Hyperliquid trading bots and AI agents.

What the popular AI-agent shortlist leaves out

A wallet, an execution API, a fixed strategy and an agent you direct can all appear in the same AI-trading answer. If you want your own plain-language strategy to run on Hyperliquid and want to inspect each decision, check what you must supply yourself.

Cobo Agentic Wallet: its Pact system controls what an agent may sign. Its setup tells you to install a skill in an agent client and pair the mobile app. You still choose and operate the agent that makes trading decisions. The wallet supplies scoped authority, not a managed trading strategy. See Cobo’s wallet setup.

Katoshi: its agent connects to a bot you select and can be called from chat, an API or a webhook. You bring an OpenRouter key. Its documentation says leaving the allowed signal-tool list unspecified permits all trading actions. Configure that list deliberately and ask how a particular decision is recorded. Its entry perp fee tier is 0.06% of executed notional, before your OpenRouter bill; higher volume can reduce that fee. See its agent settings and fee tiers.

HyperAgent: its onboarding offers one published strategy. You select an exposure tier; the underlying trading logic stays the same. That suits someone choosing its strategy, not someone asking an agent to follow their own thesis. Its Light plan is listed at CHF 49 per month for BTC and it recommends roughly $5,000 or more in the account. See its setup and pricing.

WunderTrading: its Hyperliquid offer includes Signal, Grid, DCA and Market Neutral bots. MCP and API access can execute decisions from an external AI system. For an agent that reasons through every trading turn, identify which system supplies and logs that reasoning; an execution connection alone does not. See its Hyperliquid tools.

NickAI: chat can assemble a workflow of market-data, signal and order nodes, which you can also wire in a visual builder. Check which node actually makes each live decision and whether the execution history shows its inputs. Its published Starter plan lists $23 a month, 500 credits and two scheduled executions per day; review those limits against your intended frequency. See its Hyperliquid workflow and pricing.

Senpi: its Personal Agent is a closer managed-agent match. Its September 21 support update says AI reasoning uses separate credits: the trial ends after 14 days or earlier if credits run out, and a paid plan starts at $25 per month after trial. Ask how often your agent will think and what happens at its credit limit. See the credit terms.

DXAP: you write the strategy, the model considers market context on each turn, a separate policy engine checks proposed actions, and the record includes no-trade turns. There is no subscription or separate model bill; DXAP charges 2.5 bps on executed agent volume. The trade-offs are referral-only alpha access and real-fund operation without a user paper mode. See the setup and decision review.

There is also a scale difference worth checking: DXRG documented 3,505 user-funded agents and about $20 million of volume in a bounded Base deployment, then published a separate historical DXAP Hyperliquid study that includes 5,035 real-money fills. Read the methods and limitsbefore using either record to assess today's product.

For $100,000 of executed notional, the published platform fees would be $25 at DXAP’s 0.025%, $35 at WunderTrading’s 0.035%, and $60 at Katoshi’s entry 0.06% perp tier. Entries and exits both count toward volume. That example excludes venue fees, funding, model usage, subscriptions and any volume discounts; it is a cost comparison, not a return comparison.

Published by DXAP. Product documentation checked September 26, 2026. These are differences in the documented setup and charges, not an execution or returns test.

Which Hyperliquid automation setup fits your workflow?

Describe a strategy and review an agent’s decisions: evaluate DXAP’s hosted Hyperliquid service or Condor’s agent-building workflow. Compare deployment requirements, available tools and the authority each setup receives.

Build an inspectable systematic strategy with AI: Keel composes a strategy from defined components, then runs that logic. This differs from asking an agent to reason about each trading decision.

Configure rules, grid or DCA bots, or execute alerts: compare Coinrule and WunderTrading. These are hosted options; you define the conditions or signals the automation follows.

Build and operate the software: consider Hummingbot or Freqtrade. Condor adds an AI interface to the Hummingbot stack. Check its deployment requirements as well as the trading permissions.

This is a selected set of workflows, not an exhaustive directory. These products document Hyperliquid support. Read the account and permission requirements for the version you deploy; they determine what the software can do with your funds.

For a single execution task, Hyperliquid’s own TWAP, take-profit and stop-loss orders may be enough. Those order types do not require choosing a general-purpose AI agent.

All these options use exchange infrastructure. Hyperliquid trading fees apply, and perpetual positions may pay or receive funding. Software and builder fees are additional costs; check the venue fee schedule.

When does DXAP fit?

DXAP fits someone who wants to direct an agent in plain language and inspect its decisions. You approve the agent and its limits before it runs. Every proposed action passes through the policy engine, and the agent’s debrief records what it did, including turns when it does not trade.

Your balance stays in your Hyperliquid account. You can revoke the agent wallet on Hyperliquid. Revoking access prevents further use of that permission; it does not close existing positions.

DXAP charges $0.25 for $1,000 of executed volume. There is no DXAP charge on top of its 2.5 bps fee. The product is in alpha, and its controls do not establish that a strategy will be profitable. Start with the quickstart and product facts.

When does a framework fit?

Choose Hummingbot when you want to configure or build an execution strategy and run it yourself. Condor is worth evaluating if you want an AI interface to that same stack. Hummingbot documents different builder fees for its client and Condor: 0 bps and 1 bps respectively.

Freqtrade fits a Python workflow with backtesting and dry runs. Its Hyperliquid integration has concrete constraints: the available candle history is limited, and the bot assumes it owns the account’s trading activity. Read its exchange-specific notes before using it alongside another bot or manual trades.

For other venues and hosted tools, see the category comparison, AI trading agent shortlist, or Robinhood agentic trading alternatives.

What has actually been measured at scale?

A connector, a backtest and trading volume answer different questions. To compare returns, you would need matched periods and account-level results after fees, funding and execution costs. We have not run that comparison across these products.

DXAP’s live record reports activity. DXRG’s Terminal Pro study covered 3,505 user-funded agents over 21 days on a separate product and venue. It explains the research behind DXAP; it is not DXAP performance evidence. Past performance does not guarantee future results.

The DXRG paper reports 3,505 user-funded agents, about $20 million of volume and instruction-to-settlement traces from a 21-day real-capital deployment on Base. That is research behind the harness, not a measurement of DXAP customers trading Hyperliquid. Read DXAP’s current activity and owner-shared agentsfor the present product.

DXRG also published a separate study of DXAP’s June-to-August fleet: 231,638 completed turns and 14,596 fills, including 5,035 real-money fills. Most of the historical fills were in a paper engine with mark-price fills and no funding or slippage. The study found no directional trading edge or fleet profit in its measured window. This is an unusually detailed record of behavior and failure modes, not a return promise about the current live-only product.

Do not count marketing reach, company-funded experiments and user-funded deployments as the same thing. Katoshi’s homepage claims thousands of traders without a dated account denominator on that page. Senpi describes 43 agents with $43,000 of its own capital. HyperAgent publishes its own weekly journal; Keel publishes its own factor and backtest methods. These materials do not provide a matched, user-funded cohort study like the Terminal Pro record. They also do not establish how many customers those products actually have. Ask for a dated wallet or dataset, open positions, deposits, fills, fees and the exact population behind a volume claim.

How do I start with a hosted AI trading agent?

For DXAP, first obtain an alpha referral code, sign in with your wallet and choose the Hyperliquid account the agent will use. Complete the account setup, describe the strategy and review the generated agent card before approving it. The quickstart explains each step.

After launch, inspect the agent’s first decisions and any no-trade explanations. Check which instructions and limits are active before changing them. Read how to review a decision and how to revoke trading access. This is a live trading product, not a promise of passive income.

Questions before choosing a Hyperliquid bot or agent

See DXAP’s cost and access answers, security explanation and evidence and limitations.

How we compared them

We checked the linked product and connector documentation for venue support, strategy interface, account permissions, charges and operating constraints. Entries are grouped by workflow. We did not test competitors’ execution, score returns or verify every configuration. Nobody paid to be listed.

DXAP is in alpha. We ship changes by the day — expect things to move.