← Resources

AI that works every day.
For less than a cent a month.

Any AI job that repeats, from research to summaries, from classification to reports, can run on its own on free models. A paid model stays only as fallback, and a check written in code blocks mistakes before they go out. On my system, models cost $0.0069 this month.

Get the free guideTwo ways ↓
Spent on models$0.0069 this monthTestedon a clean server, 29 September 2026ForClaude or ChatGPT

Same work, very different bills.

A real example: my system that every day reads the news, picks the best, writes and publishes. About 44 million tokens a week. On a paid model, for the same work, this is the bill.

if the model reuses text it already read (cache)if it does not

Free models: real spend. Paid models: estimate on the same token volume. Another model would use a different number of tokens.

Five pieces. None of them is a bigger model.

The trigger
The job starts on its own: at a set time, or when something new arrives, like an email, a file or a form.
↓
The filter
A free model scores each item on fixed yes/no questions. Only what deserves work reaches the real job, and tokens drop: in my case from 11 to 4 million per run.
↓
The job
Writing, summarising, classifying, extracting data, answering: a free model does it. If it does not finish, the second free model tries, then the paid fallback, then you get an alert.
↓
The check
A program, not an AI: it checks the result against your rules. Numbers against sources, format, length, required fields. If it fails, nothing goes out.
↓
The output
The result goes where it should, or an alert reaches your phone. Silence means it worked.

Two ways.

Install it with your AI agent, or have me do it on your job.

The kit

Do it yourself

$19

→The ready folder: open it with your agent (Claude or ChatGPT), say "install the kit" and it does the rest.
→The chain of free models with a paid fallback.
→The filter, the check, the alerts and the automatic start on your server.
→An example that works in 3 minutes and $0, and a sheet to calculate what you save.
→A 14-page complete guide, in English and Italian.
Buy the kit

The service

I do it for you

€290 €490

→I take one of your recurring AI jobs and move it to free models, with fallback and checks.
→I measure the cost before and after, on a week of real work.
→I hand it over working, with instructions.
→If you already bought the kit, its price comes off.
Tell me about your job

Launch price.

Before buying, read the free guide.

8 pages: the bill, the five pieces, what free models can and cannot do, the limits said plainly. Enough to know if it is for you.

Get the free guide

What you are probably thinking.

“Don't free models make more mistakes?”

Yes. That is why a job counts as done only when a check written in code passes, not when the model says it is done. A weaker model with a strict check beats a stronger one you trust blindly.

“What can I use it for?”

Any AI job that repeats and can be checked: summarising documents or news, sorting incoming requests, extracting data from text, preparing drafts and reports, answering questions on public material.

“Can I send it my clients' data?”

No. Free providers may log prompts. This system is for material you could publish. Private data needs the paid model, and the guide says so.

“Do I need to code?”

No. Your AI agent, Claude or ChatGPT, does it all following the instructions in the kit: install, test, adapt to your job. If you would rather not touch anything, there is the service.

“Does it work with my agent?”

Yes, with Claude or ChatGPT. The kit contains the instructions for the agent, which installs it and adapts it to your case.

Where it comes from.

It is the engine of Poltronave, a site that posts on X every day and on its blog every week without me writing a line. The kit is that system rewritten in a generic form, for any job, and tested from scratch on a clean server.

Deep dive: Claude Code free with OpenRouter: setup and real costs.