Wikigen — Demo
The standalone repository is private, so this is a walkthrough of the retained outputs rather than a runnable public tutorial.
Start with the graph
The main project page shows a static PNG made from 55 conversations and an animation that reveals concepts in conversation order. The final positions are computed from the complete graph first, then held still while nodes appear. That is why the animation shows the history instead of a layout constantly rearranging itself.
What the no-model mode shows
Wikigen also has a --no-api mode. It creates one node name per filename, so the
expected result is unconnected dots with zero edges. That is not the useful graph;
it is a quick way to check that transcript discovery, graph creation, and rendering
can run before adding concept extraction.
The graph becomes meaningful when different conversations can share a concept. Extraction provides those shared nodes, and the consolidation pass merges cautious near-duplicates before layout begins.
What a full run produces
A full run can create:
- a static PNG of the complete graph;
- a standalone interactive HTML page; or
- a GIF with concepts appearing in chronological order.
If one extraction fails, that conversation contributes nothing and the graph is smaller. If consolidation fails, the run continues with visible duplicate labels. Both failure modes degrade the output instead of discarding the whole run.
A privacy note
Generated concept labels still reflect the input conversations even when no transcript text is shown. I review outputs before sharing them and select only transcripts I intend to send to the provider for extraction.
See the retained outputs · Architecture · Technical notes