EveStudioDocs

Create an agent

An Eve agent is a folder. Studio can scaffold a fresh one for you, or register an agent you already have on disk. Both live on the left rail, and both are offered on the welcome screen the first time you open the app (Add existing agent and Create new).

Create a new agent

Click the + button at the top of the left rail (its tooltip reads "Create a new agent"). The Create a new agent dialog asks for:

  • Location: the parent folder for the new agent. Choose a folder… opens a native picker.
  • Agent name: lowercase letters, digits, ., _, and -. It becomes the folder name and how the agent is listed in the rail.
  • Model: the AI Gateway model id the agent starts on. The default is Eve's own, openai/gpt-5.6-luna-fast; chips offer other common ids, or type any provider/model. You can change it later from Instructions → Model.
  • Add a Web Chat app (Next.js): tick this to include the Web Chat application, so the agent ships a browser chat UI alongside the built-in HTTP channel. Leave it off for a headless agent you'll reach over HTTP or a platform channel. You can add Web Chat later from Integrations with eve add channel/web.

A Command preview shows exactly what will run, for example:

eve init my-agent --model openai/gpt-5.6-luna-fast --channel-web-nextjs

Below it: "Installs eve@latest (0.49+); needs Node 24 (provisioned automatically when missing)."

Click Create agent. Studio provisions the Node 24 runtime if your machine lacks one, then runs eve init through npx eve@latest in the chosen folder and streams a live log: the scaffold, dependency install, and Git init happen in order.

First run takes a minute. Creating your first agent downloads Eve and its dependencies. That's a one-time cost; later agents reuse the cached runtime and packages. If it fails on a fresh machine, check that you're online.

When it finishes, the new agent appears in the rail, already active, and you land in the workspace. The header reads eve 0.49.x | Vercel · not linked | local · stopped: the next steps are Connect to Vercel and Start.

What gets scaffolded

eve init writes a self-contained Eve app. The part you'll work in is the agent/ folder:

my-agent/
├── package.json
├── agent/
│   ├── instructions.md      # the system prompt
│   ├── agent.ts             # model + config
│   ├── channels/eve.ts      # the built-in HTTP channel
│   ├── tools/               # typed actions the agent can call
│   ├── skills/              # load-on-demand instructions
│   ├── subagents/           # specialist child agents
│   ├── hooks/               # observe the runtime event stream
│   ├── connections/         # MCP / OpenAPI integrations
│   ├── schedules/           # cron-driven jobs
│   ├── memory/              # long-term memory slots
│   └── extensions/          # mounted extension packages
└── evals/                   # scored test cases

Eve discovers every capability from the filesystem: a tool at agent/tools/get_weather.ts is the tool get_weather. A minimal agent is just instructions.md, agent.ts, and the built-in HTTP channel at agent/channels/eve.ts; the other folders fill in as you add capabilities.

For the full mental model behind each folder and how the tabs map onto it, read The agent, explained.

Add an agent you already have

If you built an agent elsewhere (cloned a repo, scaffolded with the Eve CLI, or already work in Eve) use Add existing on the left rail. It opens a folder picker; choose the app root (the folder containing agent/) and Studio registers it as-is. Nothing in the folder is modified. Studio is built for eve 0.49 and newer; an older agent still opens, and the header offers an in-place upgrade (see Keep eve current).

Next

  • Connect to Vercel: give the agent a model credential so it can reply.
  • The workspace: the rail, header, and tabs you'll use to build it out.
  • Chat: start talking to your new agent.