usepr/eva

单文件智能体，Single-file zero-dependency AI agent for Linux/Windows ops

What it solves

EVA is a lightweight AI agent designed to automate technical tasks like writing scripts, creating test cases, executing shell commands, and analyzing data. It aims to provide a highly portable, "single-file" alternative to more complex agent frameworks, making it easy to deploy in restricted environments like Docker or remote servers.

How it works

EVA operates as a single Python file (eva.py) that connects to LLMs via OpenAI-compatible APIs (such as DeepSeek, vLLM, or Ollama). It manages context through directory-scoped sessions, meaning it remembers previous conversations when restarted in the same folder. To ensure safety, it defaults to a read-only mode for commands, requiring user confirmation for any write or execution actions unless the --allow-all flag is used.

Who it’s for

It is built for developers and system administrators who need a quick, portable AI assistant for shell-based automation and data analysis without the overhead of complex installation processes.

Highlights

  • Single-File Architecture: The entire agent is contained in one eva.py file for "paste and go" deployment.
  • Directory-Scoped Sessions: Automatically persists and resumes conversations based on the current working directory.
  • Flexible LLM Integration: Supports any OpenAI-compatible API, including local models.
  • Safety Controls: Built-in security review that requires confirmation for potentially dangerous commands.
  • Customizable Behavior: Uses an EVA.md file to define the agent's persona, settings, and specific skills.

Related

  • Project
    evotai/evot

    Evot is a lightweight terminal‑based coding agent that lets a LLM read, edit, write files and run shell commands. With a tiny prompt, four core tools, background execution, and plug‑in support for many LLM providers, it offers fast, low‑cost AI‑assisted programming directly from the command line.

  • Project
    SWE-agent/mini-swe-agent

    A minimal AI software engineering agent that uses only bash commands to solve coding tasks, providing a high-performance, hackable alternative to complex agent frameworks.

  • Project
    fastclaw-ai/fastclaw

    A lightweight AI agent runtime written in Go that allows users to create, manage, and deploy multi-agent systems with sandboxed tool execution and IM channel integrations.

  • Project
    vercel/eve

    eve is a Vercel‑backed, filesystem‑first framework for building, running, and deploying durable AI agents. Agents are defined by a clear folder layout (prompts, typed tools, skills, channels, schedules) and can be scaffolded with `npx eve init`. The CLI handles model selection, local dev (`npm run dev`), and Vercel deployment with web‑chat integration.

  • Project
    huggingface/OpenEnv

    OpenEnv is a Hugging Face‑hosted framework that lets you create, deploy (as Docker images) and interact with isolated RL environments via a Gymnasium‑style API, supporting async/sync calls, tool‑exposing MCP environments, and integrations with many RL libraries.