Agent Memory
The PyVax transpiler natively handles persistent onchain memory storage by interpreting standard Python dictionaries and lists as persistent EVM state trees mapped directly into Avalanche blocks.
State Abstraction
By isolating your autonomous actors' logic away from the blockchain and strictly using PyVax contracts as a verifiable database, agents can securely memorize parameters.
python
from pyvax import Contract, agent_action
class AgentMemory(Contract):
# Initializes an EVM mapping(address => string => bytes)
state: dict = {}
@agent_action
def write_state(self, key: str, value: bytes):
"""Only the agent can mutate their specific storage partition"""
self.state[msg.sender][key] = value
@action
def read_state(self, target: address, key: str) -> bytes:
"""Anyone can read the state verifying the data publicly"""
return self.state[target][key]
Immutable Event Logging
Dictionaries map to raw storage which can become expensive during highly recurrent update operations. For append-only logging (example: recording a neural network's decision weights during an epoch), utilize PyVax's built in Event emitters.
python
class AgentLogger(Contract):
@agent_action
def log_decision(self, model_confidence: int, target_node: str):
# Natively transpiles to a Solidity `emit Event(args)` signature
self.emit("DecisionLogged", msg.sender, model_confidence, target_node)
The offline Agent script can stream these events instantly using standard WebSockets.
python
from pyvax import AgentWallet
wallet = AgentWallet("ai-node-1")
def on_log(event):
print(f"Agent {event.sender} executed with confidence: {event.model_confidence}")
wallet.subscribe(
target="0xAgentLogger...",
event="DecisionLogged",
callback=on_log
)