ANative-Lab/EvoAgentX

🚀 EvoAgentX: Building a Self-Evolving Ecosystem of AI Agents

What it solves

EvoAgentX is designed to move developers beyond static prompt chaining and manual workflow orchestration. It solves the problem of creating and optimizing complex AI agent systems by automating the construction, evaluation, and iterative improvement of agentic workflows based on a user's goal.

How it works

The framework operates as a self-evolving ecosystem using several core components:

  • WorkFlowGenerator: Automatically assembles a multi-agent workflow from a natural language goal.
  • AgentManager: Instantiates the agents defined in the generated workflow.
  • Self-Evolution Engine: Uses iterative feedback loops and self-evolving algorithms to optimize workflows based on task-specific criteria and automatic evaluators.
  • Tool Integration: Agents can be equipped with a wide array of built-in toolkits (e.g., code interpreters, search engines, databases, and browser automation) or MCP tools to interact with the real world.
  • Memory & HITL: It supports both short-term and long-term memory for persistence and a Human-in-the-Loop (HITL) manager to allow human review and guidance at critical checkpoints.

Who it’s for

It is built for AI researchers, workflow engineers, and startup teams who want to quickly transition from a vague idea to a functional, optimized agentic system with minimal manual engineering.

Highlights

  • Automated Workflow Construction: Generates structured multi-agent workflows from a single prompt.
  • Self-Evolution: Iteratively improves agent behavior and workflows using feedback loops.
  • Extensive Tooling: Includes built-in toolkits for Python/Docker execution, web search, filesystem access, and database operations.
  • Flexible LLM Support: Compatible with OpenAI, Qwen, Claude, DeepSeek, and others via LiteLLM, SiliconFlow, or OpenRouter.
  • Human-in-the-Loop: Supports approval gating and user input collection during execution.

Related

  • Project
    scaleapi/scale-agentex

    A framework for building and deploying AI agents that scales from simple chatbots to fully autonomous systems, supporting both synchronous and asynchronous workflows.

  • Project
    YeQing17-2026/OmniAgent

    An open-source self-evolving agent framework that dynamically updates its skills, memory, and brain model through interaction while employing a multi-layer security scaffold.

  • Project
    wanxingai/LightAgent

    LightAgent is an open‑source Python framework for building lightweight, modular LLM‑driven agents. It supports memory (via mem0), tool integration, multi‑agent routing (LightSwarm), deterministic DAG workflows (LightFlow), streaming OpenAI‑compatible APIs, optional durable sessions, trace observability, guard‑rails, human approval hooks, and a tool‑generator for rapid API‑to‑tool conversion. The library works with any OpenAI‑style provider and is released under Apache‑2.0.

  • Project
    victordibia/designing-multiagent-systems

    A comprehensive multi-agent framework and educational resource that teaches how to build, orchestrate, and evaluate AI agent systems from first principles.

  • Project
    InternLM/lagent

    A PyTorch-inspired framework for building multi-agent AI applications with support for synchronous and asynchronous workflows and integrated tool execution.