EvoScientist

Active
GitHub Python Apache-2.0

Description

A self-evolving AI-scientist framework where six sub-agents (plan, research, code, debug, analyze, write) collaborate to complete end-to-end research workflows.

Key Features

  • Six sub-agents in concert - plan, research, code, debug, analyze, and write collaborate on the full research loop
  • Self-evolving memory - auto-distills user profile and observations every turn, growing across sessions
  • Multi-provider switching - Anthropic, OpenAI, Google, MiniMax, and NVIDIA share one config
  • Multi-channel - CLI is the hub, but Telegram, Slack, Feishu, and WeChat share the same session
  • Desktop WebUI - bundled workspace panel that starts with a single flag
  • MCP and skills ecosystem - hot-pluggable MCP servers plus installable research skills from GitHub

Use Cases

💡 Run end-to-end research workflows - literature review, reproduction, write-up
💡 Benchmark research agents on DeepResearch Bench, AstaBench, and ResearchClawBench
💡 Plug team data sources, databases, and APIs into a private research assistant
💡 Connect EvoScientist to Feishu/WeChat/DingTalk for daily research Q&A and code execution
💡 Codify reusable research skills as EvoSkills packages for the team

Tags

multi-agent research self-evolving deep-research python

Categories

🧠 Multi-Agent

Quick Start

pip install EvoScientist
export ANTHROPIC_API_KEY=<your-key>
# Run a deep research task from the CLI
evosci run "Survey the latest 2026 benchmarks for LLM agents"
# Launch the desktop WebUI
evosci --ui webui run
Visit GitHub Visit Website

Related Projects