Astreus
Open-source AI agent framework for building autonomous systems that solve real-world tasks effectively.
## What is Astreus?
Astreus is an open-source TypeScript framework designed for building autonomous AI agent systems that can solve real-world tasks effectively. It provides the complete infrastructure needed to create production-ready AI agents with minimal setup time, allowing developers to focus entirely on their agent's unique capabilities rather than dealing with boilerplate code and complex integrations.
The framework was built with a clear philosophy: AI agents should be powerful yet simple to create. Whether you're building a customer service assistant, a research agent, or a complex multi-agent system that coordinates specialized teams, Astreus provides the foundational building blocks that make development straightforward and scalable.
## Core Architecture
At the heart of Astreus lies a carefully designed architecture that balances flexibility with ease of use. Understanding these core components helps you leverage the full power of the framework.
The Agent is the foundational unit of the entire system. Each agent is an autonomous entity capable of reasoning and taking action using language models. Agents can maintain their own persistent memory, manage complex context across conversations, and be equipped with specialized capabilities through plugins and tools. They're designed to be self-contained yet composable, meaning you can create simple single-purpose agents or combine them into sophisticated multi-agent systems.
Sub-Agents enable hierarchical task delegation, which becomes essential when building complex systems. Rather than creating one monolithic agent that tries to do everything, you can design specialized agents that excel at specific domains. A parent agent coordinates these sub-agents through LLM-powered assignment logic, automatically routing tasks to the most capable handler. This architecture mirrors how effective human teams work, with specialists focusing on what they do best while coordinators ensure smooth collaboration.
The Memory System provides per-agent persistent storage with automatic context integration. Unlike simple chat history, Astreus memory uses vector-based similarity search to retrieve relevant historical information. This means agents can recall pertinent details from conversations that happened days or weeks ago, maintaining coherent long-term interactions that feel natural and contextually aware.
Graph Workflows enable complex orchestration patterns that go far beyond simple linear task execution. You can define sophisticated pipelines with conditional branching, parallel execution paths, and multi-agent coordination. Workflows adapt based on intermediate results, making it possible to build agents that handle nuanced scenarios with multiple decision points and fallback strategies.
## Key Features
Astreus comes packed with features that address the real challenges of building production AI agents.
Task Orchestration provides structured task execution with comprehensive status tracking, dependency management, and real-time streaming. Tasks can include file attachments, deadlines, and automatic progress updates. The system handles task queuing, prioritization, and failure recovery, ensuring your agents remain responsive even under heavy workloads.
The Knowledge Base with RAG integration brings retrieval-augmented generation to your agents. Documents are automatically chunked, embedded as vectors, and indexed for semantic search. When an agent needs information, it can query the knowledge base to find relevant content and incorporate it into responses. This makes it trivial to build agents that can reference documentation, company policies, or any other corpus of information.
Multi-LLM Support means you're never locked into a single provider. Astreus provides a unified interface for OpenAI, Anthropic Claude, Google Gemini, and local models through Ollama. Switching between providers requires no code changes, and you can even use different models for different agents within the same system based on cost, capability, or latency requirements.
MCP Integration implements the Model Context Protocol for seamless external tool and service connections. As MCP becomes the emerging standard for AI tool integration, your Astreus agents are ready to connect with a growing ecosystem of compatible tools and services.
The Plugin System offers extensible tool integration with JSON schema validation and automatic LLM function calling. Adding new capabilities is as simple as defining a plugin with its schema and implementation. The framework handles the complex work of presenting tools to the LLM, parsing responses, and executing the appropriate functions.
Vision Processing enables multimodal interactions with built-in image analysis and document processing capabilities. Agents can understand images, extract information from documents, and incorporate visual context into their reasoning.
The Scheduler manages temporal aspects of agent operations, supporting timing controls, recurring tasks, and automated workflow triggers. Build agents that perform actions on schedules, respond to time-based events, or maintain regular maintenance routines.
## Installation
Getting started with Astreus takes just one command:
```bash
npm install @astreus-ai/astreus
```
## Quick Start
Creating your first agent requires minimal code:
```typescript
import { Agent } from '@astreus-ai/astreus';
const agent = await Agent.create({
name: 'Assistant',
model: 'gpt-4o',
memory: true
});
const response = await agent.ask('How can you help me?');
```
From this simple foundation, you can progressively add memory, tools, sub-agents, knowledge bases, and complex workflows as your needs grow.
## Documentation
Comprehensive documentation is available at [astreus.org/docs](https://astreus.org/docs), covering everything from installation and core concepts to advanced patterns and complete API reference. The documentation includes practical examples for common use cases, helping you move from concept to production quickly.
## More Slices From This Pzza
Dive deeper into the ideas and technology behind this project:
- [The Technical Architecture of Modern AI Agents](/oven/technical-architecture-modern-ai-agents) - How modern AI agents are architected with tool use, memory, and planning capabilities
- [Orchestrating Multi-Agent Systems: Technical Patterns for Complex Problem Solving](/oven/orchestrating-multi-agent-systems) - Technical patterns for building systems where multiple AI agents collaborate
- [Understanding AI Agents: Architecture, Implementation, and Future Directions](/oven/understanding-ai-agents-architecture-implementation) - Deep dive into agent architecture from simple chatbots to autonomous systems