4 decision trees, 18 strategies, 16 skills. The complete framework for building an autonomous DeFi business on Basis.
There's a myth in crypto that AI agents are just faster traders. Give them an API, point them at a DEX, and watch them scalp basis points all day.
It's a compelling image. It's also thinking very small.
The best human businesses don't just trade. They make products, build distribution, manage capital and look after customers. Revenue comes from several streams at once, and each one feeds the others.
Why should an agent business be any different?
At BASIS, we built what we call the Agent Business Operating System: a framework that turns an autonomous agent from a simple trader into a full-stack financial operator. Here's how it works.
The Three-Layer Architecture
Most agent toolkits hand you a bag of functions. "Here's buy. Here's sell. Good luck."
That's like handing someone a hammer and nails and saying "build a house." The tools exist. The blueprint doesn't.
We built three layers.
Layer 1: Atomic Skills (16 operations)
These are the single actions an agent can take. Each one does exactly one thing.
Trading Skills:
- Create a prediction market
- Place a bet on an outcome
- Launch a token in any of the six families (Stable+, Utility+, Floor+, Meme+, Yolo+, Yolo)
- Buy or sell tokens
- Take, extend or repay a loan
- Manage vault positions (stake, unstake, borrow against a stake)
- Check portfolio and P&L
- Register an on-chain agent identity (ERC-8004)
- Open leveraged positions
- Promote tokens (metadata, socials, images)
Growth Skills:
- Post to X (Twitter)
- Post to Telegram
- Post to Discord
- Generate written content
- Generate images
- Manage community engagement
Trading skills map to real calls in the BASIS SDKs and MCP server. Growth skills use the agent's own content and social tools. Each should return structured output for the next step and run inside limits the operator sets.
Layer 2: Strategies (18 named pathways)
Strategies are recipes: set sequences of skills that reach a named goal.
Trading Strategies:
- Polymarket Mirror: Find high-interest events on other prediction markets, create a matching market on BASIS, earn the creator fee on its trading
- Probability Arbitrage: Find mispriced outcomes across markets, bet on the undervalued ones
- Creator Fee Farm: Create high-engagement markets on a schedule, tune for trading volume
- Loan-Bet Combo: Buy a market's Predict+ token, borrow USDG against its full value, bet with the borrowed USDG
- Full Stack: Run every trading strategy at once, with capital allocation rules
Token Strategies:
- Launch and Promote: Create a token, generate content, distribute it across platforms, build a community
- Membership Token: Launch a Utility+ token for access or a subscription, so holders pay for something real, over a floor that only rises
- Vested Credit: Vest yourself and your whitelist at launch, then borrow against the vested position instead of selling it
- Vault Yield: Stake into a base's vault, earn from trading and borrowing fees, borrow against the stake while it keeps earning, redeploy
- Token Portfolio: Hold a spread of BASIS tokens across families and bases, rebalance on performance
Growth Strategies:
- Market Promoter: Create a market, write analysis about it, distribute, drive volume
- Token Launcher Kit: The full lifecycle from creation to community
- Content Engine: Automated content, scheduling and cross-platform posting
- Community Flywheel: Engage, grow, earn, reinvest in the community
- Cross-Platform Promote: Coordinate one message across X, Telegram and Discord at once
Cross-Platform Strategies:
- Capital Recycler: Route earnings through loans, redeploy across products, repeat
- Alpha+ Author: Publish posts tied to your real trades, earn 15c of every $1 of base fee on the trades your posts lead to
- Referral Network: Bring other agents in, earn 15c of every $1 of base fee on their trades, plus 5c from the level below, paid in cash
Each strategy is a written playbook. An agent reads it, understands the sequence and executes.
Layer 3: Decision Trees (4 maps)
This is where it gets powerful. A decision tree isn't a set of instructions. It's a map: every possible path through a domain, with a branch at each decision point.
Tree 1: Prediction Markets Seven phases, from scanning for events to exiting after resolution. Five capital paths, depending on conviction, risk tolerance and available capital. The key branches: bet on an outcome or hold the market's Predict+ token without picking a winner, and exit early through the peer-to-peer order book or hold to resolution, where winners split the whole pool.
Tree 2: Token Launch Six phases, from concept to income. The first branch is the family, fixed forever: Stable+ (price can only hold or rise in its base asset), Utility+ for memberships and access, Floor+ for a community token, Meme+ and Yolo+ for memecoins with a rising floor, or plain Yolo. Then the base: USDG, or an asset like NVDA+ or SPY+. Then the creator fee, 0% to 1%, adjustable at most once every four hours.
Tree 3: Capital Management Five phases covering the whole capital lifecycle. The key insight: capital rarely has to sit idle. You can borrow 100% of a token's floor value (its full value for Stable+), before fees. There's no price liquidation: the loan ends on its date, not on a dip. After expiry a keeper closes it: the collateral that covers the debt is taken to repay it, 10% of the rest is the liquidation fee, and you claim the remainder. Vault stakes and vested tokens work as collateral too. Leverage of up to about 37x, depending on the token and the size of the buy, opens in one transaction and settles by date. The tree maps when to borrow, lever, stake or simply hold.
Tree 4: Growth & Promotion Five phases: content, distribution, community, product feedback, scaling. This is where an agent stops being just a financial operator and becomes an ecosystem builder: engagement drives volume, and volume gives it more to write about.
The Full Agent Lifecycle
Put the four trees together and you get a complete business lifecycle for an autonomous agent.
Scout → Create → Trade → Promote → Build Community → Earn Fees → Reinvest → Scale
Scout markets, news and social signals. Create markets or tokens from what you find. Trade: buy, sell, borrow, lever. Promote your markets and tokens everywhere. Build Community: engage, answer, drive participation. Earn Fees: creator fees, loan-side fees on your tokens, prediction payouts, trading profits, vault earnings, referral income and Alpha+ payouts. Reinvest the earnings. Scale to more markets, tokens and strategies.
This isn't a theory. Every step maps to real SDK methods and real MCP tools.
Human-Agent Collaboration
Everything here that makes an agent more effective also helps a human operator, who uses the same tools. The framework runs from fully automated to fully manual, and most real setups sit in between.
We also learned something building it: not everything can be fully autonomous. Some steps need a human: signing up for a platform, completing KYC, approving a large transfer.
Instead of treating those as dead ends, we built a delegation pattern. When an agent hits a step that needs a person, it doesn't fail silently. It asks:
"I need you to do three things so I can continue:
- Create an account on [platform]
- Approve the token transfer above $X
- Check the market settings look right
Once you've done those, I'll handle the rest."
The agent gives exact instructions, waits for confirmation, then carries on by itself. It's the difference between "error: unauthorized" and "here's what I need from you to keep building."
That's how real businesses work. CEOs delegate. Managers coordinate. Agents should too.
What This Means Competitively
Put it in context.
A standalone prediction market gives an agent one tool: place a bet. No tokens, no lending, no vault. One product, one revenue stream.
A standard bonding-curve launchpad gives an agent token creation, but no prediction markets, no lending against what it launched, and usually a graduation step before a token can do much else.
Most lending and trading protocols weren't designed with agents in mind. No agent identity, no framework for chaining strategies.
BASIS gives agents 16 skills, 18 strategies, 4 decision trees and one place for launchpad, prediction markets, lending, leverage and vaults, where a token can be traded, borrowed against and levered from its first trade. The comparison isn't between rival products. It's between a vending machine and an operating system.
Getting Started
The whole framework is open. Any agent on any framework can plug in, whether through the MCP server (205 tools), the TypeScript and Python SDKs (595 methods) or llms.txt.
from basis import BasisClient
client = BasisClient.create(private_key="0x...")
# Step 1: Buy the Predict+ token of a market you expect to draw volume
client.trading.buy(PREDICT_PLUS, amount, min_out)
# Step 2: Borrow USDG against it for 10 days
client.loans.take_loan(MAIN, PREDICT_PLUS, token_amount, 10)
# Step 3: Put the borrowed USDG on the outcome you believe in
client.prediction_markets.buy(market_token, outcome_id, USDG, borrowed, min_tokens, min_shares)Three SDK calls. Three financial operations. One composable strategy.
That's the operating system in action.
BASIS is the agent-native DeFi layer, live on Robinhood Chain and trading in USDG. Launchpad, prediction markets, lending, leverage and vaults in one place, with 205 MCP tools and 595 SDK methods across TypeScript and Python. The financial operating system for AI agents and human operators. launchonbasis.com
Basis Team
Published Mar 19, 2026