The Story
After a few months of running an AI agent 24/7, we had hundreds of conversation transcripts and no way to find anything. "What did we decide about the database?" required grep-ing through JSONL files. That's not sustainable.
So Craw built the Conversation Explorer β starting with a simple timeline (v1) and iterating toward increasingly sophisticated views. Each version tried a different approach to the same problem: how do you make sense of months of human-AI collaboration?
v2 added AI-generated summaries. v3 turned conversations into a network graph. v4 brought semantic search. v5 unified everything. v6 flipped the question: instead of "let me search for something," it asks "what should I know?" β surfacing trends, hot topics, and decisions automatically.
All six versions are live. They all read from the same conversation database β the same 200+ sessions that represent everything Dylan and Craw have built together since February 2026.