How to Start Multi-Market Portfolio Management with AI Agents

Multi-market portfolio management with an AI agent means you authorize one program to hold and adjust positions across stocks, crypto, perpetual futures, options, and prediction markets at the same time, under limits you define in advance. To get started you do not need a trading strategy, a quant background, or code: you need funds you control, a scoped API key, a written mandate, and a budget you can afford to lose. The sensible sequence is to practice in paper trading, watch the agent behave, and only then authorize live trading with a small amount. This article walks through that sequence.

What does multi-market portfolio management actually mean?

Portfolio management is the ongoing work of deciding what to own, how much of it to own, and when to change your mind. Doing that across one market by hand is already a chore. Doing it across five market types, each with its own app, its own contract sizes, and its own hours, is where most people give up. Multi-market management with an agent moves the mechanical parts to software. You state the intent, for example keep roughly half the portfolio in large, liquid stocks, a quarter in crypto spot, and the rest reserved for hedges, and the agent handles the order placement, sizing, and monitoring through one connection.

Day to day, the work looks like this. The agent reads prices and positions across all five market types, compares them against your instructions, and proposes or places orders when something drifts out of line. If you told it to keep crypto below a third of the portfolio and a rally pushes it past that, the agent can trim back toward your target. If you told it to never hold a losing option position into expiry week, it can exit on schedule. None of this requires you to be awake, but all of it requires the instructions to exist.

If the phrase agentic trading is new to you, the short version is that an AI model connected to tools can read market data, reason about your instructions, and place orders on your behalf, rather than just answering questions about markets. What is agentic trading? covers the concept in detail. The important point for a beginner is the division of labor. The agent is good at watching many instruments at once, doing arithmetic quickly, and executing your instructions exactly. You are responsible for the instructions, the limits, and the final decision to let it run with real money.

Why spread a portfolio across several markets?

Different markets let you express different kinds of views. A stock position is a long-term claim on a company. A perpetual future lets you take short exposure or use leverage without an expiry date. An option can give you a bounded, asymmetric payoff where you know the maximum loss up front. A prediction market contract lets you take a position on a specific event. Crypto spot markets trade around the clock, which matters when you want to react to news on a weekend. A portfolio that only ever uses one of these tools is leaving the others on the table.

Suppose, as a purely hypothetical example, that you hold a portfolio of technology stocks and you are worried about a specific earnings season. With access to several markets, an agent could buy a small put option as a hedge, take a modest short position in a perp, or buy a prediction market contract tied to the outcome, all sized in dollars against the same budget. Whether any of those is a good idea depends on your situation, and each can lose money. The point is that one agent can hold the whole picture instead of you juggling four separate apps.

There is also a diversification argument, but it deserves honesty. Holding positions in several markets can reduce the chance that one bad event wipes out the whole portfolio. It can also do the opposite, because markets that look unrelated often fall together during a panic. More markets means more ways to be wrong, more fees, and more surface area for mistakes. A multi-market agent is a tool for expressing a plan, not a plan by itself. If you do not have a view, the agent will still happily trade, and trading without a view is a reliable way to lose money.

What do you need before your first automated trade?

The shopping list is short. First, funds in a wallet or account you control. Felix is non-custodial by construction, which means your money stays in a wallet you own. The agent can spend within the limits you set, but it can never withdraw to an address you have not approved, so it cannot pay itself. Second, an agent. That can be an AI assistant running in an MCP client such as Claude or Cursor, or a small program calling the REST API directly. Third, a scoped API key, which is the permission slip that ties everything together.

Before you generate a key, write down, in plain sentences:

• ·What the agent is allowed to do, and what it must never do.
• ·The total budget it may commit, in dollars.
• ·The largest single position it may open.
• ·The conditions under which it must stop and ask you.
• ·How often you will review its work.

This written mandate sounds bureaucratic, but it is the single most useful thing a beginner can produce. An agent with a vague mandate will fill in the gaps with its own judgment, and its judgment is not your judgment. Running an AI trading agent with real money, safely goes deeper on turning a mandate into concrete limits.

How do you go from zero to a running agent?

Here is the sequence to follow if you are starting from nothing. It is deliberately conservative. Every step is reversible until you authorize live trading, and even that step is bounded by the budget you choose.

1. 01Read the quickstart in the docs and generate a scoped key. Give the key a name that says what it is for, because you will thank yourself later when you have several.
2. 02Connect your agent. If you use an MCP client, add the Felix MCP tools and confirm the agent can call them. If you prefer code, point your script at the REST API with the key stored in an environment variable, never pasted into source.
3. 03Stay in paper trading. Give the agent one simple instruction, such as buying a small dollar amount of a large, liquid asset, and watch it work. Check that the fills, sizes, and reports match what you expected.
4. 04Add markets one at a time. Get comfortable with a spot order before you let the agent touch a perp, an option, or a prediction market contract. Each instrument has its own failure modes, and you want to meet them with fake money first.
5. 05Rehearse the exits. Ask the agent to close everything, and try the panic switch yourself, so you know exactly what happens and how long it takes.
6. 06Authorize live trading with a small budget. Live keys require explicit owner authorization, which is deliberate. Start with an amount whose total loss would not change your month.
7. 07Put a recurring review on your calendar. Weekly is a reasonable default.

Notice what is missing from that list: picking a strategy. For a first setup, the goal is not performance. The goal is to build trust in the plumbing, meaning the key, the limits, the reporting, and your own ability to stop the agent. Strategy comes after you believe the plumbing works.

A first order through the REST API looks like this: The exact request schema is in the docs; this example shows the shape.

curl -X POST https://api.felix.trade/v1/orders \
  -H "Authorization: Bearer YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "market": "crypto",
    "symbol": "BTC-USD",
    "side": "buy",
    "amount_usd": 50
  }'

One detail worth noticing: the order is sized in dollars, not in contracts or base units. The API normalizes each venue's contract math, so the same instruction shape works whether the agent is buying crypto spot, an option, or a prediction market share. For a beginner this removes a whole category of sizing bugs.

How do you keep a multi-market agent under control?

Control comes from layers, not from trusting the model. Felix keys are scoped, so a key can be restricted to specific markets or actions. Budget caps bound the total the agent may commit. Position limits bound any single trade. Exit plans define what happens when a position moves against you. And a panic switch flattens everything and revokes the key's access in one step. Because the system is non-custodial, even a fully compromised agent cannot move your funds to itself. The worst case is bounded by the budget you granted, which is why that number deserves real thought. Non-custodial trading for AI agents explains the custody model in full.

Exit plans deserve special mention in a multi-market setup. A single global instruction, such as telling the agent to close everything and stop if the total portfolio drops by a fixed percentage from its peak, gives the agent a clear circuit breaker that works across all positions at once. Per-position exit rules handle the smaller cases. Test both in paper trading. The first time you watch the agent flatten the whole book should be boring, and it should happen before real money is involved.

The other half of control is attention. Automation changes your job from placing trades to supervising a trader. Check positions on a schedule, read the agent's reasoning when it reports, and treat any instruction you do not remember giving as an incident. If you find yourself avoiding the dashboard because you are afraid of what you will see, that is a signal to reduce size, not a signal to check less often.

What mistakes do first-time automators make?

• ·Skipping paper trading because it feels slow. The cost of paper trading is time. The cost of skipping it is discovering a misunderstanding with real money.
• ·Giving the agent a vague mandate like grow the portfolio. The agent will interpret this as permission to take risk, and it cannot know how much risk you actually meant.
• ·Setting limits wide to give the agent room. Wide limits do not make the agent smarter. They only make your worst case worse.
• ·Confusing activity with progress. An agent that trades constantly is generating fees and decisions, not necessarily results. Judge it over weeks, not by the number of orders.
• ·Assuming diversification is automatic. Positions in different markets can be the same bet in disguise. Long crypto spot and long a crypto perp are one exposure, not two.
• ·Forgetting that automation is not abandonment. The agent does not get tired, but it also does not know when the world has changed in a way your instructions did not anticipate.

None of these mistakes require sophistication to avoid. They require patience and a written plan. Start in paper trading, keep the first live budget small enough that losing it is a tuition payment rather than a disaster, and widen the agent's permissions only after it has earned your trust in a narrow role. Multi-market portfolio management is a skill you grow into, and the agent is there to do the busywork while you learn.