Skip to main content
DeFi has a usability problem that compounds its information problem. Even when users understand what they want to do — protect a position, capture a yield opportunity, respond to a market move — translating that intent into the correct sequence of on-chain actions requires significant technical knowledge. The Natural Language Interface eliminates this translation burden. You describe what you want in plain English. The NLI handles the rest.

​
What Is the NLI

The Natural Language Interface is a translation layer between human intent and on-chain execution. It sits at the boundary between the user and the Sentinel Core, accepting unstructured natural language input and converting it into structured strategy objects that the execution architecture can act on. The NLI is not a chatbot. It is not a general-purpose AI assistant. It is purpose-built for DeFi strategy expression — trained to understand the vocabulary, logic, and constraints of on-chain financial operations on Solana. It understands assets, amounts, conditions, timing, risk limits, and execution venues. It understands what is technically possible on-chain and what is not, and it generates strategies that are executable rather than aspirational. Critically, the NLI never executes autonomously. Every strategy it generates is presented to the user as a structured, human-readable plan before any on-chain action occurs. The user reviews, modifies if needed, and explicitly approves. The NLI is an intelligence layer, not an autonomous actor.

​
How It Works

1

You enter a natural language command

Type your intent in plain English through the SURCHI interface. You can be precise (‘Buy exactly 500 USDC of SOL at market’) or express conditional logic (‘Alert me if SOL drops 8% in the next 4 hours, then let me decide’). You can reference relative values (‘Sell half my position’), time conditions (‘Only between 9 AM and 5 PM UTC’), and risk constraints (‘Never spend more than 200 USDC on any single trade’).
2

NLI parses intent and extracts parameters

The NLI processes your command through its intent recognition and parameter extraction models. It identifies: the action type (buy, sell, alert, provide liquidity, withdraw, monitor), the target assets, the amounts (absolute or relative to current holdings), the trigger conditions (price levels, time windows, percentage changes), and any risk constraints you’ve specified. Ambiguous inputs are flagged with clarifying questions rather than guessed at.
3

NLI generates a structured strategy object

Extracted parameters are compiled into a structured strategy object — a precise, machine-readable representation of your intent. The NLI presents this object to you in human-readable form: a plain English summary of exactly what the system will do, under what conditions, with what limits. You can see the full parameter set, verify it matches your intent, and modify individual parameters directly.
4

You review and approve the strategy

You review the generated strategy. The interface displays the complete execution plan: the trigger conditions, the execution steps, the risk parameters, the maximum exposure, and the expected on-chain actions. You can approve as-is, edit specific parameters and regenerate, or discard the strategy entirely. No execution occurs at this stage — this is a confirmation checkpoint.
5

Execution Sentinel carries out the on-chain steps

Once approved, the strategy is handed to the Sentinel Core, which routes it to the Execution Sentinel. The Execution Sentinel monitors trigger conditions in real time and executes the approved on-chain steps when conditions are met. All execution events are recorded on-chain and surfaced in your activity feed via the NLI.

​
Example Commands

The NLI is designed to handle a wide range of DeFi intent expressions, from simple market orders to multi-condition strategies: 
# Conditional buy order
"Buy 100 USDC of SOL when price drops 5% from current level"

# Watchlist alert
"Alert me if any wallet on my watchlist moves more than 10,000 USDC"

# Conditional liquidity provision
"Provide liquidity to the SOL/USDC pool if APY exceeds 25%"

# Conditional take-profit
"Sell 50% of my BONK position if it gains 30% in 24 hours"

# Portfolio protection
"Set a stop-loss on my JTO position at 15% below my entry price"

# Multi-condition strategy
"If SOL drops below $140 AND whale wallets are net buying, buy 200 USDC of SOL"

# Liquidity withdrawal trigger
"Remove my liquidity from the USDC/USDT pool if the APY drops below 5%"

# Time-bounded monitoring
"Monitor my open positions and alert me to any changes over 2% while I'm offline"
The NLI handles relative references (‘my position’, ‘current level’, ‘my entry price’) by querying your connected wallet’s on-chain state at strategy creation time. Amounts expressed as percentages of holdings are resolved to absolute values at execution time, using the actual position size at the moment the trigger condition is met.

​
Intent Parsing and Disambiguation

The NLI is designed to ask rather than assume when intent is ambiguous. Examples of disambiguation behavior:
  • Ambiguous asset reference: ‘Buy some SOL’ → NLI asks for amount or amount range before generating strategy
  • Missing risk parameters: A strategy with no stop-loss or maximum exposure → NLI highlights the absence and offers to add default risk controls
  • Conflicting conditions: Trigger conditions that cannot logically co-occur → NLI flags the conflict and asks for clarification
  • Technically infeasible operations: Requesting execution on a venue with insufficient liquidity for the specified size → NLI explains the constraint and offers alternatives
Disambiguation interactions are conversational — the NLI explains why it is asking and what information it needs. It does not generate a partially-formed strategy and leave the user to discover the gap at execution time.
You always review and approve the generated strategy before execution. The NLI never executes without your confirmation.