CtrlK

Community Documentation Log in Get started

tessl/maven-com-embabel-agent--embabel-agent-starter

Base starter module for the Embabel Agent Framework providing core dependencies for building agentic flows on the JVM with Spring Boot integration and GOAP-based intelligent path finding.

Overview

Eval results

Files

Actions Concept

Name: tessl/maven-com-embabel-agent--embabel-agent-starter
Author: tessl

Actions are executable operations that agents perform. GOAP planner selects and sequences actions based on preconditions, postconditions, cost, and value.

Definition

@Action(
    description = "What this action does",  // Required
    pre = {"condition1", "condition2"},     // Preconditions
    post = {"newCondition"},                // Postconditions
    cost = 10.0,                            // Execution cost
    value = 50.0,                           // Result value
    canRerun = true,                        // Allow multiple executions
    outputBinding = "resultVar"             // Blackboard variable name
)
public ResultType actionMethod(ParamType param, @Provided Ai ai) {
    return result;
}

Parameters

Regular Parameters: Retrieved from blackboard (agent state)

public Result processData(Dataset data, String config) {
    // 'data' and 'config' must be in blackboard
}

Provided Parameters: Injected by platform using @Provided

public Result useAi(Input input, @Provided Ai ai, @Provided ActionContext ctx) {
    // 'ai' and 'ctx' provided by platform
}

Common provided types: Ai, ActionContext, AgentPlatform, PromptRunner, OutputChannel

Preconditions & Postconditions

GOAP uses these to plan action sequences:

@Action(description = "Fetch data", post = {"dataLoaded"})
public Data fetchData(String id) { }

@Action(description = "Validate", pre = {"dataLoaded"}, post = {"dataValidated"})
public ValidationResult validate(Data data) { }

@Action(description = "Process", pre = {"dataValidated"}, post = {"processed"})
public Result process(Data data) { }

Planner automatically sequences: fetch → validate → process

Cost & Value

Static values:

@Action(description = "Quick check", cost = 1.0, value = 10.0)

Dynamic computation:

@Action(description = "Complex operation", costMethod = "computeCost")
public Result operate(Input input) { }

@Cost(name = "computeCost")
public double computeCost(Input input) {
    return input.isComplex() ? 100.0 : 10.0;
}

Output Binding

Store action results in blackboard:

@Action(description = "Load config", outputBinding = "config")
public Config loadConfig() {
    return configService.load();
}

@Action(description = "Use config", pre = {"config"})
public Result useConfig(Config config) {
    // 'config' available from blackboard
}

Tool Access

Actions can use tools:

@Action(
    description = "Process with tools",
    toolGroups = {"data-tools"},
    toolGroupRequirements = {"math-tools"}
)
public Result processWithTools(Input input, @Provided Ai ai) {
    return ai.withTools(tools).createObject("Process: " + input);
}

Retry Policies

Override agent-level retry:

@Action(
    description = "Critical - no retry",
    actionRetryPolicy = ActionRetryPolicy.NO_RETRY
)
public void criticalOp() { }

@Action(
    description = "Resilient operation",
    actionRetryPolicy = ActionRetryPolicy.exponential(10, 500)
)
public Result resilientOp() { }

Conditions

Define reusable condition checks:

@Condition(name = "hasStock", cost = ZeroToOne.of(0.1))
public boolean checkStock(String productId) {
    return inventory.hasStock(productId);
}

@Action(description = "Fulfill order", pre = {"hasStock"})
public Order fulfillOrder(String productId) { }

See Also: Tools | Goals | API Reference

tessl i tessl/maven-com-embabel-agent--embabel-agent-starter@0.3.1