OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Agent Frameworks

Agentevolver

AgentEvolver is an open-source Python framework for training self-evolving AI agent systems. It combines automatic task generation, experience-guided exploration and attribution-based credit assignment, with services for connecting environments, language models and experience management.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Agentevolver — OpenViST project profile in Agent Frameworks.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use it when you are developing or researching agent training workflows and want to combine task generation, experience reuse and trajectory attribution. The README’s concrete setup example uses AppWorld.

Who it is for

  • Developers building LLM-agent training systems.
  • Researchers exploring agent self-improvement and reinforcement-learning workflows.

Documented uses

  • Generate diverse training tasks from an environment rather than relying solely on manually assembled datasets.
  • Summarize and reuse cross-task experience to guide agent exploration.
  • Analyze long trajectories to attribute intermediate steps and support policy optimization.
  • Connect agent training workflows to environments and tool APIs through standardized interfaces.
Before you choose

Requirements & limits

The public GitHub repository identifies the project as open source under the Apache-2.0 license.

  • The setup uses Conda and the CUDA toolkit.
  • Training configuration requires an API key and Conda path.
  • ReMe is optional; the README documents both a basic run without it and a fuller ReMe-enabled run.
  • The framework’s architecture connects environment sandboxes, LLMs and experience management; it is a developer framework rather than a hosted training service.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Set up the AppWorld training example

Prepare the documented AppWorld example and start AgentEvolver training.

Before you start

  • Conda
  • CUDA toolkit
  • An API key and configured Conda path
  • Python 3.11+ is indicated by the repository

Steps

  1. Run the command below to set up the training environment.

    Documented command
    bash install.sh

  2. Run cd env_service/environments/appworld && bash setup.sh to prepare the AppWorld environment.

  3. Copy example.env to .env and edit the API key and Conda path.

  4. Activate the agentevolver Conda environment and launch either the command below or, with ReMe configured, the command below.

    Documented command
    python launcher.py --conf examples/basic.yaml --with-appworld
    Documented command
    python launcher.py --conf examples/overall.yaml --with-appworld --with-reme

Expected outcome

Starts the documented AppWorld training workflow; the basic option uses built-in environment datasets, while the ReMe option enables questioning, navigating and attributing.

Guide limitations

  • The repository requires Conda and the CUDA toolkit.
  • ReMe is optional and must be set up separately for the ReMe-enabled option.
  • The README does not specify API provider requirements beyond an API key.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
11 October 2026
Original website title
modelscope/AgentEvolver · Official homepage title · retained in its source language
How we review sourcesSuggest a correction