Short answer: "AI trading agent" covers four different products. Curated agents, such as Deploy Finance's, run a defined strategy from your own wallet; vaults pool deposits under a manager or protocol; toolkits automate rules you write; chat assistants suggest trades but never place them. Choose by who holds your capital and who decides each trade, not by the label.
Disclosure: Deploy Finance publishes this comparison and offers competing products, the Income: Funding Rates and Superstar agents, which are included in it.
Not financial advice. Perpetual futures use leverage and can be liquidated, delta-neutral does not mean risk-free, and funding can turn negative. Read the risk disclosure.
"AI trading agent" is used for at least four different products, and the differences matter more than the label. Some are curated strategies that run for you. Some are pooled vaults where a human manager trades everyone's capital together. Some are toolkits that automate rules you write yourself. And some are chat assistants that suggest trades but never place them.
What separates them is who holds your capital while the strategy runs, and who decides what it does.
Compare the four models
| Curated autonomous agents | Manager-led vaults | DIY bot toolkits | LLM assistants | |
|---|---|---|---|---|
| Examples | Deploy Finance | Hyperliquid HLP and user vaults, Drift vaults | 3Commas, self-hosted bots | ChatGPT, Claude with a read-only integration |
| Who sets the strategy | The platform's quant team, published as a defined mandate | The user-vault leader or delegate; HLP uses protocol strategies | You, rule by rule | Nobody; it suggests, you execute |
| Where your funds sit | Your own wallet, individually held | Pooled into a shared vault address | Your exchange account or connected wallet | Wherever they already are |
| Your result depends on | Your own allocation and the agent's mandate | The vault's aggregate book, shared with every other depositor | The quality of the rules you wrote | Your own execution |
| Charges | No subscription or management fee; gas, exchange fees, and Superstar builder fee | Hyperliquid user vaults: 10% leader profit share; HLP: no leader fee; Drift: vault-specific | Subscription pricing | None |
| Exit | Revoke trading access, use the agent exit flow, then withdraw | Follow the vault's withdrawal rules and lock or redemption period | Stop execution, review open positions, then withdraw if needed | No execution to exit in a read-only setup |
| Work required from you | Choose an agent and an allocation | Evaluate a manager's track record | Design, backtest, and maintain the strategy | Everything |
Curated autonomous agents
A curated agent is a strategy someone else designed, tested, and published with a fixed mandate, and then runs on your capital without pooling it with anyone else's.
Deploy Finance runs two live agents, funded and settled in USDC:
- Income: Funding Rates is market-neutral. It holds long spot against a short perpetual on the same asset so price movement cancels between the legs, and collects the funding leveraged longs pay to shorts.
- Superstar is directional. It trades Hyperliquid spot, perpetuals, and HIP-3 markets, with selective outcome-market positions. Position sizing follows a fixed loss budget.
Your USDC stays in a wallet you control. The agent receives a scoped session key that lets it open, close, and manage positions but never transfer or withdraw funds, and you can revoke it at any time. There is no subscription or management fee on deposits. Gas and exchange fees apply; Superstar also charges a dynamic builder fee of 1 to 10 basis points per trade. See fees.
The trade-off: you take the mandate as published. You do not tune the entry rules, and you do not pick the venue.
How to read an agent's performance record
Use a record with explicit dates and accounting. This fictional scorecard illustrates the questions to ask; it describes no provider or actual strategy.
| Field | Illustrative disclosure | Question it answers |
|---|---|---|
| Evidence type and dates | Live account, 1 January–31 March, one specified year | Is this executed trading or a simulation, and which regime does it cover? |
| Gross and net return | +4% gross; +2.5% after identified costs | Which fees, funding, slippage and withdrawals enter the calculation? |
| Maximum drawdown | 6% from an earlier equity peak | How much did account value fall before recovering? |
| Worst period | Worst calendar week: −3% | What loss could a holder actually experience between updates? |
| Exposure | Directional perpetuals; daily leverage disclosed | Was the return earned with the risk you intend to take? |
| Benchmark | Unlevered underlying return for the same dates | Did active trading improve the outcome for comparable exposure? |
For a funding strategy, also request funding receipts and total capital used. For a directional strategy, inspect open risk and unrealized PnL alongside closed trades. A short sample, an omitted cost or a different benchmark can change the interpretation without changing the displayed return.
Manager-led vaults
Vaults pool depositor capital under a manager or protocol strategy. That is a different ownership and exit model from an individually held agent allocation.
Hyperliquid vaults include HLP, which follows protocol strategies without a leader profit share, and user vaults, whose leaders receive 10% of profits. HLP has a four-day deposit lock from the latest deposit; user vaults have a one-day lock. See Hyperliquid's vault documentation.
Drift vaults pool funds on Solana under a delegate that can trade but cannot directly withdraw depositor capital. Managers configure fees and redemption periods; read the particular vault's terms rather than assuming a standard fee or one-to-seven-day exit.
In both cases your outcome depends on the vault's aggregate book, not a position sized for you. You get the manager's skill, and depositor flows can affect position management and liquidity.
DIY bot toolkits
3Commas is the reference point here: non-custodial software that connects to exchange accounts through API keys, plus a Web3 connection to Hyperliquid. It offers DCA, Grid, and Signal Bots, SmartTrade for manual execution, and an AI Assistant that can draft and backtest a DCA configuration from a description.
A toolkit gives you the most control and asks for the most work. You are the strategist. The automation only executes rules you are responsible for designing, testing, and maintaining as market conditions change.
LLM assistants, and where they stop
A general chat assistant can explain funding rates, read a chart you paste in, or reason about a position. On its own it cannot hold a key, sign a transaction, or place an order.
What has changed is the bridge. Deploy Finance exposes a Model Context Protocol server at https://mcp.deploy.finance/mcp with tools grouped into learn, explore, fund, earn, and exit. Read-only tools (asking what Deploy is, browsing live agents, simulating yield on an amount) need no sign-in. Tools that require account access use OAuth authorization and operate within approved scopes. A connected assistant's execution authority depends on the tools and permissions you grant. The limits are covered on Can ChatGPT or Claude manage your crypto?.
How to choose
Choose a curated agent when you want a defined strategy executed on capital that stays in your own wallet, and you would rather evaluate a published mandate than a manager's discretion.
Choose a vault when you have evaluated a specific manager or protocol strategy, and you are comfortable pooling with other depositors and accepting the lock-up.
Choose a toolkit when you have a strategy of your own and want automation to execute it faithfully.
Choose an assistant when you want help thinking, and you intend to keep execution in your own hands.
Automation is not a risk reduction. Products that execute perpetual-futures strategies carry liquidation, funding-reversal, venue, and smart-contract risk. Market-neutral is not riskless, and a backtest is not a track record. Read Risks before allocating.
Three reader profiles: which model actually fits?
You want funding income. Start by evaluating a market-neutral mandate such as Income: Funding Rates. You choose the allocation and accept funding, hedge and venue risks; the agent handles the published execution process. Ask for net funding captured over a stated period and the capital committed, including margin and idle balances.
You want directional trading. Evaluate a directional agent's loss budget, exposure and exits. Superstar's documented minimum is $10,000 USDC of working capital, separate from service charges. You retain the allocation and withdrawal decisions while delegating trade management. Confirm the current minimum and mandate before funding.
You already have rules. A DIY toolkit fits when you can specify entries, sizing and exits and maintain them as conditions change. Request venue support and a cost-aware test of your exact configuration. Automation delegates order handling; strategy design and ongoing monitoring remain your work.
What happens after a bad week?
Suppose losses make you want to stop. First decide whether you are reviewing the allocation or ending it. A control that pauses new orders, where available, should not be assumed to close existing positions.
For Deploy, you review the open positions and revoke delegated trading access. Revocation prevents further use of that permission; it does not itself settle the book or move capital. You then use the supported exit flow, wait for the balance to return to the Funding Wallet, and choose whether to withdraw or reallocate. Position closing and settlement depend on execution and liquidity.
Follow the full stop-and-exit sequence. Decide how to handle the open book before relying on an inactive agent to manage it.
Learn more
- What are Deploy Finance agents?
- How to choose an agent
- Deploy Finance vs Hyperliquid vaults
- Deploy Finance vs Drift vaults
- Deploy Finance vs 3Commas
Start with Deploy Finance
Create a self-custodial Deploy Finance wallet and review the live agents.
