GCP Computer
GDG HACKATHON PROTOCOL

Your Secure Sandbox on Google Cloud Platform

GCP Computer provides AI coding agents with isolated Linux workspaces in milliseconds. Execute terminal commands, edit directories, mount repositories safely, and let the agent work without compromising your local operating system.

Google Cloud compute backend
|
Vercel EVE Agent loops
|
Zero Local Setup Required
terminal-emulator://gcp-sandbox-vm
STATE:Instant Workspace Spawning
gcp-computer:~$ spawn-sandbox --provider gcp-compute-engine --image node-22-alpine

⚡ Initializing Sandbox Manager...

✔ Resolving Compute Engine credentials [gcp-key.json]

✔ Spin-up sequence initiated for VM 'gcp-sandbox-us-west1-b'

✔ VM status: RUNNING (142ms latency)

✔ Local directory mounted at: /workspace -> c:/Users/ethan/project

🚀 Sandbox secure container is ready.

gcp-computer:~$

Platform Engine

Provisioned for Agents. Designed for Isolation.

A developer platform targeting sandboxed security vectors. We bridge the gap between heavy agent tasks and zero-trust local constraints.

Compute Engine Sandboxing

Dynamically provision high-performance Linux VMs on demand. Keep execution nodes cleanly separated and isolated from your primary operating system.

Bi-Directional Volume Mounts

Mount single project folders directly inside the remote container or VM. The agent edits files natively while rest of your system remains completely secure.

Smart Inactivity Reaper

Keep cloud budgets tight. The background coordinator tracks agent task states and automatically suspends or terminates inactive VM nodes after 10 minutes.

Vercel EVE Loop Orchestration

Integrate multi-turn reasoning loops. Leverages smart `stopWhen` rules to stop tool execution loops as soon as target objectives are satisfied.

Live Streamed Server-Sent Logs

Watch your agent think, debug, and write files in real-time. Features structured visual dropdown logs for inner command calls and API evaluations.

Postgres State Persistence

All interactions, session states, sandboxing configurations, and agent chats are fully logged and safely tracked via our structured database layer.

System Topology

Zero-Trust Architecture Diagram

GCP Computer provides clear, decoupled routing between client actions, EVE core loop execution, and actual isolated host shells.

1

Frontend client sends task query

Secure SSE streams update the dashboard container in real-time.

2

EVE Core initiates multi-turn tool resolution

Evaluates code edits and CLI tools using Gemini/Vertex API.

3

Sandbox driver executes command inside safe container

Runs on either a local Docker instance or a VM on Google Compute Engine.

NEXT.JS DASHBOARD
UI Console State
SSE Stream
EVE CORE ORCHESTRATOR
Gemini Multi-Turn AI SDK
Tool-Call
SANDBOX ENGINE
Dedicated Runtime Driver
Isolated
DOCKER DRIVER

Alpine containers

GCP DRIVER

Compute Engine VMs

User Success

Life with GCP Computer

See how AI engineers, developers, and founders configure their secure coding agents.

Anthea

Autonomous Agent Architect

Docker alpine-node driver
I used to run coding agents locally on my machine, which felt like letting a stranger run arbitrary CLI commands on my host system. Spawning isolated sandboxes on GCP Computer in milliseconds keeps my code clean, protected, and fully observable.
Verified Deployment Case12 local tools replaced

Get Started in Emulation Mode

No GCP API keys ready yet? Run the app in offline emulator mode. Credentials authentication and memory emulation work out-of-the-box.

.env config
APP_MODE=local-emulated
NEXTAUTH_SECRET=a-secure-random-secret
NEXTAUTH_URL=http://localhost:3000