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

Model Providers

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

Type-safe LLM model identifiers and constants for various providers available through embabel-agent-api.

Importing Model Constants

Java

import com.embabel.agent.api.models.GeminiModels;
import com.embabel.agent.api.models.AnthropicModels;
import com.embabel.agent.api.models.OpenAiModels;
import com.embabel.agent.api.models.DeepSeekModels;
import com.embabel.agent.api.models.OllamaModels;
import com.embabel.agent.api.models.MistralAiModels;
import com.embabel.agent.api.models.GoogleGenAiModels;
import com.embabel.agent.api.models.LmStudioModels;
import com.embabel.agent.api.models.DockerLocalModels;

Kotlin

import com.embabel.agent.api.models.GeminiModels
import com.embabel.agent.api.models.AnthropicModels
import com.embabel.agent.api.models.OpenAiModels
import com.embabel.agent.api.models.DeepSeekModels
import com.embabel.agent.api.models.OllamaModels
import com.embabel.agent.api.models.MistralAiModels
import com.embabel.agent.api.models.GoogleGenAiModels
import com.embabel.agent.api.models.LmStudioModels
import com.embabel.agent.api.models.DockerLocalModels

Google Gemini Models

Model Constants

public class GeminiModels {
    // Gemini 3.0 Series
    public static final Llm GEMINI_3_PRO_PREVIEW;

    // Gemini 2.5 Series
    public static final Llm GEMINI_2_5_PRO;
    public static final Llm GEMINI_2_5_FLASH;
    public static final Llm GEMINI_2_5_FLASH_LITE;

    // Gemini 2.0 Series
    public static final Llm GEMINI_2_0_FLASH;
    public static final Llm GEMINI_2_0_FLASH_LITE;

    // Embedding Models
    public static final EmbeddingModel TEXT_EMBEDDING_004;
    public static final EmbeddingModel DEFAULT_TEXT_EMBEDDING_MODEL;
}

Model Characteristics

Model	Use Case	Context Window	Speed
GEMINI_3_PRO_PREVIEW	Advanced reasoning, complex tasks	Large	Moderate
GEMINI_2_5_PRO	Production workloads, balanced	Very Large	Moderate
GEMINI_2_5_FLASH	Fast responses, simple tasks	Large	Fast
GEMINI_2_5_FLASH_LITE	Lightweight, high-throughput	Medium	Very Fast
GEMINI_2_0_FLASH	Legacy fast responses	Medium	Fast
GEMINI_2_0_FLASH_LITE	Legacy lightweight	Small	Very Fast

Usage Example (Java)

import com.embabel.agent.api.models.GeminiModels;
import com.embabel.agent.api.llm.Ai;

public class MyAgent {
    public Result process(Input input, Ai ai) {
        return ai
            .withLlm(GeminiModels.GEMINI_2_5_PRO)
            .createObject("Process this input: " + input);
    }

    public Embedding embed(String text, Ai ai) {
        return ai
            .withEmbeddingModel(GeminiModels.TEXT_EMBEDDING_004)
            .embed(text);
    }
}

Configuration

spring:
  ai:
    gemini:
      api-key: ${GOOGLE_STUDIO_API_KEY}

Requires embabel-agent-starter-gemini dependency.

Anthropic Models

Model Constants (Kotlin)

object AnthropicModels {
    // Claude 3.7 Series
    val CLAUDE_37_SONNET: Llm

    // Claude 3.5 Series
    val CLAUDE_35_HAIKU: Llm

    // Claude 4.0 Series
    val CLAUDE_40_OPUS: Llm
    val CLAUDE_41_OPUS: Llm

    // Claude 4.5 Series (Latest)
    val CLAUDE_SONNET_4_5: Llm
    val CLAUDE_HAIKU_4_5: Llm
}

Model Characteristics

Model	Use Case	Context Window	Speed
CLAUDE_SONNET_4_5	Latest, most capable	Very Large	Moderate
CLAUDE_HAIKU_4_5	Fast, lightweight	Large	Fast
CLAUDE_41_OPUS	Advanced reasoning	Very Large	Slower
CLAUDE_40_OPUS	Complex tasks	Large	Slow
CLAUDE_37_SONNET	Balanced performance	Large	Moderate
CLAUDE_35_HAIKU	Quick responses	Medium	Very Fast

Usage Example (Kotlin)

import com.embabel.agent.api.models.AnthropicModels

fun processWithClaude(input: Input, ai: Ai): Result {
    return ai
        .withLlm(AnthropicModels.CLAUDE_SONNET_4_5)
        .createObject("Process this: $input")
}

Configuration

spring:
  ai:
    anthropic:
      api-key: ${ANTHROPIC_API_KEY}

Requires embabel-agent-starter-anthropic dependency.

OpenAI Models

Model Constants (Kotlin)

object OpenAiModels {
    // GPT-5 Series (Latest)
    val GPT_5: Llm
    val GPT_5_MINI: Llm
    val GPT_5_NANO: Llm

    // GPT-4.1 Series
    val GPT_41: Llm
    val GPT_41_MINI: Llm
    val GPT_41_NANO: Llm

    // Embedding Models
    val TEXT_EMBEDDING_3_SMALL: EmbeddingModel
}

Model Characteristics

Model	Use Case	Context Window	Speed
GPT_5	Most capable, latest	Very Large	Moderate
GPT_5_MINI	Balanced cost/performance	Large	Fast
GPT_5_NANO	Lightweight, high-throughput	Medium	Very Fast
GPT_41	Previous generation, capable	Large	Moderate
GPT_41_MINI	Efficient, cost-effective	Medium	Fast
GPT_41_NANO	Ultra-fast responses	Small	Very Fast

Usage Example (Kotlin)

import com.embabel.agent.api.models.OpenAiModels

fun processWithGPT(input: Input, ai: Ai): Result {
    return ai
        .withLlm(OpenAiModels.GPT_5)
        .createObject("Process: $input")
}

Configuration

spring:
  ai:
    openai:
      api-key: ${OPENAI_API_KEY}

Requires embabel-agent-starter-openai dependency.

DeepSeek Models

Model Constants (Kotlin)

object DeepSeekModels {
    const val DEEPSEEK_CHAT = "deepseek-chat"
    const val DEEPSEEK_REASONER = "deepseek-reasoner"
    const val PROVIDER = "Deepseek"
}

Model Characteristics

Model	Use Case
DEEPSEEK_CHAT	General-purpose chat and completion
DEEPSEEK_REASONER	Advanced reasoning and complex problem-solving

Usage Example (Kotlin)

import com.embabel.agent.api.models.DeepSeekModels

fun processWithDeepSeek(input: Input, ai: Ai): Result {
    return ai
        .withLlm(DeepSeekModels.DEEPSEEK_CHAT)
        .createObject("Analyze: $input")
}

fun complexReasoning(problem: Problem, ai: Ai): Solution {
    return ai
        .withLlm(DeepSeekModels.DEEPSEEK_REASONER)
        .createObject("Solve this problem: $problem")
}

Ollama Models

Model Constants (Kotlin)

object OllamaModels {
    // Ollama-hosted local model constants
}

Usage Example (Kotlin)

import com.embabel.agent.api.models.OllamaModels

fun processLocally(input: Input, ai: Ai): Result {
    return ai
        .withLlm(OllamaModels.LLAMA3_2)
        .createObject("Process: $input")
}

Configuration

spring:
  ai:
    ollama:
      base-url: http://localhost:11434

Requires embabel-agent-starter-ollama dependency and Ollama running locally.

Mistral AI Models

Model Constants (Kotlin)

object MistralAiModels {
    // Mistral AI model constants
}

Usage Example (Kotlin)

import com.embabel.agent.api.models.MistralAiModels

fun processWithMistral(input: Input, ai: Ai): Result {
    return ai
        .withLlm(MistralAiModels.MISTRAL_LARGE)
        .createObject("Process: $input")
}

Google GenAI Models

Model Constants (Kotlin)

object GoogleGenAiModels {
    // Google GenAI model constants
}

LM Studio Models

Model Constants (Kotlin)

object LmStudioModels {
    // LM Studio model constants
}

Configuration

LM Studio uses the OpenAI-compatible client. Ensure your LM Studio server is running.

Docker Local Models

Model Constants (Kotlin)

object DockerLocalModels {
    // Docker-hosted model constants
}

Using Model Constants in Actions

Selecting Models at Runtime

import com.embabel.agent.api.models.GeminiModels;
import com.embabel.agent.api.models.OpenAiModels;
import com.embabel.agent.api.llm.Ai;
import com.embabel.agent.api.llm.LlmOptions;

@Action
public Result intelligentRouting(ComplexInput input, Ai ai) {
    // Use cheaper, faster model for simple classification
    Category category = ai
        .withLlm(GeminiModels.GEMINI_2_5_FLASH_LITE)
        .createObject("Classify: " + input.getPreview(), Category.class);

    // Use more capable model for complex processing
    if (category.isComplex()) {
        return ai
            .withLlm(OpenAiModels.GPT_4_TURBO)
            .createObject("Deep analysis: " + input.getFullContent(), Result.class);
    } else {
        return ai
            .withLlm(GeminiModels.GEMINI_2_5_FLASH)
            .createObject("Quick process: " + input.getFullContent(), Result.class);
    }
}

With Advanced Options

import com.embabel.agent.api.llm.LlmOptions;

@Action
public CreativeResult generateCreative(Prompt prompt, Ai ai) {
    LlmOptions options = LlmOptions
        .withModel(GeminiModels.GEMINI_2_5_PRO)
        .withTemperature(0.9)  // High temperature for creativity
        .withMaxTokens(2000);

    return ai
        .withLlm(options)
        .createObject("Generate creative content: " + prompt, CreativeResult.class);
}

Kotlin DSL

@Action
fun processWithOptions(input: Input, ai: Ai): Result {
    val options = LlmOptions
        .withModel(GeminiModels.GEMINI_2_5_PRO)
        .withTemperature(0.2)  // Low temperature for consistency
        .withMaxTokens(1000)

    return ai
        .withLlm(options)
        .createObject("Analyze: $input")
}

Default Model Configuration

Configure default models in application.yml:

embabel:
  models:
    default-llm: gpt-4.1-mini  # Default LLM for all operations
    default-embedding-model: text-embedding-3-small  # Default embedding model

Using Default Models

@Action
public Result processWithDefault(Input input, Ai ai) {
    // Uses default LLM configured in application.yml
    return ai
        .withDefaultLlm()
        .createObject("Process: " + input);
}

Model Selection Strategies

Cost Optimization

// Use fast, cheap models for simple tasks
GeminiModels.GEMINI_2_5_FLASH_LITE  // Simple classification
GeminiModels.GEMINI_2_5_FLASH        // Quick Q&A

// Use capable models for complex tasks
GeminiModels.GEMINI_2_5_PRO          // Complex reasoning
OpenAiModels.GPT_4_TURBO             // Advanced analysis

Privacy and Local Models

// For sensitive data, use local models
OllamaModels.LLAMA3_2               // Local inference
LmStudioModels.LOCAL_MODEL          // LM Studio hosted
DockerLocalModels.CONTAINER_MODEL   // Docker container

Latency Optimization

// Fastest models
GeminiModels.GEMINI_2_5_FLASH_LITE
GeminiModels.GEMINI_2_0_FLASH_LITE

// Moderate speed
GeminiModels.GEMINI_2_5_FLASH
OpenAiModels.GPT_3_5_TURBO

// More capable but slower
GeminiModels.GEMINI_2_5_PRO
OpenAiModels.GPT_4_TURBO

Multi-Model Consensus

Use multiple models for critical decisions:

import com.embabel.agent.api.consensus.ConsensusBuilder;

@Action
public Decision criticalDecision(Problem problem, Ai ai) {
    return ConsensusBuilder
        .forDecision(Decision.class)
        .withModels(
            GeminiModels.GEMINI_2_5_PRO,
            OpenAiModels.GPT_4_TURBO,
            AnthropicModels.CLAUDE_3_5_SONNET
        )
        .requireConsensus(0.67)  // 2 out of 3 must agree
        .execute(ai, "Decide on: " + problem);
}

Provider Configuration

OpenAI

spring:
  ai:
    openai:
      api-key: ${OPENAI_API_KEY}

Requires embabel-agent-starter-openai dependency.

Anthropic

spring:
  ai:
    anthropic:
      api-key: ${ANTHROPIC_API_KEY}

Requires embabel-agent-starter-anthropic dependency.

Google Gemini

spring:
  ai:
    gemini:
      api-key: ${GOOGLE_STUDIO_API_KEY}

Requires embabel-agent-starter-gemini dependency.

Ollama (Local)

spring:
  ai:
    ollama:
      base-url: http://localhost:11434

Requires embabel-agent-starter-ollama dependency and Ollama running locally.

AWS Bedrock

spring:
  ai:
    bedrock:
      aws:
        region: us-east-1
        access-key: ${AWS_ACCESS_KEY_ID}
        secret-key: ${AWS_SECRET_ACCESS_KEY}

Requires embabel-agent-starter-bedrock dependency.

Resilience Configuration

Provider-Specific Retry Configuration

embabel:
  agent:
    platform:
      models:
        anthropic:
          max-attempts: 10                  # Maximum retry attempts
          backoff-millis: 5000              # Initial backoff in milliseconds
          backoff-multiplier: 5.0           # Backoff multiplier
          backoff-max-interval: 180000      # Maximum backoff interval (3 minutes)
        openai:
          max-attempts: 10
          backoff-millis: 5000
          backoff-multiplier: 5.0
          backoff-max-interval: 180000

Backoff Calculation

Delay = min(backoff-millis * (backoff-multiplier ^ attempt), backoff-max-interval)

Example Retry Sequence:

Attempt 1: 5000ms (5 seconds)
Attempt 2: 25000ms (25 seconds)
Attempt 3: 125000ms (125 seconds)
Attempt 4+: 180000ms (180 seconds, capped)

Best Practices

Start with cheap models - Use fast, inexpensive models for simple tasks
Route intelligently - Direct complex tasks to capable models
Use local models for sensitive data - Privacy-critical operations should use local models
Configure defaults - Set reasonable defaults in application.yml
Test with multiple models - Different models have different strengths
Monitor costs - Track usage and optimize model selection
Use embeddings appropriately - Choose embedding models based on use case
Consider latency - Balance capability with response time requirements

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

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

integration-model-providers.md.css-3qkkll{font-size:var(--chakra-font-sizes-sm);font-weight:var(--chakra-font-weights-normal);color:var(--chakra-colors-gray-300);}docs/

Model Providers

Importing Model Constants

Java

Kotlin

Google Gemini Models

Model Constants

Model Characteristics

Usage Example (Java)

Configuration

Anthropic Models

Model Constants (Kotlin)

Model Characteristics

Usage Example (Kotlin)

Configuration

OpenAI Models

Model Constants (Kotlin)

Model Characteristics

Usage Example (Kotlin)

Configuration

DeepSeek Models

Model Constants (Kotlin)

Model Characteristics

Usage Example (Kotlin)

Ollama Models

Model Constants (Kotlin)

Usage Example (Kotlin)

Configuration

Mistral AI Models

Model Constants (Kotlin)

Usage Example (Kotlin)

Google GenAI Models

Model Constants (Kotlin)

LM Studio Models

Model Constants (Kotlin)

Configuration

Docker Local Models

Model Constants (Kotlin)

Using Model Constants in Actions

Selecting Models at Runtime

With Advanced Options

Kotlin DSL

Default Model Configuration

Using Default Models

Model Selection Strategies

Cost Optimization

Privacy and Local Models

Latency Optimization

Multi-Model Consensus

Provider Configuration

OpenAI

Anthropic

Google Gemini

Ollama (Local)

AWS Bedrock

Resilience Configuration

Provider-Specific Retry Configuration

Backoff Calculation

Best Practices

integration-model-providers.mddocs/