vercel/eve

eve is a framework for building AI agents, software that can hold conversations, call external tools, run on schedules, and connect to platforms like Slack or Discord. Instead of writing a lot of glue code to wire an AI model to these capabilities, you organize your agent's parts into folders and files, and the framework handles the connections. The core idea is that the filesystem is the interface. You create an agent/ directory with specific subfolders: instructions.md holds the system prompt telling the agent how to behave, tools/ contains typed functions the AI can call (like fetching weather data), skills/ holds procedures loaded on demand, channels/ connects the agent to messaging platforms, and schedules/ runs recurring tasks. Each piece lives in a predictable location, so you can look at a project folder and immediately understand what the agent does. Getting started is a single command, npx eve@latest init my-agent, which scaffolds the project, installs dependencies, and launches an interactive terminal UI. From there, you write a markdown file with your agent's instructions, optionally add tool functions, pick an AI model like Claude, and run it. The README shows a minimal weather assistant built this way in just a few lines. You can then layer in features like human-in-the-loop approval prompts, subagents, and cron-style scheduled jobs as needed. This is built for developers and teams who want to ship production AI agents without reinventing the plumbing each time. A startup building a customer support bot that needs to escalate to humans, or a team creating an internal assistant that posts weekly Slack recaps, could use this to stand up the agent quickly and keep the codebase legible as it grows. One notable detail: the package ships with its full documentation bundled inside, so coding assistants working in your project can read the docs locally rather than relying on external lookups. The project is currently in beta, so APIs and behavior may shift before a stable release.