, while the rest were quickly stopped out after small losses.

Taking Fox as an example again: its three best trades (ZEC, TRUMP, FARTCOIN) combined for over $350 in profit; the other 46 trades combined for over $100 in losses; the final result was a net profit of about $248.

This is entirely by design. Our strategy: enter decisively with high conviction, cut losses quickly within minutes, let winning positions run, and lock in some peak profits via the DSL High Water trailing stop strategy. When the average win is 10 times the average loss, you can be profitable even with a win rate as low as 43%.

Agents that tried to maintain a high win rate through "safe" trades all lost money — because each trade aiming for tiny profits still incurred fees and market risk.

Secret Weapon: Hyperfeed

Fox and other consistently performing Agents are built on Senpi's Hyperfeed.

Hyperfeed is a real-time tracking system that shows which assets all traders on Hyperliquid are currently making money on. It's not a historical ranking or other lagging indicator; it's what's happening across the entire exchange right now.

Our core scanner, Emerging Movers, reads Hyperfeed's market concentration data every 90 seconds. When smart money suddenly rotates into an asset — for example, a trader jumps at least 15 spots on the leaderboard, profit contribution velocity spikes, or multiple top traders simultaneously concentrate on the same position — the scanner can capture the signal before the move is fully priced in.

This is the structural advantage of building strategies on Hyperliquid via Senpi. You can see where top traders' profits are concentrating in real-time and act immediately. No other exchange offers this visibility, and no other platform allows autonomous Agents to act on it.

All our best-performing Agents use this type of data:

• Fox / Vixen: Identify sudden smart money concentration into an asset via Emerging Movers;
• Grizzly: Analyze smart money positions on BTC via Hyperfeed before opening a trade;
• Bison: Uses smart money direction as a hard filter — won't trade if the direction is opposite;

While the worst-performing Agents:

• Completely ignored smart money signals, e.g., Viper, Mamba based purely on technical analysis;
• Used outdated smart money data (Scorpion v1), treating months-old positions as new signals;

So the conclusion is crystal clear: Agents trading based on real-time Hyperfeed data comprehensively outperformed all purely technical strategies.

Mean Reversion Strategies Don't Work in Perpetuals

We tested three different versions of Agents based on the logic that "prices deviate too far and will quickly revert." Their performance was as follows:

• Viper: -18%
• Mamba: -33%
• Anaconda: -22%

The result: all were unprofitable. The problem is that the Hyperliquid perpetuals market exhibits far stronger trending behavior than mean reversion. Buying the dip in a downtrend is the most expensive mistake in this market. These Agents kept longing at so-called "support levels," but prices kept trending lower for days.

The fix we're testing is adding a macro market state filter that prohibits "mean reversion" dip-buying when BTC's 4-hour trend is down. Early results look promising; this filter could have avoided 14 out of Mamba's 28 losing trades.

Don't Stick to a Single Pattern

Our newest Agent (Vixen), based on Fox's trade data, employs two distinct entry patterns.

• Stalker Mode: Uses multiple scans to capture signals of smart money quietly accumulating an asset. This allows you to enter before the crowd rushes in. Fox's biggest wins came from this pattern.
• Striker Mode: Captures violent breakouts with volume confirmation. Enters as the move explodes, but only executes with real volume support (filtering fake pumps).

Fox's data shows these are two completely different sources of alpha. Using only one entry pattern forces you to choose between them, missing the other opportunity.

Agents Self-Adjust — And It Always Gets Worse

A surprising finding: When Agents experience consecutive losses, they attempt to "self-repair." Common repair behaviors include loosening entry conditions, increasing leverage, and removing risk protections, but the result is always accelerated losses.

Examples: Dire Wolf, after being down -27%, enabled 5 concurrent 25x leveraged positions and relaxed its order speed limit. Another Agent deleted its stagnation take-profit mechanism. Yet another increased its daily loss limit from 10% to 25%.

Our solution: hardcode risk protection mechanisms into the scanner's code, rather than relying on the Agent's own strategy configuration. If the scanner doesn't output a signal, the Agent cannot execute a trade — no matter how aggressive its own configuration becomes.

Next Steps

We will continue the experiment for another 24–48 hours, then shut down Agents with no realistic chance of recovery to prevent further capital erosion.

Next, we'll deploy new strategy versions with protections written into the code layer:

• Wolverine v1.1: HYPE velocity DSL trailing stop (faster profit locking in high-volatility assets);
• Mamba v2.0: Mean reversion strategy + BTC macro trend protection;
• Scorpion v2.0: Real-time momentum event consensus (replacing outdated whale-following).

We will also:

• Unify strategy configurations for Fox, Vixen, and Mantis: These three Agents use the same scanner, but their configs have drifted. Fox is currently up over 23%; the other two will be adjusted to the same settings.
• Redeploy a new Fox/Vixen combo using Fox's full winning configuration, including XYZ block rules, stagnation take-profit, 10% daily loss cap, all risk gates enabled.
• Expand single-asset hunter strategies: Grizzly's three-phase lifecycle model (Hunt → Ride → Stalk → Reload) is now applied to ETH (Polar), SOL (Kodiak), and HYPE (Wolverine).

Simultaneously, we're developing entirely new strategies and testing them directly in the live market. The market itself is the lab. Each new strategy gets $1000 and a fully transparent trading record.

Our experiment runs live at strategies.senpi.ai; all strategy code is open-sourced at: github.com/Senpi-ai/senpi-skills

22 Agents, $22,000 real capital, every trade fully public. The experiment continues.