Agent Framework

uni-agent provides a framework for building LLM-powered agents with tool calling, chat sessions, and provider integrations.

Overview

Component	Description
LLM Agent	Agent configuration and creation
Chat Sessions	Conversation management
Tool Integration	Function calling support
AWS Bedrock	Bedrock chat model integration

Quick Start

import wvlet.uni.agent.*
import wvlet.uni.agent.chat.*
import wvlet.uni.agent.runner.AgentRunner
import wvlet.uni.agent.bedrock.BedrockChat

// Define an agent
val agent = LLMAgent(
  name = "assistant",
  description = "A helpful assistant",
  model = LLM.Claude3Sonnet
).withSystemPrompt("You are a helpful assistant.")
 .withTemperature(0.7)

// Create an agent runner with a chat model
val runner = AgentRunner(BedrockChat())

// Create a session and chat
val session = agent.newSession(runner)
val response = session.chat("Hello, how are you?")
println(response.text)

Key Concepts

LLMAgent

Defines an agent with:

• Name and description
• LLM model selection
• System prompt
• Tools (functions)
• Model configuration

ChatSession

Manages conversations:

• Send messages
• Continue conversations
• Stream responses
• Handle tool calls

ToolSpec

Defines callable functions:

• Name and description
• Parameter definitions
• Return type

ChatMessage Types

Type	Description
SystemMessage	System instructions
UserMessage	User input
AIMessage	LLM response (may include tool calls)
AIReasoningMessage	Reasoning process output
ToolResultMessage	Result of tool execution

Modules

// Core agent framework
libraryDependencies += "org.wvlet.uni" %% "uni-agent" % "2026.1.0"

// AWS Bedrock integration
libraryDependencies += "org.wvlet.uni" %% "uni-bedrock" % "2026.1.0"

Package

import wvlet.uni.agent.*
import wvlet.uni.agent.chat.*
import wvlet.uni.agent.tool.*
import wvlet.uni.agent.runner.*