Find your next open-source project

Search by name, purpose, or keyword — or paste a GitHub repository URL

vercel/eve

Developer Tools

A filesystem-first TypeScript framework for building durable AI agents with tools, skills, channels and schedules.

5,970
665
TypeScript
Apache-2.0
5 months old

What it is

eve is a filesystem-first framework for durable AI agents, published by Vercel. Instead of burying agent logic in configuration, the project keeps core capabilities in conventional directories under agent/: instructions.md (the always-on system prompt), optional agent.ts for model and runtime config, tools/ for typed functions the model can call, skills/ for procedures loaded on demand, channels/ for HTTP, Slack and Discord message channels, and schedules/ for recurring cron jobs. The aim is that a project stays inspectable, extendable and operable.

It targets TypeScript/JavaScript developers who want to define agents as code and Markdown, run them locally, and later deploy them without rewriting the project layout. Models are referenced by AI Gateway model IDs.

Quick start

Create a new agent from the CLI:

npx eve@latest init my-agent

This creates my-agent, installs dependencies, initializes Git and opens an interactive terminal UI. The scaffold defaults to openai/gpt-6-luna-fast with high reasoning; select another model with --model (and optionally --reasoning). To add eve to an existing project, run npx eve@latest init . from that project's directory.

The generated agent/instructions.md is the system prompt. A tool is a TypeScript file exporting defineTool with a description, a Zod input schema and an execute function, for example a mock get_weather.ts returning { city, condition: "Sunny", temperatureF: 72 }. The model is set in agent/agent.ts via defineAgent({ model, reasoning }). After editing, start the scaffold with npm run dev to get a working agent.

To add a web chat, run npx eve link and npx eve add channel/web, choose a hosting layout and sign in with Vercel to provision authentication for the project team; npx eve deploy publishes. Local development does not require browser sign-in. See the Web Chat setup guide for permissions and recovery.

Constraints and project details

The framework is TypeScript-based and the CLI scaffolds a project with dependencies and Git initialized; rewriting the agent requires the documented agent/ layout rather than arbitrary file names. Web Chat and deployment flows depend on linking a Vercel project and authenticating team members, so that channel is tied to Vercel rather than being a standalone option. The package ships its documentation inside node_modules/eve/docs, which lets coding agents read it locally.

The project is released under Apache-2.0. Security issues should be reported privately per SECURITY.md rather than through public issues. Contributions go through CONTRIBUTING.md and GitHub Discussions.

HotRepo's take

On the 2026-11-18 TypeScript daily chart, eve recorded 5,958 stars and 665 forks, with 38 stars gained and a best position of 15 on that list. That is a single snapshot, so it establishes where the project placed that day but not a trend. For readers evaluating it: the value is in the explicit agent/ directory convention and a short path from npx eve@latest init to a running agent, not in abstract "framework" positioning. Teams already on Vercel get the tightest integration; those avoiding Vercel should weigh the deployment and Web Chat coupling before adopting.

Written by AI from the project's README, updated . The project's own docs are the source of truth.

You might also like

  1. cordiverse/cordisA TypeScript meta-framework for building plugin systems with runtime dependency injection and hot-reloadable components.TypeScript9,687
  2. vercel/aiA TypeScript library for calling many LLM providers through one API and wiring models into UI frameworks and agents.TypeScript27,500
  3. vuejs/vueProgressive JavaScript framework for building user interfaces, now in end-of-life Vue 2.TypeScript212,509
  4. mastra-ai/mastraA TypeScript framework for building AI agents and workflows, with model routing, memory, RAG, evals, and observability.TypeScript29,561