Operations demo · live AI
Paste in messy meeting notes and get back a clean table — who owns what, and by when. Unlike the other tools here, this one can't run on rules alone: it calls Claude directly from your browser to actually read the notes. Built by directing an AI assistant in plain English — one of a few small tools at nateeisenstein.com.
Real, slightly messy notes from a facilities vendor review, and what a good extraction pulls out of them. This result is fixed — it's here so you can see the shape of the output before deciding whether to try it live.
Meeting summary
Assignments
| Owner | Task | Due | From the notes |
|---|
Ready to send — by person
This section makes a real API call to Claude using the key you provide below. Nothing is sent anywhere else, and your key isn't saved past this browser tab.
No key entered — the notes below won't be sent anywhere until you add one.
The SLA tracker, coverage planner, supply tracker, and CSV helper are all rules running
on structured data — I wrote the logic, your browser runs it, no AI involved once it's
built. Reading someone's actual meeting notes and figuring out who owns what isn't a rule
a script can follow; it needs real language understanding at the moment you use it. So
this page calls Claude's API directly from your browser, using
anthropic-dangerous-direct-browser-access — a header Anthropic added
specifically so a client-only page like this one can call the API without a server in
between.
This page is public. If it called the API with a key I control, anyone could run unlimited requests on my account. So each visitor brings their own key, which is sent straight to Anthropic and kept only in this browser tab's memory — never logged, stored on a server, or sent anywhere else. A real product solves this the other way: the key lives on a server you don't expose, and the app calls your own backend instead of the API directly.
The notes go to Claude with a fixed instruction — find every concrete commitment, name the owner if one's stated, resolve any date phrase against the meeting date you give it, and quote the source sentence — using Claude's structured tool-calling so the response comes back as clean data (owner / task / due / source) instead of prose I'd have to parse myself.