The £0 AI Business Case Study: What Actually Works When You Build on Free Tools

Disclosure: This article promotes products from Rook. Links may point to paid products.

I spent the last six months trying to run a fully autonomous AI operation using only free-tier tools. No credit card swipes, no paid upgrades, no "free trial" that turns into a £50/month subscription the moment I forget to cancel. Just free APIs, free compute, and a lot of trial and error.

The result? A side project that now earns £150–£300/month in passive revenue with zero ongoing costs. The experiment is documented in The £0 AI Business, a £14 case study that breaks down exactly what worked, what didn’t, and why most "free" AI business advice is either misleading or incomplete.

If you’re an indie hacker, solopreneur, or developer wondering whether an autonomous AI operation on free tools is actually realistic, this is the breakdown you need.

The Myth of "Free" in AI Business

The internet is full of articles promising you can start an AI business for free. They usually follow this formula:

Pick a niche.
Use free APIs.
Build an agent.
Watch the money roll in.

The problem? They skip the part where you actually run the agent 24/7, handle failures, and deal with rate limits. Free tiers are designed for experimentation, not production. Most "free" AI businesses hit a wall when:

Their agent gets rate-limited and stops working.
They realize the free API only allows 100 requests/day.
They need to store data somewhere, and free tiers cap storage at 1GB.
They try to scale and hit hidden costs (e.g., free GPU hours run out).

I learned this the hard way. My first attempt used a free-tier LLM, a free vector database, and a free cloud function. It worked great—for about three days. Then the free LLM tier hit its request limit, and my entire pipeline collapsed. No warnings, no grace period. Just silence.

That’s when I decided to document the real experiment: What does it actually take to run an autonomous AI operation on free tools, long-term?

The Stack That Actually Worked

After three failed attempts, I landed on a stack that has now run for six months without a single paid upgrade. Here’s what it includes:

1. Free LLM: Mistral 7B via Hugging Face

Why? Most free LLMs are either too small (3B parameters) or too restricted (e.g., Llama 2’s commercial license). Mistral 7B hits a sweet spot: powerful enough for most tasks, and Hugging Face’s free inference API doesn’t throttle aggressively.
Cost: £0. The only limitation is queue times during peak hours.

2. Free Vector Database: Qdrant Cloud (Free Tier)

Why? Pinecone and Weaviate have free tiers, but they’re either too small (10K vectors) or require a credit card. Qdrant’s free tier gives 1M vectors and no credit card needed.
Cost: £0. The catch? You have to remember to clean up old collections manually.

3. Free Compute: Fly.io (Free Tier)

Why? Most free cloud providers (e.g., Render, Railway) give you a free VM for 30 days, then cut it off. Fly.io’s free tier gives you 3 shared-CPU VMs forever, as long as you don’t exceed 3GB RAM total.
Cost: £0. The trade-off is slower cold starts, but for a side project, it’s fine.

4. Free Scheduling: GitHub Actions

Why? Cron jobs on free tiers are unreliable. GitHub Actions gives you 2,000 free minutes/month—enough to run a lightweight agent every hour.
Cost: £0. The only limitation is that jobs can’t run longer than 6 hours.

5. Free Storage: Cloudflare R2 (Free Tier)

Why? AWS S3’s free tier is a trap (12 months, then £0.023/GB). Cloudflare R2 gives 10GB free forever, no credit card.
Cost: £0. The catch? You have to set up lifecycle rules to auto-delete old files.

6. Free Monitoring: UptimeRobot (Free Tier)

Why? If your agent stops working, you won’t know unless you check manually. UptimeRobot pings your endpoint every 5 minutes and emails you if it’s down.
Cost: £0. The free tier covers 50 monitors.