agents-features-snapshot
Provides checkpoint functionality for AI agents, allowing saving and restoring agent state at specific points during execution.
Overview
The agents-features-snapshot module enables checkpoint capabilities for AI agents in the Koog framework, allowing you to:
Resume agent execution from a specific point
Roll back to previous states
Persist agent state across sessions
Implement continuous persistence of agent state
Key features include:
Checkpoint creation and management
Multiple storage providers (in-memory, file-based)
Automatic checkpoint creation after each node execution (optional)
Complete state capture including message history, current node, and input data
Using in your project
To use the checkpoint feature in your project, add the following dependency:
dependencies {
implementation("ai.koog.agents:agents-features-snapshot:$version")
}Then, install the Persistence feature when creating your agent:
val agent = AIAgent(
// other configuration parameters
) {
install(Persistence) {
// Configure the storage provider
storage = InMemoryPersistenceStorageProvider()
// Optional: enable automatic checkpoint creation after each node
enableAutomaticPersistence = true
// Use `RollbackStrategy.Default` if you want to checkpoint the whole state machine and continue from the same node in your strategy graph,
// or `RollbackStrategy.MessageHistoryOnly` if you only want to checkpoint messages
rollbackStrategy = RollbackStrategy.Default
// Optional -- if you want to also roll back all side-effects produced by some tools when rolling back in time to some checkpoint,
// you can also provide a `RollbackToolRegistry` and register rollback-tools for each relevant agent tool in your main `ToolRegistry`
rollbackToolRegistry = RollbackToolRegistry {
registerRollback(::someTool, ::howToRollbackThisTool)
registerRollback(::anotherTool, ::sideEffectsRemover)
}
}
}Example of usage
Here's an example of running the agent with checkpoints:
val agent = AIAgent(
executor = executor,
llmModel = OllamaModels.Meta.LLAMA_3_2,
strategy = singleRunStrategy(ToolCalls.SEQUENTIAL),
) {
install(Persistence) {
storage = snapshotProvider
enableAutomaticPersistence = true
}
}Provides checkpoint functionality for AI agents, allowing saving and restoring agent state at specific points during execution.
Overview
The agents-features-snapshot module enables checkpoint capabilities for AI agents in the Koog framework, allowing you to:
Resume agent execution from a specific point
Roll back to previous states
Persist agent state across sessions
Implement continuous persistence of agent state
Key features include:
Checkpoint creation and management
Multiple storage providers (in-memory, file-based)
Automatic checkpoint creation after each node execution (optional)
Complete state capture including message history, current node, and input data
Using in your project
To use the checkpoint feature in your project, add the following dependency:
dependencies {
implementation("ai.koog.agents:agents-features-snapshot:$version")
}Then, install the Persistence feature when creating your agent:
val agent = AIAgent(
// other configuration parameters
) {
install(Persistence) {
// Configure the storage provider
storage = InMemoryPersistenceStorageProvider()
// Optional: enable automatic checkpoint creation after each node
enableAutomaticPersistence = true
// Use `RollbackStrategy.Default` if you want to checkpoint the whole state machine and continue from the same node in your strategy graph,
// or `RollbackStrategy.MessageHistoryOnly` if you only want to checkpoint messages
rollbackStrategy = RollbackStrategy.Default
// Optional -- if you want to also roll back all side-effects produced by some tools when rolling back in time to some checkpoint,
// you can also provide a `RollbackToolRegistry` and register rollback-tools for each relevant agent tool in your main `ToolRegistry`
rollbackToolRegistry = RollbackToolRegistry {
registerRollback(::someTool, ::howToRollbackThisTool)
registerRollback(::anotherTool, ::sideEffectsRemover)
}
}
}Example of usage
Here's an example of running the agent with checkpoints:
val agent = AIAgent(
executor = executor,
llmModel = OllamaModels.Meta.LLAMA_3_2,
strategy = singleRunStrategy(ToolCalls.SEQUENTIAL),
) {
install(Persistence) {
storage = snapshotProvider
enableAutomaticPersistence = true
}
}