For Developers

Write the agent.
We host the rest.

Scaffold an agent, run guuey deploy, and get a hosted endpoint on isolated infrastructure — sandboxed, scaled to zero when idle. Bring Claude Agent SDK, OpenAI Agents, or Google ADK code, or skip code entirely with a declarative config.

Hosted

No servers.

Your agent runs on gVisor-isolated pods that scale to zero when idle. Conversations stream live and persist as history.

Batteries

MCP + LLM, managed.

Hosted MCP servers deploy alongside your agent. Managed LLM keys with metered usage — no provider account required to start.

Console

Apps, env, billing.

The builder console — manage apps, env vars and secrets per agent, deployments, and usage-based billing in one place. Console access comes with your beta invite.

Ship it

Scaffold, then deploy.

npx @guuey/create-agentic-app my-agent
cd my-agent
pnpm install
pnpm dev            # local: your agent + your MCP servers, hot reload

guuey login
guuey deploy        # hosted: agent + MCP servers live on guuey

The scaffold gives you a code-mode agent worker (pick a framework with --framework), an MCP server template built and hosted alongside it, and generative UI wired in. The same worker runs locally and on the hosted runtime. Everything up to pnpm dev works today, straight from npm — the scaffold, your MCP servers, and generative UI all run locally. guuey deploy needs a platform account: private beta now, open to everyone in the August 2026 public beta.

Or skip the code

One config file is a whole agent.

guuey.json

{
  "schema": "1",
  "agent": {
    "mode": "declarative",
    "framework": "claude-agent-sdk",
    "model": "claude-sonnet-5",
    "systemPrompt": "You are a plant-care assistant. Check the local weather before recommending a watering schedule."
  }
}

Declarative mode deploys with no source to build — the config is the agent. The default MCP server is mcp.ggui.ai, so declarative agents get generative UI out of the box.

What you get

A production agent, not a demo.

A hosted agent endpoint — invocations stream over SSE straight from your agent's pod.
Isolated infrastructure: gVisor sandboxing per agent, scale-to-zero when idle.
Hosted MCP servers, deployed from your repo with the agent — plus generative UI via the default mcp.ggui.ai server.
Managed LLM keys with metered usage, per-agent env vars and secrets, and automatic conversation history.

Embed it in your app

Your agent, in your UI.

Every hosted agent answers a public POST /agent/invoke endpoint over SSE — call it directly, or drop in @guuey/agent-client's React hook for a working chat surface in about fifteen lines. There's also a drop-in script embed — one tag, no React required — already live for private-beta builders; the embed guide has the snippet.

Read the embed guide →

Private beta

Public beta: August 2026.

We're onboarding builders in a private beta now. Drop your email and we'll send an invite when a spot opens.

See the open protocol

Write the agent.We host the rest.