EvoScientist aims to advance AI-driven scientific discovery by creating self-evolving AI scientists capable of autonomous exploration, insight generation, and iterative self-improvement. Designed for researchers and power users, it offers a living research system that integrates evolving agent skills, toolsets, and memory, moving towards a human-on-the-loop paradigm where AI acts as a co-evolving research partner that internalizes scholarly judgment.

How It Works

The system employs a multi-agent architecture comprising six specialized agents (plan, research, code, debug, analyze, write) that collaborate on scientific tasks. It features persistent memory across sessions, supports multiple LLM providers (Anthropic, OpenAI, Google, NVIDIA) via a single configuration, and enables multi-channel communication through CLI, Telegram, Slack, and other platforms. EvoScientist follows a structured scientific workflow: Intake → plan → execute → evaluate → write → verify, with extensibility through MCP servers and on-the-fly skill installation from GitHub.

Quick Start & Requirements

Installation is straightforward using uv or pip: uv tool install EvoScientist or pip install EvoScientist. Development installs require cloning the repository and using uv sync --dev or pip install -e ".[dev]". Conda environments are also supported. The project requires Python 3.11+ (specifically versions less than 3.14). Configuration involves an interactive wizard (EvoSci onboard) or manual setup via environment variables or a .env file for API keys. The primary execution command is EvoSci, which launches an interactive TUI by default, with options for single-shot mode (-p), specific work directories (--workdir), or headless operation (serve).

Highlighted Details

• Recognized with the Best Paper & Appraisal Award and ranked #1 on the DeepResearch Bench II.
• Features a multi-agent team with persistent memory and multi-provider LLM support.
• Supports a wide array of communication channels including CLI, Telegram, Slack, Feishu, and WeChat.
• Integrates with MCP servers and offers installable "EvoSkills" for end-to-end research lifecycle coverage.

Maintenance & Community

Officially debuted on March 13, 2026, EvoScientist has a core team of listed contributors and welcomes contributions from the wider open-source community. Inquiries can be directed to EvoScientist.ai@gmail.com. The project provides detailed contributing guidelines for both human and AI agents.

Licensing & Compatibility

The project is licensed under the Apache License 2.0, which permits commercial use and integration into closed-source projects.

Limitations & Caveats

The system requires Python versions between 3.11 and 3.13. Features such as OAuth sign-in, a web application UI, and a dedicated benchmark suite are listed as upcoming on the roadmap. By default, shell command execution requires explicit human approval, though this can be configured for automated execution or specific command whitelisting.