Google AI Studio — Gemini-powered AI Web IDE + app builder

Google AI Studio (aistudio.google.com) is the in-browser home for the Gemini family. It started as a prompt playground, but since 2024 it has grown into a notebook + app builder + auto-deploy environment — squarely in the same AI Web IDE category as Replit.

1. Four faces in one tab

Face	What it is
Prompt playground	Single-shot Gemini API tests. Tweak system instruction, temperature, safety filters.
API key issuance	Free key from one Google account. Separate from Vertex AI (no org/billing required).
AI notebook (formerly Studio Library)	Auto-generates code for RAG, function calling, multi-turn chat.
Build (NL → app)	Prompt → Vite/Next app → auto-deploy to Cloud Run.

notes/ai/04-gemini-api covers the API side. This note focuses on the AI Web IDE and app builder faces.

2. Where AI Studio fits

Year	Event
2023-12	MakerSuite launches alongside Gemini Pro 1.0.
2024-02	Rebranded to Google AI Studio.
2024-05	Gemini 1.5 Pro/Flash · 1M-token context, free key.
2024-12	Multimodal Live API (image/audio/video).
2025+	Build mode — generate and deploy apps from prompts.
2025+	Stitch (UI design) · NotebookLM hooks.

Core value:

• All Gemini 1.5/2.0 models + 1M-token context, free (within rate limits).
• Auto-generates code (Python · Node.js · cURL · Swift · Kotlin · Dart).
• Auto-deploy (Cloud Run · Firebase Hosting).
• Tight Workspace · Drive · Maps integration.

3. AI Studio vs Vertex AI

Same Gemini models, two entry points:

Aspect	AI Studio	Vertex AI
Entry	Single Google account	GCP org · project · billing
URL	aistudio.google.com	GCP Console → Vertex AI
Billing	Free (rate limit) + paid key	GCP usage
Data policy	Free key may train models (opt-out available)	Enterprise — no training (default)
SDK	@google/genai	@google-cloud/vertexai
Auth	API key	OAuth · service account
Deploy hooks	Cloud Run · Firebase	Vertex AI Endpoints
Sweet spot	learning, side projects, PoCs	production, compliance

Common pattern: start with AI Studio key → graduate to Vertex AI. Same model IDs, so the migration is mostly an SDK swap.

4. Build mode — NL → app → deploy

aistudio.google.com/apps for Build mode.

"An app where I upload a photo and Gemini writes a description"
       ↓
1. Gemini picks the stack (Vite + React + Tailwind).
2. Generates file tree, installs deps automatically.
3. Auto-injects the Gemini API key as an env var.
4. Preview in the in-browser webview.
5. Click "Deploy to Cloud Run" → auto-deploys + URL.

What you get:

• Frontend — Vite/Next or static HTML.
• Backend — Cloud Run Functions (when needed) or direct client→Gemini calls.
• Auth — Firebase Auth (optional).
• DB — Firestore · Cloud SQL (optional).

GitHub auto-sync (Build can export to your repo). Best for learning · hackathons · PoCs.

5. AI notebook (Live API · function calling)

Use the GUI to try complex flows — RAG · tool use · multi-turn chat. Then click Get code to copy the equivalent Python/Node.js SDK code.

# Get code result (Python)
from google import genai

client = genai.Client(api_key="AIzaSy...")
response = client.models.generate_content(
    model="gemini-2.0-flash-exp",
    contents="Hello",
)
print(response.text)

Same flow for Node.js · Swift · Kotlin · Dart.

6. Free quota — actual limits

As of 2026-05 (subject to change):

Model	RPM (per min)	TPM (per min)	RPD (per day)
Gemini 2.0 Flash Exp	10	1M	1,500
Gemini 1.5 Pro	2	32K	50
Gemini 1.5 Flash	15	1M	1,500
Gemini 2.5 Pro / Flash	(varies)	(varies)	(varies)

Free-key inputs may be used for model training (opt-out available). For sensitive data, prefer Vertex AI.

Switching to a paid key (Pay-as-you-go) → no training + larger quotas.

7. AI Studio vs Replit / Bolt.new / v0.dev

Aspect	AI Studio Build	Replit Agent	Bolt.new	v0.dev
Model	Gemini only	Anthropic + opt	Anthropic	OpenAI · Anthropic
Backend	Cloud Run · Firebase	Reserved VM	partial (WebContainer)	✗
DB	Firestore · Cloud SQL	Replit DB · PG	✗	✗
Deploy	Cloud Run auto	Replit Reserved	Vercel / Netlify	Vercel
Free	Gemini quota	sleep + 1 Static	limited	✓
Strength	Google ecosystem · 1M-token context	learning · collab	quick SPA PoCs	UI components
Weakness	Google lock-in	costly when busy	thin backend	front-end only

Pick by:

• Already on GCP + 1M-token context → AI Studio Build.
• Learning / pair programming + collab → Replit.
• Quick Vite/Next PoCs → Bolt.new.
• shadcn/ui components → v0.dev.

8. Good fit / poor fit

Fits:

• Gemini experiments / PoCs.
• Combining Workspace · Drive · Maps · YouTube APIs.
• RAG over very long documents (1M-token context shines).
• Auto-generated learning apps.
• Function calling · Live API trials.

Poor fits:

• Strict compliance worries about training-on-input — use Vertex AI.
• Tasks where Anthropic/OpenAI models clearly win (varies by task).
• Above-quota traffic.
• Avoiding Google ecosystem lock-in.

9. Data / training policy (important)

AI Studio free key:

• Inputs/outputs may be used for model training (default on).
• Privacy controls let you opt out.
• 30-day retention (for debugging).

AI Studio paid key (Pay-as-you-go):

• No training (default).
• Short retention (24–48h).

Vertex AI:

• No training (default).
• Separate retention policy.

Use a paid key or Vertex AI for any work / sensitive data.

10. Further reading

• Google AI Studio
• AI Studio Build
• Gemini API note — the API side
• Replit — another AI Web IDE
• AI coding IDE comparison — desktop side
• AI Web IDE roundup — browser side