EVA

If an agent's execution layer is as small as a single script, it has the same viral spread potential.

EVA is a small but fully-featured agent, a lightweight alternative to Claude Code. It can help you write scripts, write test cases, run shell commands, analyze data, and more. I'm a heavy EVA user myself and use it very frequently in my daily work.

Check out the showcase in this repository for all kinds of fun use cases 🦖🦖🦖

Features

• Local deployment support: Connect to locally deployed OpenAI API-compatible models like vLLM, or external online models
• Ultra lightweight: It's a single file (only eva.py) and runs as long as you have Python installed
• Directory-level sessions: When you launch it in the same directory next time, it will continue your previous conversation
• Built-in security review: By default, only read-only commands are executed automatically, and all other commands require your explicit confirmation
• High portability: It's very easy to integrate EVA into your existing automation workflows. For example: eva -au 'Calculate the sum of all prime numbers under 1 million and write the result to /tmp/result.txt'. I've currently integrated EVA into a WeChat bot using the -asu flag.