The Free AI Business Stack: Gemini Credits, Local Tools, and Client-First Delivery

Albert Olgaard's video is framed as "Never Pay for AI Ever Again." I would make that claim more practical: you can remove a lot of startup cost if you understand free credits, local tools, free tiers, and client-first delivery.

The useful idea is not that software costs disappear. The useful idea is that a beginner can validate a simple AI service business before stacking up $20, $100, and $200 monthly subscriptions.

Source Note

Credit for the walkthrough goes to Albert Olgaard. The Mac and Windows setup details come from the attached guides and the public Google Docs versions: Gemini Credits Mac and Gemini Credits Windows.

I checked the official Google Cloud, Gemini CLI, Vercel, OpenWhispr, Cap, Stripe, and Google Maps policy sources where they matter. This is not legal, tax, billing, or business advice. Treat it as a practical setup and offer-design note.

Link Map

Here is the stack from the video, organized by job.

The Honest Take

The strongest lesson in the video is not "free tools make you rich." It is "do not buy a tool stack before you have a service people want."

That is good advice. A beginner does not need five subscriptions to test whether local businesses want a better website, a faster landing page, a simple booking flow, a better contact form, or a clearer service page.

But the title needs a caveat. You can avoid paid subscriptions while validating. You cannot run a serious client business forever by pretending every cost is someone else's problem. Eventually you need billing hygiene, client-owned infrastructure, a maintenance plan, support time, taxes, payment fees, and clear terms.

Gemini Credits Setup

The technical trick in the guides is routing Gemini usage through Google Cloud / Vertex AI so it draws from Google Cloud trial credits. The key pieces are:

• a Google Cloud account with active trial credits,
• a project linked to billing,
• Google Cloud CLI installed,
• Application Default Credentials configured,
• Vertex AI API enabled,
• environment variables set for the project, location, and Vertex AI usage,
• Gemini CLI opened with Vertex AI authentication.

The important environment variables from the guides are:

GOOGLE_CLOUD_PROJECT=YOUR_PROJECT_ID
GOOGLE_CLOUD_LOCATION=global
GOOGLE_GENAI_USE_VERTEXAI=True

The attached guides also include a raw Vertex AI test call using gcloud auth print-access-token. That is useful because it proves the chain works before you start relying on a coding CLI.

Mac and Windows Guide

Here is the cleaned-up version of the Mac and Windows flow from the supplied guides.

Windows-specific note from the guide: use curl.exe in PowerShell for the raw API test. Bare curl can be an alias for Invoke-WebRequest, which handles flags differently.

Finding Clients

Albert's business workflow is straightforward:

1. Find local businesses with weak websites or no website.
2. Prioritize businesses with proof of demand, such as good reviews and active operations.
3. Take a screenshot or note one specific issue.
4. Send a low-pressure message offering a free redesign preview.
5. Do not quote price before showing value.
6. Get the interested business on a call.
7. Show the demo live, explain the business value, then quote.

That is a stronger beginner offer than "I build AI automations." A plumber, barber, cleaner, dentist, or local service business does not need to understand your tool stack. They need to understand that their current website is losing trust, missing testimonials, loading slowly, or hiding contact information.

One warning: the video demonstrates scraping Google Maps results. Google Maps Platform terms restrict scraping, exporting, and saving Maps content outside the services. For a safer version of the workflow, use your own network, referrals, business websites, public directories with permissive terms, chamber of commerce lists, manual research, or permissioned lead sources. If you do use scraping tools, review the source terms, outreach law, and privacy rules in your region.

Delivery Stack

The delivery example is a local-business website redesign. The AI stack does the first build, but the business value still comes from judgment:

• clear service headings,
• visible phone and booking actions,
• trust signals and testimonials,
• fast loading,
• mobile layout,
• local SEO basics,
• forms that actually send somewhere,
• client-owned domain and hosting access.

Albert uses Gemini credits and a website-building skill to generate a polished demo quickly. That is fine for the first version. The client deliverable still needs review: copy accuracy, accessibility, contact details, form handling, analytics, legal pages, performance, and deployment ownership.

For hosting, the video uses Vercel. That is a good developer-friendly option, but check plan rules. Vercel's Hobby tier is free, but its docs describe it as personal/non-commercial. If this is commercial client work, a Pro plan or client-owned paid workspace may be the cleaner setup.

Free Tool Layer

The useful bonus tools from the video are:

• OpenWhispr: local voice-to-text for dictation, prompting, call notes, and writing faster without sending every word to a paid service.
• Cap: open-source screen recording for client demos, quick walkthroughs, bug reports, and async updates.
• Google Sheets: simple CRM for a first outreach list.
• Google Meet: sales call and demo surface.
• Stripe Startups: possible fee credits and startup benefits, depending on eligibility.

These tools do not make the service free forever. They make the first few reps cheaper, which is exactly what a beginner needs.

Risks and Guardrails

A free-stack post needs sharper guardrails than a normal tool list.

• Billing: set Google Cloud budgets and alerts before running anything expensive.
• Client ownership: do not build client infrastructure inside your personal account unless both sides understand the arrangement.
• Credit farming: do not create or encourage fake accounts to multiply credits. A client-owned project is different from abusing trials.
• Lead sourcing: check platform terms before scraping. Google Maps terms restrict extracting and saving Maps content.
• Outreach: follow CAN-SPAM, GDPR/UK GDPR, and local rules. Keep sender identity clear and honor opt-outs.
• Deployment: do not launch with fake testimonials, wrong service claims, broken forms, or missing contact paths.
• Maintenance: a website sale creates future support. Price or document that from the start.

JQ Checklist

Here is the version I would actually run.

1. Set up Gemini CLI through Vertex AI with a budget alert.
2. Pick one offer: local-business homepage redesign, not "anything AI."
3. Find 20 prospects through compliant sources.
4. Manually inspect each website and write one specific improvement note.
5. Send 5 to 10 human-reviewed messages per day.
6. Build only after interest, not before every cold lead.
7. Show the demo on a call.
8. Quote a clear fixed price and define what is included.
9. Deploy on client-owned infrastructure or a clearly agreed hosting setup.
10. Document the workflow so the second client takes half the time.

The real edge is not the free credit. The real edge is using the free credit to practice the full business loop: prospect, diagnose, demo, sell, deliver, and support.

Sources

• Albert Olgaard: Never Pay for AI Ever Again
• Albert Olgaard on YouTube
• Gemini Credits Mac guide
• Gemini Credits Windows guide
• Google Cloud Free
• Google Cloud free trial docs
• Google Cloud CLI install docs
• Gemini CLI authentication docs
• Google Gen AI SDK and Vertex AI docs
• Data Miner
• Google Maps Platform Terms
• Vercel pricing
• Vercel Hobby plan docs
• OpenWhispr
• OpenWhispr GitHub
• Cap
• Cap GitHub
• Stripe Startups
• Stripe Startups FAQ