Run it on your company, not ours.
Drop in a Slack export, a CSV, or a few hundred lines pasted straight out of a channel. VANAV reads it into the same shape the demo uses, shows you what it understood, and waits for you to say go before it derives anything.
Used to name the workspace.
The title, even if nobody has done it yet.
We do not look your company up. Without this, the derivation only knows what is in the corpus below — which works, but a sentence here makes the role noticeably more specific.
Drop a file here, or
Up to 6MB. Nothing leaves this server.
This step is free and instant — no model call. You will see exactly what was understood before anything is derived.
- Slack export
- The real JSON: an array of message objects with user / user_profile.real_name, text and a unix ts. The { "messages": [...] } wrapper and per-channel files both work.
- Pasted log
- Select a channel, copy, paste. #channel Name 10:32 message, or Name: message, or [10:32] Name: message — mixed together is fine.
- CSV
- A header row, then rows. Columns named author / user / name, text / message / body, date / ts / timestamp, and channel are picked up automatically.
Join and leave events, bot bookkeeping and duplicate messages are dropped. Anything the parser cannot read is reported as a warning rather than an error — a partial corpus you were told about is more useful than a failed upload. The corpus is capped at 1,500 messages or 200,000 characters: the whole thing goes to the model in one prompt, with no retrieval step, so size is literally cost. Over the cap we keep the most recent slice and say so.