Last updated
Haiku Intents enable any user to achieve their trading and portfolio management outcomes efficiently. Traditional DeFi workflows require traders to manually execute swaps, lending, borrowing, bridging, and liquidity provision across multiple protocols. This not only takes time but also increases gas costs, forces manual calculations, and leaves users with dust.
Haiku Intents remove this complexity by empowering users to define their trading objectives, while handling execution. Haiku uses an instruction-tuned large language model (LLM), a constrained optimisation engine, and an atomic transaction bundler to process user-defined intents into single-click transactions. This approach enables a Web-2 trading experience in Web3, enabling users to execute financial strategies without deep technical knowledge.
Haiku offers three modes to trade with Haiku Intents, giving users flexibility in how they interact with the system. Every mode provides access to Haiku's proprietary convex optimisation engine, which determines the most efficient transaction path - saving traders time and reducing trading costs.
Haiku Agent empowers users to express their trading goals in plain language and ensures efficient trade execution.
Users enter their trading strategies as natural language inputs, which are processed by Haiku's 9 Billion parameter instruction-tuned large language model (LLM). These inputs are then converted into structured intents before passing them to the convex optimiser, which then determines the best execution path. Haiku then returns an optimised transaction bundle for the user to sign.
Example: A user enters "Derisk my portfolio into USDC".
Haiku Agent processes the intent and performs the following actions:
Parses the intent using the LLM.
Identifies the user's eligible holdings and calculates the optimal assets to trade.
Determines the best trade or exit route using Haiku's convex optimiser.
Returns a transaction bundle for the user to sign and execute.
Haiku Classic combines the familiarity of manual trade entry with optimised execution, ensuring traders achieve the best outcomes while maintaining full control.
Haiku Classic is a manual trading UI for users who want full control over their trades. Users manually input their transactions, selecting tokens, lending collateral, debt positions, cross-chain tokens and liquidity provision (coming Q2'25). Haiku's convex optimiser then structures the trade for the best execution.
Example: A trader manually rebalances their portfolio in the Haiku Classic UI.
The user selects inputs and the amounts of each (e.g. 500 $USDC on Base, 1 $WBTC on Berachain).
The user selects their desired outputs (e.g. $USDC collateral on AAVE Arbitrum, $ETH debt on AAVE Arbitrum, $LINK token on Arbitrum) and the percentage that each of these positions will represent in their desired portfolio construction.
Haiku's convex optimiser calculates the most efficient transaction sequence.
Haiku Classic returns an optimised transaction for the user to sign.
Haiku API enables high-performance, intent-driven execution for AI Agents, protocols and institutions.
Haiku API provides direct programmatic access to the Haiku convex optimiser. Partners can submit structured intent requests - in either natural language or trade instruction format - and receive optimised trade executions in response.
