chatviz
✦ analyze

Natural-language analysis

Ask a question in plain language → an open-weight LLM maps it to a vetted analysis plan → the plan runs on the full corpus or exports as a script. The model never writes SQL. Follow-ups refine the plan, up to 5 rounds per analysis.
analyze maps the question to a vetted plan · search looks the words up in the conversations
examples:
·
·

Paper demos (ShareChat, arXiv:2512.17843 — live previews; scripted walkthroughs, no LLM needed)

Source links

ShareChat keeps each message's outbound links — the paper's SourceLink field — so citation behaviour is directly measurable.

grok 25.2%
chatgpt 0%
claude 0%
gemini 0%
% of conversations with links/citations — “What share of conversations include links or citations, per …”
read all 5 steps on one page →

Timestamps

Per-message timestamps — the paper's Timestamp field — give the corpus a time axis for usage trends and assistant latency.

2024-01 2024-12
conversations — “How many conversations are there per month?”
read all 4 steps on one page →

Conversation completeness

Does a dialogue end resolved, or on a dangling user message? A measurable proxy for the paper's completeness analysis of multi-turn chat.

98.6%
% of conversations ending with an assistant reply — “What share of conversations end with an assistant reply?”
read all 5 steps on one page →