Bolt.new vs Bolt.diy ⚡ ¿Cuál es Mejor para Crear Apps con Inteligencia Artificial?

Summary of the Video (Bolt.new vs Bolt.diy / Bolt.diy DIY)

Purpose & overview

• The video compares Bolt.new (paid, cloud-based) with Bolt.diy (open-source/open alternatives) for building web and mobile apps using AI coding agents.
• It highlights key differences in setup and workflow, then runs practical tests by generating real apps/pages from prompts.

---

Bolt.new (paid, cloud)

Core product features

• Browser-based development: build directly in the browser—no local setup required.
• Uses containers to provide a complete development environment.
• Projects are generated from natural-language descriptions (optionally with image attachments as references).
• Workflow integrations mentioned:
  • Netlify for deployment
  • Supabase for databases/authentication

Pricing / limits (trial experience)

• The creator notes a Pro subscription cost of $20/month, including 10,000 tokens/month.
• Despite this, the test account starts with a very large free allocation:
  • 2,000,000 tokens/month, with a daily cap of 300,000 tokens (previously higher, per the video).
• Practical emphasis: iterative prompting can consume tokens quickly; the demo used a significant portion of the daily cap during translation/updates.

Demo: AI-generated portfolio site (photographer)

The agent generates a portfolio site that includes:

• Landing hero section with a strong background image
• Top navigation + section navigation
• About section with cards
• Portfolio section with functional filters
• Services / pricing section
• Testimonials carousel/swipeable cards
• Contact form (can email directly; optionally store leads in Supabase or other databases)
• Footer with links

Reported outcome:

• “Professional” result in under ~6 minutes
• Model mentioned: Bolt.new “generally uses Claude/Cloud Set 3.7” (as stated in subtitles)

Iteration test: translate UI to Spanish

• The agent can change existing output (translate page text into Spanish).
• Result: mostly translated, but some sections remained in English until additional prompting.
• Drawback highlighted: iterative changes consumed a lot of tokens.

Recommended workflow suggestion

• Use Bolt.new to generate an initial template, then download the project and finalize it with other tools (the video references an additional coding agent tool for final adjustments).

---

Bolt.diy / open-source alternative (local, self-hosted)

Core product features

• Open-source tool for AI coding using many model providers, including:
  • OpenAI, Anthropic, Gemini, DeepSeek, and more
• Key advantage: flexibility in choosing/configuring AI providers.
• Supports model execution:
  • Cloud-based (via API keys)
  • Locally using runtimes like Ollama or LM Studio

Installation options shown

1. GitHub/manual install (requires Node.js)
2. Docker install (isolated container)
3. Pinokio (simplified “app store”-style installer used in the demo)

Pinokio ecosystem setup

• Pinokio acts as an installer/manager:
  • Search/install Bolt locally
  • Provides a starter guide
  • Helps configure:
    • Local model runtime (Ollama / LM Studio)
    • Provider API keys for cloud model access
  • Includes update checks and provider configuration UI

Local model execution: hardware & model sizing

• Local models require enough GPU RAM. Examples mentioned:
  • High-end: e.g., RTX 4090 (16GB laptop / 24GB desktop) for better results
  • Smaller options for weaker machines (example given: 7B variants)
• Model examples referenced:
  • Qwen 2.5 Coder variants (7B / 14B / 32B depending on GPU)
  • Yen / “Yema 3” sized variants
• Local runtimes:
  • Ollama (command-line approach)
  • LM Studio (visual approach; recommended for beginners)

API key / cloud model options for the open-source setup

Getting keys from providers mentioned in the video:

• Google AI Studio
• OpenRouter
• Grok (free trials mentioned)
• DeepSeek (very low-cost access during off-peak hours; specific discount/price claims noted)

Also noted:

• Free-tier trials may include daily limits, requests-per-minute limits, and possible throttling, which can cause the agent to hit limits.

Demo results: open-source portfolio site (photographer)

• The same portfolio prompt was run locally/open-source.
• Generated a similar structure:
  • Hero image + navigation
  • Portfolio filters
  • Pricing/services section
  • Testimonials carousel
  • Contact form + footer
  • Responsive layout (mobile + desktop)
• A “reference error” occurred, but the agent fixed it automatically without a server restart.
• Overall quality described as very good for a first version.

Demo results: additional pages/apps

• Pastry shop site
  • Added: top menu, hero image, close-up menu section, special offers, gallery, testimonials, contact, footer
  • Claimed completion in under ~10 minutes
• Generic landing page
  • Added: CTA + registration form, features list, testimonials
  • Includes FAQ section with accordion UI
  • Footer
• React/TypeScript task app
  • Prompt: “add/remove tasks” in Spanish
  • Demonstrated adding and removing multiple tasks correctly
  • Completion claim: under ~5 minutes

Deployment mention with Netlify (open-source workflow)

• Bolt.diy can integrate with Netlify by entering a Netlify token, enabling deployment toward production readiness.

---

Key comparison takeaways (as presented)

• Bolt.new

  • Easiest setup (cloud, browser-only)
  • Strong output quality for quick “professional” results
  • Main limitation: token limits and iterative prompt cost
  • Integrations: Netlify + Supabase

• Bolt.diy (open source)

  • More setup effort (install locally)
  • Higher flexibility: choose providers and potentially run locally
  • Avoids Bolt.new’s daily/monthly token constraints by controlling local compute or API costs
  • Quality depends heavily on the model selected and available hardware

---

Main speakers / sources

• Speaker / channel host: Mauricio Perdomo (introduced as the narrator)
• Primary tools referenced:
  • Bolt.new
  • Bolt.diy (open-source counterpart)
  • Pinokio, GitHub, Docker
  • Ollama, LM Studio
  • Model providers mentioned: OpenAI, Anthropic, Gemini, DeepSeek, Google AI Studio, OpenRouter, Grok
  • Integrations: Netlify, Supabase