PDFs to Agents.

Over 2000 hours working with ChatGPT. Not just code and projects — personal topics too. As a journal, as a coach, for processing, for everything. In the last year alone, 50,000 pages of PDF material accumulated. At some point it was clear: everything's in there. How I think, decide, communicate. Patterns I can't see myself. Values that reveal themselves over months.

But I can't do anything with those 50,000 pages. Can't feed them to any AI — way too much. If I just dump the PDFs into Gemini, ChatGPT, or Claude, processing quality tanks. Standard AI has a context window, and everything that doesn't fit is lost. The important information disappears in the noise.

So I built a system that processes, distills, and makes all that context portable.

Thousands of pages of PDF material.
Processed into what you actually need.

What it does.

PDFs to Agents takes exported chat logs — from ChatGPT, Claude, Gemini, any platform — and extracts structured personality profiles. 30 categories by default: communication style, decision logic, values, strengths, patterns, relationships, routines. Everything an AI needs to understand you.

The system works in two phases. First, a fast model extracts raw data from every conversation. Then a more precise model synthesizes everything into coherent profiles. 4,500 pages of PDF processed for about 10 dollars. Incremental — new PDFs just get added, nothing is processed twice.

At the end I have files I can load into any AI. ChatGPT, Claude, Gemini — doesn't matter. The profiles are platform-independent. My context travels with me.

Why I built it.

I work with AI as a thinking partner. Not as a tool I query and it answers — as a mirror that shows me what I'm missing. That only works with context. And building context takes months. Everything I put in there — the personal topics, the decisions, the patterns — that's not just a chat log. It's a documentation of my thinking.

Every time a platform changes, a chat gets too long, or a new model drops, that context is gone. Or worse: I don't notice it's missing, and the answers get more generic without me knowing why. There needs to be a way to actually work with this material.

PDFs to Agents solves that. Process once, use everywhere. And it doesn't just work for me — the categories are fully configurable. Sales call analysis, coaching profiles, research evaluation. The pipeline is the same.

Where it stands.

The system runs. Python, Claude API, everything local. I've been using it myself for months for my own personality profile. The code is open — if you want to build it yourself, everything's on GitHub.