Perpetual futures never expire, so a leveraged position can stay open for weeks. That convenience explains why they now dominate crypto derivatives. It also explains why a bad trade can compound quietly through funding payments.

The growth is hard to ignore. Combined perpetual volume rose from $4.14 trillion in January 2024 to $7.24 trillion by January 2026, per perpetual futures statistics. Onchain venues climbed from a 2.0% share to roughly 10% over the same stretch.

AI agents promise to handle the hard parts: scanning markets, sizing positions, and managing risk. The real question is how much control you hand over. A tool that suggests a trade is very different from one that places it while you sleep.

This guide ranks six agents that work with perpetual markets in 2026. They range from open-source toolkits to exchange-native assistants. Each is judged on autonomy, risk controls, pricing, and who actually uses it.

What are the best AI agents for perpetual futures?

StableJack leads for research-driven traders who want to approve every order. OKX Agent Trade Kit suits builders who want an open-source bridge between an AI assistant and one exchange. Bitget GetAgent fits traders who prefer a chat assistant inside an exchange app.

Hummingbot gives technical users an open-source framework with a built-in risk gate. 3Commas offers rule-based bots with an AI helper on top. Freqtrade is the free, self-hosted option for coders who want machine learning.

Platform

Starting Price

Best For

Key Differentiator

Free Trial

StableJack

Free; paid from $16/mo

Research-led traders who approve each order

Structured AI analysis plus multi-asset execution

Yes, free plan

OKX Agent Trade Kit

Free (exchange fees apply)

Developers wiring an AI assistant to one exchange

Open-source MCP toolkit with 82 tools

Free toolkit, demo mode

Bitget GetAgent

Free tier

Exchange-native beginners

Chat-based futures assistant with approval step

Free tier

Hummingbot

Free (self-hosted)

Technical operators and quants

MCP agents behind a risk gate

Free, open source

3Commas

From $20/mo ($15 billed annually)

Rule-based bot users

DCA, Grid and Signal bots with AI setup help

Not confirmed

Freqtrade

Free (self-hosted)

Python-comfortable traders

FreqAI adaptive machine learning

Free, dry-run mode

How to judge a perps agent before you trust it

Leverage changes the question. With spot trading, a bad call costs you the position. With perpetuals, a bad call can cost you the whole margin.

Check five things before connecting any agent. Does it ask before placing orders, and can its API key withdraw funds? Does it support isolated margin, leverage caps, and full action logs?

Autonomy is the biggest dial. Rule-based automation follows instructions you wrote, while AI agents interpret goals and decide. Our guide to AI trading versus algorithmic trading explains where each approach fits.

Regulators have been blunt about the hype. The CFTC's AI trading bot advisory warns that AI cannot forecast sudden market shifts. It also calls guaranteed returns a red flag.

A joint investor alert from the SEC, FINRA, and NASAA adds that AI output can be inaccurate, outdated, or fabricated.

1. StableJack: Research-first copilot that keeps you in control

Full transparency: StableJack is our product, so we're not neutral here. We've judged it by the same checklist as every other tool on this list.

StableJack is an AI investment copilot that connects research, planning, and execution in one workflow. It is not an autonomous bot. You investigate a market, build a structured plan, and place the order yourself.

For perpetuals, trading inside the terminal runs on an onchain perpetuals venue. You can also connect your own exchange or broker API keys and trade through the same interface. Stocks, crypto, forex, and commodities are all tradable.

That breadth is the main contrast with the other agents here. Several rivals stop at crypto. StableJack treats perpetuals as one market among many.

How the agent works

Navigator is the conversational analyst. Ask about any asset and it synthesizes fundamentals, technical structure, orderbook conditions, and sentiment into one market view.

AI Insights run on their own roughly every four hours. Each brief carries a thesis, a confidence score out of 100, and a verdict from Strong Sell to Strong Buy. It also lists bullish and bearish cases plus risk flags.

The Strategy Builder Copilot turns a thesis into entry, stop-loss, take-profit, and invalidation levels. Position Management then watches open trades against the original thesis. Indicator Tracker monitors technical conditions over time.

Nothing executes automatically. You choose the asset, direction, size, leverage, order type, and risk settings for every order. StableJack never custodies funds, so assets stay in your own wallet, exchange, or broker account.

MCP integration lets you reach StableJack's market intelligence from inside the AI assistant you already use.

Pricing and fees

Plans are billed annually: Free, Basic at $16 per month, Pro at $40, and Elite at $80. Monthly credits run from 1,500 on Free to 300,000 on Elite. Every tier includes AI Insights and Copilots.

Exchange connections are limited on Basic and fully available from Pro. Enterprise is available now on a contact-for-quote basis.

Trading fees are separate. Perpetual orders start at 0.0644% maker and 0.0932% taker at the lowest volume tier. They fall to 0.02% and 0.043% at the top tiers.

The full fee schedule lists every tier. Referral-code users get a 5% discount.

Strengths and limits

Pros: Every order needs your approval, which removes the unattended-agent risk. AI is native at every tier rather than a paid add-on. Outputs are structured, with scores, scenarios, and risk flags instead of open-ended chat.

Cons: Trading fees add a second cost layer on top of the subscription. Dedicated AI Insights cover stocks and crypto only, although other assets can still be analyzed. Traders who want unattended execution won't find it here.

Who relies on it

StableJack names day traders, researchers and analysts, long-term investors, fund managers, and institutions as its core users. It suits anyone who wants institutional-style research without building a tool stack.

It is a weaker fit for people who want a bot running overnight with no input.

2. OKX Agent Trade Kit: Open-source bridge for AI assistants

OKX Agent Trade Kit is an open-source toolkit that lets AI assistants trade on one exchange. It ships as two npm packages: an MCP server and a command-line tool. Both are released under the MIT license.

It is infrastructure, not a finished agent. You bring the assistant and the strategy. The kit supplies the tools.

Compared with StableJack, there is no research layer. Compared with an exchange-native assistant, you choose the model and keep control of it.

How the agent works

The kit exposes 82 tools across seven modules. They cover spot, perpetual swaps, and futures, plus options market data, positions, and fee queries. Algo orders such as OCO and trailing stops are included, along with grid and DCA bots.

Security is the standout. API keys stay in a local config file, and requests are signed locally with HMAC-SHA256. The AI model only sends trading intent and never sees the keys.

The kit also reads your key's permissions at startup. If the key cannot trade, the order tools are never registered. A demo mode lets agents practice with simulated assets.

What it costs

The toolkit itself is free. You pay standard exchange trading fees plus whatever your AI assistant costs.

Strengths and limits

Pros: It is open source under MIT, and keys never reach the model. Permission-aware tool registration and a demo mode reduce accidents. Coverage spans perpetual swaps, futures, and spot.

Cons: It works with one exchange only. It has no built-in research or strategy logic, so output quality depends on your assistant. Setup requires technical comfort.

Who relies on it

OKX has not published adoption figures that we could find. The design points to developers and technical traders who already use an MCP-compatible assistant. They want precise control over what it can touch.

3. Bitget GetAgent: Chat assistant built into an exchange app

GetAgent is a chat-based AI assistant inside the Bitget app. It handles market analysis, strategy building, and order placement from one conversation. Bitget opened it to the public in August 2025.

It works best for traders who already keep funds on the exchange. The convenience is real. So is the lock-in.

How the agent works

GetAgent reads indicators such as RSI and volume, plus tokenomics and sentiment data. It monitors signals like RSI shifts, funding-rate changes, and volume surges. You can open, close, or adjust futures positions through prompts.

It proposes entry levels, risk controls, and order options. You approve before anything executes.

In June 2026, Bitget added GetAgent Playbook. It offers a library of ready-made strategies that run in user-authorized, isolated sub-accounts. Per coverage of the update, an Agent Harness enforces limits on execution paths, position sizing, and anomalies.

The harness also logs every action for audit.

Pricing and fees

A free tier covers analysis of the top 50 trending coins and basic futures strategy suggestions. Higher tiers add personalized strategies and natural-language order placement.

Eligible VIP users get those tiers free. Everyone else pays a monthly fee in the exchange's native token. Daily interaction limits depend on your VIP level.

Strengths and limits

Pros: A free entry point. An approval step before execution. Everything lives where you already trade, and Playbook adds isolated sub-accounts.

Cons: Your assets sit on the exchange. Feature access is tied to VIP status. It works on one venue only, and reviewers note it cannot guarantee accuracy or returns.

Who relies on it

Bitget reported more than 1 million users completing AI-powered trades earlier in 2026. Cumulative volume passed $1.2 billion. Few agents on this list publish adoption figures at that scale.

The audience skews toward exchange-native traders who want help, not a toolkit.

4. Hummingbot: Open-source framework with a risk gate

Hummingbot is an open-source framework for running trading bots, and it now speaks MCP. Its MCP server lets assistants manage trading infrastructure through natural language. It connects to centralized and onchain exchanges that offer perpetuals.

The newer Condor layer adds agents on top. Those agents work through tools served by two MCP servers.

It is the most operator-focused option here. You host it, secure it, and tune it.

How the agent works

Through the MCP server, an assistant can check balances, order history, prices, and funding rates. It can place orders, manage positions, and deploy bots. Multi-step workflows cover strategy deployment and liquidity management.

Condor agents run in three seats: tick with 37 tools, agent with 43, and full with 45. Actions that move money need your approval in chat. Unattended loops pass through a risk gate that checks session limits.

The documentation treats MCP as a privileged control plane. It advises keeping the API behind a private network, never on a public IP.

What it costs

Hummingbot is free to run. Your costs are hosting, exchange fees, and AI model usage.

Strengths and limits

Pros: The code is transparent and open source. Money-moving actions need your approval, and unattended loops pass a risk gate. Connector coverage is wide.

Cons: You must self-host. The learning curve is steep, and securing the server is entirely your job.

Who relies on it

The project appeals to quants, developers, and liquidity providers comfortable running servers. Public adoption numbers were not in the documentation we reviewed.

5. 3Commas: Rule-based bots with an AI assistant

3Commas is cloud bot software built around DCA, Grid, and Signal bots. It also offers SmartTrade, a manual terminal for entries and exits. Futures support starts on the Pro plan.

Its AI Assistant helps set up bots and navigate the platform. It is a helper, not an autonomous trader. That makes 3Commas automation first and AI second.

You write the rules, and the platform runs them. That is the opposite of an interpretive agent, and some traders prefer it.

How the agent works

DCA bots average into positions over time. Grid bots trade within a price range. Signal bots act on external alerts.

The AI Assistant drafts and helps backtest DCA configurations. One third-party review notes a cap of roughly one million tokens per month on paid plans.

Pricing and fees

Third-party reviews list Starter at $20 per month, or $15 billed annually. That tier is spot only. Pro costs $50 monthly or $40 annually and adds futures.

Expert costs $140 monthly or $110 annually. It adds read-and-write API access and full-history backtesting.

An Asset Manager plan starts around $223 per month. Check the vendor's pricing page before buying.

Strengths and limits

Pros: Broad exchange support. Several bot types plus a manual terminal. Mobile apps. Review sites score it between 4.3 and 4.8 out of 5.

Cons: It is closed-source and cloud-only, and exchange API keys sit on its servers. A 2022 API-key incident drew wide scrutiny. Backtesting is gated by plan, and the AI helper is token-capped.

Who relies on it

Its users are mostly retail traders running grid and DCA bots. Professionals who manage client accounts use the Asset Manager plan.

6. Freqtrade: Free, self-hosted, machine-learning ready

Freqtrade is a free, open-source crypto trading bot written in Python. You install it yourself with Docker or a native install. You control it through a web UI or chat commands.

Its FreqAI module self-trains adaptive machine-learning models against market data. It supports futures on several major exchanges, both centralized and onchain.

This is the do-it-yourself end of the spectrum. There is no vendor, no subscription, and no support desk.

How the agent works

Setting trading mode to futures enables long and short positions with leverage. You choose isolated or cross margin for each position. A liquidation buffer reserves part of your balance so less is exposed.

Leverage is defined in your strategy code. The documentation suggests starting at 2x to 5x and raising it gradually. It warns that cross margin can liquidate the entire account if positions are mismanaged.

Backtesting and a dry-run mode let you test before risking money. The project itself states it is for educational purposes.

What it costs

Freqtrade is free. You pay for hosting, exchange fees, and any compute that machine-learning training needs.

Strengths and limits

Pros: There is no subscription, and you control all logic and risk settings. FreqAI adds adaptive machine learning. Backtesting and dry-run mode are built in.

Cons: It requires Python skills and offers no managed support. Models need data and tuning. No approval step exists unless you build one.

Who relies on it

The audience is developers and quant hobbyists comfortable with Python. The repository page we reviewed does not publish user counts.

Common questions about AI agents for perpetual futures

Are AI agents safe to use with leveraged perpetual futures?

No agent makes leverage safe, because the risk lives in position size, margin mode, and liquidation price. Look for approval prompts, scoped API keys, and hard leverage limits. The CFTC is clear that AI cannot guarantee returns.

Should I choose an autonomous agent or one that asks first?

Start with one that asks first. Approval prompts cost you seconds and prevent runaway errors. Consider autonomy only after you've tested the logic in a demo or dry-run mode.

What leverage should an AI agent use?

Keep it low, since Freqtrade's documentation suggests starting around 2x to 5x. Higher leverage pulls your liquidation price closer to entry. Set the cap in the tool's settings, not in a prompt.

Do I need to code to use these agents?

Not for all of them, since StableJack and GetAgent work through chat and point-and-click interfaces. The OKX kit, Hummingbot, and Freqtrade expect technical setup. 3Commas sits in between.

Do these agents hold my funds?

It depends. StableJack never custodies funds, and the open-source tools use your own API keys. An exchange-native assistant like GetAgent operates on funds held at that exchange. 3Commas stores exchange API keys on its servers.

How do funding rates affect an agent's trades?

Funding payments move between long and short holders at regular intervals. When funding is positive, longs pay shorts. Over time those payments can drain margin on a position everyone forgot about.

So which agent should you pick?

If you want research, structure, and the final say on every order, pick StableJack. Its Navigator, AI Insights, and Copilots cover the work before the trade. You still press the button.

If you build your own stack, the choice depends on how much you want to maintain. The OKX kit fits one-exchange builders. Hummingbot suits operators who want approval gates and a risk layer.

Freqtrade suits coders who want machine learning and no subscription. If you prefer to stay inside an exchange app, GetAgent is the easiest start. If you want rules you control with light AI help, 3Commas fits that style.

Whichever route you take, test in demo or dry-run mode first. Size positions as if the agent will be wrong. If you want a research-first workflow with approval on every order, start on StableJack's free plan. You can explore Navigator and AI Insights before paying anything.