Available now

Run the pipeline on your own ChatGPT or Claude export, locally.

DOL_Analyser is a standalone Streamlit app that ingests your conversations.json export and generates a self-contained HTML report — the same structure shown below. Open source, MIT-licensed, fully operational.

Your data never leaves your device. No accounts, no signup, no API key required. Python 3.10+ on Windows, macOS, or Linux.

Download DOL_Analyser → Full README & instructions

§ 01 Ten-minute path

Four steps from conversations.json to a personal report.

The full instructions live in the DOL_Analyser README. The four-step summary below covers what most users need.

01

Export your conversations

ChatGPT: Settings → Data Controls → Export. Claude: Settings → Privacy → Export. Both produce a conversations.json file (sometimes inside a ZIP) that DOL_Analyser ingests directly. Either source works.

Clone and install

git clone https://github.com/RayanBVasse/DOL_Analyser.git
cd DOL_Analyser

# Windows:
install.bat

# macOS / Linux:
chmod +x install.sh
./install.sh

One-shot installer creates a local virtualenv and pulls dependencies. Nothing global, nothing to clean up later.

03
Run the app
```
# Windows:
run.bat

# macOS / Linux:
./run.sh
```
The Streamlit app opens at http://localhost:8501. Drop your conversations.json into the upload form. Processing runs entirely on your machine — no API calls, no telemetry, no network egress.
04

Read your report

A self-contained HTML report (the same format shown below) is generated locally. All figures are embedded as base64; you can save it, email it, or open it in any browser. Close the app and your data evaporates with the process.

§ 02 What you'll get

A real sample output.

Below: an actual DOL_Analyser report generated from one user's ChatGPT collaboration (8,945 messages across May–December 2025). Thirteen figures across activity, cognitive language, vocabulary specialisation, topic dynamics, dyadic alignment, and rolling entropy. Yours will look like this — but it will be yours. Scroll the panel to read.

Live sample · DOL_Report.html

§ 03 For research replication

Adapt, extend, replicate.

Bring your own corpus

The DOL research pipeline (separate repo: github.com/RayanBVasse/DOL) accepts JSON-lines corpora per the schema in data/README.md. One participant per file, one message per line, ISO timestamps.

Modify the lexicons

The structural-thinking and epistemic-uncertainty lexicons live in lexicons/ as plain-text term lists. Add language-specific terms, replace the construct, or substitute a published lexicon (LIWC categories, MFD, NRC). The analytical machinery is lexicon-agnostic.

Replace any layer

Elicitation, analysis, and synthesis are decoupled. Use only DOL on your own elicited corpus. Use only the Prometheus protocol with a different analytical backend. The three-layer separation is enforced at the interface boundary.

Open an issue

The point of an open instrument is that the failure modes are visible. We’d rather hear about a divergent result than not. Issues, pull requests, and questions all welcome.

Open source. Local-only. Run it on your own machine in ten minutes.

Open DOL_Analyser repository File an issue