Async Agents Framework

A lightweight typescript AI agents framework for building concurrent applications with strong flow control.

Features

• Lock-less Architecture: Stateless agents operate without shared memory locks
• Multi-Core Performance: Utilizes worker threads for true parallelism
• Strong Flow Control: Managed conversation workflows with hooks
• Modular Design: Pluggable agents with tool integration support
• Async Processing: Non-blocking operations with promise-based APIs. Can be integrated with broadcast channel, kafaka, sqs, pubsub, etc.

Agents Patterns

1. Round Robin: Simple task distribution
2. Handoff: Agent-to-agent conversation transfer
3. Tool Calling: Function execution with reflection
4. Selector: Intelligent agent assignment
5. Workflow: Managed multi-step processes

Design Overview

This is an email-like messaging system. Key design elements explained below.

Email-like Messaging:

Each entity (Manager/Agents) has its own virtual mailbox (persistent storage)

Messages have explicit types: NEW, FORWARD, REPLY

All communications are asynchronous and stored

Message Flow:

NEW: End User initiates conversation with Manager

FORWARD: Manager routes messages to agents

REPLY: Agents respond to Manager (not directly to End User)

Final REPLY: Manager responds to End User

Get Started

NPM: npm i async-agents-core or

YARN: yarn add async-agents-core

Try this AI native way to get started. Copy below files and paste them to ask AI.

1. Core code all in one file: https://github.com/ggzy12345/async-agents/blob/main/packages/shared/async-agents-core/project-ai.md

2. Example code all in one file: https://github.com/ggzy12345/async-agents/blob/main/packages/backend/async-agents-backend-example/project-ai.md

3. Example prompt: https://github.com/ggzy12345/async-agents/blob/main/packages/shared/async-agents-core/ai-prompt-examples.md

4. Example coding with AI: