Bookie’s NLP engine parses natural language input and extracts transaction parameters without requiring complex UI interactions.
Supported Commands
Token Swaps
Exchange 100 USDC for BONK with 2% slippage
Token Transfers
Send 5 USDC to 7xKXtg2CW87d97TXJSDpbD5jBkheTqA83TZRuJosgAsU
Transfer 0.1 SOL to alice.sol
Portfolio Queries
What's my portfolio worth?
Display transaction history
Price Checks
The intent parser identifies key components from your command:
| Entity | Example | Description |
|---|
| Action | swap, send, show | Operation type |
| Input Token | SOL, USDC | Source token |
| Output Token | BONK, JUP | Destination token |
| Amount | 10, 0.5 | Quantity to transact |
| Slippage | 1%, 2% | Maximum acceptable slippage |
| Address | 7xKXtg… | Recipient wallet |
Processing Flow
Tokenization
Input text is split into tokens and normalized
Entity Recognition
NLP model identifies tokens, amounts, and addresses
Intent Classification
Command categorized as swap, transfer, or query
Parameter Validation
Extracted values checked for validity and completeness
Transaction Construction
Valid parameters passed to routing or execution layer
Slippage Configuration
Default slippage is 0.5% (50 basis points). Override with explicit commands:
Swap 100 USDC for SOL with 1% slippage
Trade 50 USDC to BONK, max slippage 2%
Higher slippage increases execution probability but may result in worse prices. Use minimum slippage necessary for your trade size.
- Full Solana Address:
7xKXtg2CW87d97TXJSDpbD5jBkheTqA83TZRuJosgAsU
- SNS Domains:
alice.sol, bob.sol
- Shortened Format:
7xKX...gAsU (must be unambiguous)
Error Handling
If the parser cannot extract required parameters, Bookie will prompt for clarification:
User: “Swap SOL”
Bookie: “How much SOL would you like to swap, and for which token?”
User: “10 SOL for USDC”
Bookie: Proceeds with transaction
| Metric | Value |
|---|
| Parse Time (p50) | 42ms |
| Parse Time (p99) | 78ms |
| Entity Accuracy | 97.3% |
| Intent Accuracy | 99.1% |
The NLP model is continuously improved based on user interactions. Ambiguous commands help train better entity recognition.
