My AI Automation Costs $0 per Month - Here's How

Last year I was spending $200/month on API calls for various AI automations. CRM updates, email drafting, browser tasks, document summarization. Every single action was a round trip to an API endpoint with per-token billing. Then I switched to local models and an open source agent running entirely on my Mac.

Total monthly cost now: $0.

The Setup

The stack is simple. A local model running through Ollama handles the reasoning. An open source desktop agent handles the execution - clicking buttons, filling forms, navigating between apps. Everything runs on Apple Silicon, which turns out to be more than enough for 7B-13B parameter models.

For browser automation, the agent uses the accessibility tree rather than screenshots. For document tasks, it reads and writes files directly. For CRM updates, it navigates the web app the same way you would - click, type, submit.

The Trade-Off Is Real

Local models are not as capable as GPT-4 or Claude for complex reasoning. Multi-step planning with many dependencies can stumble. Nuanced writing tasks produce decent but not great output.

But here is the thing - 80% of daily automation tasks do not need frontier-model intelligence. Updating a CRM field, organizing files into folders, filling out a form, sending a templated email. These are mechanical tasks where a 7B model performs just fine.

Where It Breaks Down

Anything requiring deep analysis - summarizing a complex legal document, writing original marketing copy, debugging subtle code issues - still benefits from a larger model. For those tasks, I make occasional API calls. Maybe $10/month total.

The 80/20 Split

Run local for the mechanical 80%. Use API calls for the complex 20%. Your monthly bill drops from hundreds to nearly zero, and the automation quality stays high for the tasks that matter most - the repetitive ones eating your time.

Fazm is an open source macOS AI agent. Open source on GitHub.