ADR-0001: MVP exchange is Hyperliquid

• Status: accepted, refined by ADR-0015
• Date: 2026-05-27
• Refined by: ADR-0015 — testnet deprecated; mainnet is the only live target. The "testnet is the default" clause below no longer applies; the rest of this ADR is in force.

Context

The platform must integrate with at least one exchange to be useful. Picking N exchanges day one quadruples integration work, doubles testing surface, and slows down everything else. We need exactly one for MVP.

Requirements that narrow the choice:

• Perpetuals — most "AI trading" interest is in perps (24/7, leverage, liquid, programmatic)
• API-first — clean REST + WS, no scraping
• Testnet available — we can't develop against real money
• Programmatic withdrawals/deposits not required (we don't custody)
• Liquid on majors (BTC, ETH, SOL minimum) for backtests to be meaningful

Decision

Hyperliquid is the MVP exchange. All Skills in MVP trade Hyperliquid perpetuals. Mainnet integration is gated behind explicit per-deployment confirmation. (Testnet was originally the default; deprecated in ADR-0015 — mainnet is now the only live target.)

We'll integrate via @nktkas/hyperliquid (maintained TypeScript SDK). Adapter sits in packages/brokers/hyperliquid-mainnet.

Alternatives considered

Alt A — Binance Futures

• Most liquid perp market in the world
• Geographic + KYC restrictions add user-onboarding friction
• API is mature but heavy; many endpoints, regional variants
• Not picked: friction for users, larger integration scope

Alt B — Bybit / OKX

• Similar profile to Binance, slightly smaller
• Not picked: same reasons; nothing here that beats Hyperliquid for our use case

Alt C — dYdX

• Decentralized perp DEX
• Smaller liquidity than Hyperliquid in current era
• Migration to v4 chain complicates SDK story
• Not picked: smaller market + integration churn

Alt D — Multi-exchange from day one

• Forces clean abstraction early
• Quadruples integration + ops surface
• Not picked: premature; the abstraction can be designed without being implemented

Consequences

Positive

• Single exchange means single SDK, single rate-limit budget, single testnet, single fee schedule to reason about
• Hyperliquid's info API exposes leaderboards, OI, funding cleanly — valuable as tools for agents
• Crypto-native users (our early target) already know Hyperliquid
• Testnet is good and free

Negative / trade-offs

• We're locked into one exchange's quirks (order types, margin model, downtime)
• A Hyperliquid incident takes the entire platform's live trading down
• Users wanting non-Hyperliquid assets must wait for Phase X
• Smaller TAM than a multi-exchange platform

Things we'll need to revisit

• After MVP, plan a second exchange. The BrokerAdapter interface is designed to accommodate this — but it isn't proven until we add a second adapter and discover what doesn't fit.
• If Hyperliquid changes ownership / has a major incident / shuts down, plan B is "add Binance or Bybit fast." Keep the adapter interface clean.

References

• docs/architecture/execution-engine.md — broker adapter interface
• packages/brokers/ — adapter implementations (when scaffolded)
• Hyperliquid docs: https://hyperliquid.gitbook.io/hyperliquid-docs