eve

eve is an open-source, filesystem-first framework developed by Vercel for building durable AI agents. Designed to simplify the creation and deployment of autonomous agents, eve treats the filesystem as the primary authoring interface, allowing developers to inspect, extend, and operate agent projects easily. ### Key Features and Project Structure In an eve project, core agent capabilities are organized in conventional directories under an `agent/` folder: - **instructions.md**: The always-on system prompt that defines the agent's behavior. - **tools/**: A directory for typed functions (tools) that the AI model can call to interact with the external world. - **skills/**: Procedures and prompts that the agent loads on demand to perform complex tasks. - **channels/**: Configuration for message channels such as HTTP, Slack, or Discord. - **schedules/**: Recurring cron jobs to trigger agent executions periodically. - **agent.ts**: Optional configuration file specifying the AI model and runtime settings. ### Getting Started To scaffold a new eve project, users can run: bash npx eve@latest init my-agent This command creates the project directory, installs dependencies, initializes Git, and starts an interactive terminal UI. Developers can specify a preferred AI model during setup, such as `openai/gpt-5.6-terra` or `spacexai/grok-4.7`. ### Community and Ecosystem eve is currently in beta and integrates closely with Vercel's ecosystem. The community discusses the framework and shares projects on GitHub Discussions. Contributions are welcome, and security vulnerabilities should be reported through the responsible disclosure email provided in the repository.