jeffgreen311/eve-agent-v2-unleashed

Eve Agent V2 Unleashed is a Python application that gives you an autonomous AI coding assistant running on your own computer. It connects to Ollama, a tool for running large language models locally, and uses those models to plan and execute programming tasks from start to finish without you having to guide each step. You describe what you want built, and the agent reads your project, writes code across multiple files, runs tests, and fixes errors on its own. The agent runs up to 40 rounds of tool calls per task. The tools it can use include running shell commands, reading and writing files, searching through code with grep, making git commits, and fetching pages from the web. It also supports live web search through a service called Tavily. The interface is a web page served at localhost:7777 with a cyberpunk visual theme. The project ships several custom AI model variants, ranging from a 2.6 GB model that runs on most GPUs to an optional 480B parameter cloud model accessible through Ollama. The default models are fine-tuned versions of Qwen models with a custom persona. Installation requires Ollama to already be installed and a model pulled from the project's Ollama or Hugging Face pages. The README describes a test where the agent was given a single prompt to build a FastAPI server with a test suite, and all nine tests passed on the first attempt without corrections. Additional features include slash commands for common tasks like code review and refactoring, a quest system where you drop a Markdown file into a folder and the agent runs it automatically, Telegram notifications for mobile access, and Windows-native support with a one-click batch file launcher. Licensed under MIT.