The Indicator guide to vibecoding for OSINT

For 20 years, Eitan Livne amassed a collection of regional Arabic dialects, including Egyptian, Moroccan, and Iraqi.

His dream was to start a company that could train on the corpus of writing to build an Arabic dialect detector.

"I wanted that to be my first startup, my first company: searching by Arabic dialect," he said. "We call them dialects but the difference between let's say Moroccan and Iraqi is bigger than English and German or Dutch and German.”

He eventually decided the VC-funded startup life wasn’t for him. Instead, he started a company that trains people on OSINT techniques and provides intelligence and digital exposure analysis. But Livne kept collecting Arabic texts. Over the years, he licensed the material to companies that wanted to train on the data. He one day hoped to use the corpus for himself.

Thanks to vibecoding, that day arrived last year. Livne loaded the linguistic examples into Base44, a vibecoding app, and told the product what he wanted to create. Soon he had a working prototype. After a bit of tweaking he released his tool, the Arabic Dialect Identifier.

Twenty years of collecting and planning came to life in a matter of days — no coding knowledge required.

"It's amazing, really," Livne told me.

Vibecoding is a cliché, but it describes a real phenomenon with remarkable potential for investigators. Using a dedicated platform like Base44 or Lovable, or a chatbot like Claude or ChatGPT (or their dedicated coding variants), someone with no coding, design, or product expertise can build a tool that speeds up their workflow, gathers useful information, visualizes data, performs OSINT tasks, or more.

"If you can describe it, you can make it,” said Henk van Ess, a longtime OSINT trainer who has vibecoded several useful tools and leads multi-day workshops for newsrooms on how to use AI.

In the OSINT community, vibecoding has led to a flowering of new tools. I looked at one slice of the output in my recent guide to 38 "situation monitoring” dashboards, many of which were created by people with no coding skills or knowledge.

Building tools for public release is just one use case, and it comes with risk and expectations. Most investigators will get more out of building something that lives on their own computer (or in their AI account) and handles specific tasks and workflows. I'd also argue you're vibecoding when you create a custom GPT or even the Skills for Chrome I outlined a couple of months ago.

To help investigators experiment with vibecoding, I spoke to five people in the OSINT and journalism community who have vibecoded their own tools. None of them are software engineers. Two describe themselves as having zero technical background. Some, like Livne and van Ess, built public facing tools. Others, like Omer Benjakob, an investigative reporter with Haaretz, built more than a dozen tools that never leave his laptop.

I think anyone could benefit from using AI to create personalized tools and processes. This guide offers advice on the easiest ways to get started, how to scale up, cautions about security and privacy, and firsthand accounts of how people are building. I’m grateful to Benjakob, Jeremy Caplan, Livne, van Ess, and Sarah Womer for taking the time to talk.