Tooling guide · Paris 2026

Build on Google Cloud, Gemini, and Antigravity.

Google Cloud, Gemini, and Antigravity 2.0 are the platform core for the hackathon. The dev tools your team already uses slot in around them.

On the day

Every team arrives to a pre-configured environment. The log-in flow is three steps:

  1. Sign in to console.cloud.google.com with the @gcplab.me credentials we send you in advance. Use a new Chrome profile or an incognito window so it doesn't collide with your own Google account.
  2. Select your assigned Google Cloud project (you'll have one).
  3. Start building.

There is no activation step and nothing to configure before you can build. Gemini API keys, Google Cloud billing, and your Workspace account (Gmail, Drive, Chat, Calendar, NotebookLM) are all live from the moment you sign in.

A 60-minute pre-event Enablement Session walks every team through the credentials, environments, and APIs you'll have on July 9 and 10. Sessions run on 19 and 22 June.

What Google provides

Every team gets a pre-configured Google Cloud environment with the full stack live from sign-in.

Google Cloud Platform

PROVIDEDGOOGLE

A pre-configured GCP project per team with temporary credentials and billing already in place. Cloud Run, Cloud Functions, Cloud Storage, Firestore, BigQuery, Compute Engine, GKE, Cloud SQL, AlloyDB, Spanner, and the rest of the Google Cloud stack are available from the start. Resources are torn down after the event.

Google Workspace

PROVIDEDGOOGLE

Each @gcplab.me account is a full Workspace account: Gmail, Drive, Chat, Calendar, and NotebookLM enabled by default. Use these for team coordination during the build without leaving the Google environment.

Gemini API

PROVIDEDGOOGLE

API credits with high quotas configured for the build day. Gemini is Google's multimodal AI model family and the primary model for the hackathon. Use it for text, image understanding, code generation, structured data extraction, and more.

Two quota tiers operate. The default tier is generous but capped; a high-throughput, no-throttling tier can be enabled at the project level for teams that need it. Capacity issues (HTTP 429s) are possible at peak; the fastest fix is to switch to a less-loaded model rather than retry the same one.

Beyond the core Gemini models, Google's media-specialist models are available: Nano Banana for image generation, Veo for video generation, and Lyria for music generation.

Google AI Studio

PROVIDEDGOOGLE

A browser-based playground for prototyping with Gemini. Test prompts, compare model outputs, and generate API code without writing any code first. Also includes a vibe-coding experience with an Export to Antigravity feature - useful as a starting point that hands off to a full agentic build environment.

Agent Platform (formerly Vertex AI)

PROVIDEDGOOGLE

Google Cloud's AI and agent platform. Provides managed access to Gemini, plus third-party models (Anthropic Claude, Meta Llama, Mistral, and others) through the Model Garden. Also offers Agent Studio, an Agent Registry, an Agent Garden, ADK runtimes, MCP servers, a governance Gateway, and memory management - the full agent infrastructure stack on Google Cloud.

Cloud Shell and Cloud Shell Editor

PROVIDEDGOOGLE

A browser-based terminal and IDE with the common tools pre-installed (gcloud, git, Docker, Python, Node, Gemini CLI). No local setup needed. Open the editor from the console toolbar or directly at ide.cloud.google.com.

Chromebooks

PROVIDEDGOOGLE

Available on request for teams without company laptops or for non-technical team members who want to skip local setup entirely.

Mentors and on-site support

Mentors and engineers circulate during the build to help teams get unstuck: answering questions, pointing you to the right service, and clearing blockers. They support the build, they don't co-build it, so the prototype stays your team's own work. The exact mix of on-site support is being finalised.

Antigravity 2.0 - the lead build surface

Antigravity 2.0 is Google's agentic development platform and the recommended starting point for teams of all skill levels at the hackathon. It's three surfaces sharing the same harness:

  • Antigravity Agent ManagerA command centre for orchestrating coding agents. Most teams will work here.
  • Antigravity IDEA companion VS Code-based editor for when you need to hand-edit code directly.
  • Antigravity CLIThe same capabilities from the command line for terminal-native workflows.

All three surfaces support @gcplab.me accounts under the Organization plan and come with easy-to-enable Google MCP servers out of the box. Download for macOS, Linux, or Windows from antigravity.google/download.

If you start in AI Studio's vibe-coding mode, you can export directly into Antigravity when you're ready for a fuller build environment.

AI models

Gemini is the primary AI model family for the hackathon. Other models are accessible through Agent Platform's Model Garden, so everything stays on Google Cloud infrastructure regardless of which model you choose.

Gemini 3.5 Flash

PROVIDEDGOOGLE

Generally available. Fast responses at lower cost. Strong default for speed-sensitive applications, multimodal tasks, and high-throughput agent loops.

Gemini 3.5 Pro

PROVIDEDGOOGLE

Coming soon - available for the hackathon. Google's most capable model for complex reasoning, long context, and multimodal work where output quality matters more than latency. Preview versions of Gemini may also be available on the day; no special setup or allow-listing is needed to use them.

Anthropic Claude

VIA AGENT PLATFORM

Claude models accessible through Agent Platform's Model Garden. Useful for long-form analysis or nuanced instruction-following while keeping billing and infrastructure on Google Cloud.

Open-source models

VIA AGENT PLATFORM

Llama, Mistral, and other open-source models are hosted in the Model Garden. Useful if your prototype needs to run on self-hosted infrastructure post-hackathon, or if you want to compare outputs across model families.

Build paths

Three routes to a working prototype in 24 hours. Pick the one that matches your team's makeup. The judging criteria are the same regardless: a working build, a problem worth solving, and AI doing real work.

Path A: Agentic build with Antigravity 2.0

RECOMMENDED

The default path. See section 02. Antigravity Agent Manager orchestrates the build; the IDE and CLI are there when you need them. Works for engineering-heavy and engineering-light teams.

Path B: No-code and low-code platforms

For teams without engineers, or for teams that want a polished front-end fast. These platforms produce real, working applications - you're not at a disadvantage by using them.

Google AppSheet

GOOGLEFREE TIER

Google's own no-code builder, integrated with Workspace and Gemini. Connects to Sheets and Cloud SQL. Solid for internal tools, forms, and approval workflows when the data lives in a spreadsheet.

Base44

RECOMMENDEDFREE TIER

Describe what you want in plain language; Base44 generates a working web application with database, UI, and logic. Fast from zero to a working demo.

Lovable

RECOMMENDEDFREE TIER

AI-powered app builder that generates full-stack web applications from natural language. Clean, production-quality code. Supports databases, auth, and deployment.

Bolt.new

RECOMMENDEDFREE TIER

Browser-based AI development environment. Describe your app and Bolt generates and runs it in a full-stack sandbox. Iterate by chatting with the AI.

v0 by Vercel

RECOMMENDEDFREE TIER

Generates UI components and pages from text or image descriptions. Best paired with another tool: build your interface in v0, wire it up to a backend elsewhere.

Glide

FREE TIER

Turns Google Sheets or Excel data into mobile and web apps. Strong for spreadsheet-rooted problems.

Path C: Code with engineers

Bring the IDE and toolchain your team already knows.

Visual Studio Code

BRING YOUR OWN

The most widely used editor. The Cloud Code extension connects it directly to your GCP project.

Cursor

BRING YOUR OWNFREE TIER

AI-native editor built on VS Code. Conversational code generation across your codebase.

Replit

RECOMMENDEDFREE TIER

Browser-based IDE with built-in AI assistance, deployment, and live collaboration. No local setup; good middle ground between full-code and no-code.

Frameworks worth knowing for a 24-hour build

Flutter

GOOGLE

Google's cross-platform UI framework. Web, mobile, and desktop from a single codebase. The right choice if your prototype needs to feel like a native mobile app.

Streamlit

RECOMMENDEDFREE

Turns Python scripts into shareable web apps in minutes. Ideal for data-centric prototypes, dashboards, and ML model demos.

Gradio

RECOMMENDEDFREE

Web UIs for ML models in a few lines of Python. Upload an image, get a prediction. Paste text, get a classification. The fastest path for 'AI model plus interface' prototypes.

Next.js / React

BRING YOUR OWN

If your engineering team already builds with React, keep using it. Next.js deploys easily to Cloud Run.

Other AI coding assistants

If your team already pays for one of these and prefers it to Antigravity for your stack, that's fine - they all work alongside the Google environment.

Gemini Code Assist

PROVIDEDGOOGLE

Google's AI coding assistant, available in Cloud Shell, VS Code, and JetBrains. Pre-configured as part of your GCP project.

GitHub Copilot

BRING YOUR OWN

Bring your own licenses.

Claude Code

BRING YOUR OWN

Anthropic's terminal-based agentic coding tool. Useful for teams with complex codebases.

Data, backend, and deployment

Where to store data, run backend logic, and deploy your prototype. Google Cloud services are pre-configured in your project.

Databases and storage

Firestore

PROVIDEDGOOGLE

NoSQL document database with real-time sync. Good default choice for most hackathon prototypes. Minimal setup, works well with Firebase Auth and Hosting.

BigQuery

PROVIDEDGOOGLE

Google's data warehouse. If your prototype involves analysing large volumes of data or running SQL queries at scale, BigQuery handles it without infrastructure work.

Cloud Storage

PROVIDEDGOOGLE

Object storage for files, images, documents, and unstructured data. Pair with Gemini for document processing or image analysis workflows.

Cloud SQL

PROVIDEDGOOGLE

Managed relational databases - MySQL, PostgreSQL, and SQL Server. Good fit when you want a familiar SQL stack with no infrastructure overhead.

AlloyDB

PROVIDEDGOOGLE

PostgreSQL-compatible relational database with better performance characteristics than vanilla Postgres. Worth considering for heavier transactional workloads.

Spanner

PROVIDEDGOOGLE

Globally distributed relational database. Overkill for most 24-hour builds, but available if your prototype needs it.

Compute and deployment

Cloud Run

PROVIDEDGOOGLE

Deploy any containerised application with a single command. Scales automatically including to zero. The recommended way to deploy your prototype's backend or full application on GCP.

Cloud Run now supports GPUs in a serverless paradigm - useful if your prototype needs GPU inference without provisioning a full VM. See docs.cloud.google.com/run/docs/configuring/services/gpu.

Cloud Functions

PROVIDEDGOOGLE

Serverless functions triggered by HTTP requests or events. Good for simple APIs, webhooks, and background processing. Technically a subset of Cloud Run, but kept as a separate top-level option because it's the cleanest fit for many small backend needs.

Google Kubernetes Engine (GKE)

PROVIDEDGOOGLE

Managed Kubernetes, with Autopilot mode for hands-off cluster management. Worth considering if your team is comfortable with Kubernetes or if you need advanced GPU or TPU infrastructure for the build.

Firebase Hosting

PROVIDEDGOOGLE

Fast, secure hosting for web applications. One command to deploy. Pairs with Firestore and Firebase Auth for a complete application stack with minimal configuration.

Compute Engine

PROVIDEDGOOGLE

Standard VMs when you need them - generally not necessary for a hackathon build, but available.

Alternatives you can bring

Supabase

BRING YOUR OWNFREE TIER

Open-source alternative to Firebase with PostgreSQL, auth, storage, and real-time subscriptions. If your team prefers SQL over NoSQL and already uses it, fine - but Cloud SQL or AlloyDB are the equivalent native options.

Vercel

BRING YOUR OWNFREE TIER

If your team already deploys front-ends on Vercel, that works. Keep the backend and AI model calls on Google Cloud.

Synthetic and test data

Gemini for synthetic data

PROVIDEDGOOGLE

Gemini can generate synthetic datasets from a description of your schema and business domain. Prompt it with your table structure and it produces realistic sample data. No additional tool needed.

Mockaroo

RECOMMENDEDFREE TIER

Generates realistic synthetic data in CSV, JSON, SQL, and other formats. Up to 1,000 rows free.

Design, demo, and presentation

The goal is a working build, not a slide deck - but a clear demo helps the judges see what you built and why it matters.

Design and prototyping

Figma

BRING YOUR OWNFREE TIER

Collaborative UI design. Useful for wireframing the prototype before building, and for polished screens.

Canva

BRING YOUR OWNFREE TIER

Quick graphics, mockups, and simple slide decks for non-designers.

Presentations and demos

Google Slides

GOOGLEFREE

For structuring the demo narrative. Frame the problem briefly, then switch to a live demo. Judges want to see it run.

Gamma

RECOMMENDEDFREE TIER

AI-powered presentation builder. Describe your content, Gamma generates a polished deck. Good for teams that want to spend build time on the prototype and assemble the slide layer quickly at the end.

Loom

BRING YOUR OWNFREE TIER

Screen recording with narration. If your demo involves a complex workflow, a recorded walkthrough can serve as a backup for the live run.

Collaboration and team logistics

Twenty-four hours goes fast. These tools help your team stay aligned and avoid the classic hackathon failure mode of building three things that don't connect.

Google Workspace

PROVIDEDGOOGLE

Docs, Sheets, Chat, Drive, and NotebookLM are all enabled on your @gcplab.me account from sign-in. Use them for running notes, decision logs, task tracking, and team comms without leaving the Google environment.

Bring your own

  • Miro / FigJamFREE TIERVisual whiteboarding for the first hour of scoping: map the problem, define the prototype boundary, assign workstreams.
  • NotionFREE TIERDocs, databases, and lightweight project management.
  • Slack / DiscordWhatever messaging tool your team already has.
  • GitHub / GitLabFREE TIERSource control. Create a dedicated repo for the hackathon project so the work is clean and portable when you leave.

Bring your own stack

These work fine alongside Google Cloud

  • Your IDEVS Code, Cursor, JetBrains, Vim, Emacs, Xcode, or whatever you code in. No restrictions.
  • Your programming languagePython, JavaScript/TypeScript, Go, Java, Dart, Ruby, anything else. All run on GCP.
  • Source controlGitHub, GitLab, Bitbucket.
  • AI coding assistantsGitHub Copilot, Claude Code, Cursor, Codeium.
  • No-code platformsBase44, Lovable, Bolt.new, v0, Glide, Bubble, Retool.
  • Design toolsFigma, Canva, Sketch, Adobe XD.
  • CollaborationSlack, Discord, Notion, Miro, Trello, Linear, Asana.
  • Docker and containersDeploy them to Cloud Run on GCP.

Avoid as your primary platform

  • AWS or Azure as your primary cloudKeep your cloud workload on Google Cloud. If you have an existing service running elsewhere that you need to call, that's fine - but don't set up new infrastructure on competing clouds for the hackathon build.
  • OpenAI or Anthropic APIs directlyIf you want to use GPT or Claude models, access them through Agent Platform so everything runs on GCP infrastructure and billing.

Important practices

  • Copy anything you want to keep before the end of Friday - the environment is decommissioned and there's no migration path off it.
  • Don't leak API keys. Treat your @gcplab.me credentials as you would any production credential.
  • Secure all open ports.
  • If you hit capacity issues (429s) on a Gemini model, switch models rather than retry.

Tooling questions

Reach the Eurazeo team for any tooling questions ahead of the hackathon.

Contact Eurazeo