OpenAI

OpenAI powers exactly two things in PeopleMuster, and the product page says so in one line: HashBot policy search and AI meeting signals. Nothing else in the product calls a model.

The two features it powers

HashBot answers questions from your policy library. Publishing a policy chunks and indexes it, and the count of indexed chunks shows on the policy record. Someone asks how many sick days they get, and the assistant answers from the document rather than from a manager's memory.

HashBot carries a beta label, and it is clear with people about its role. It gives a fast first answer in plain words, and the policy document stays the official source for anything that matters.

Meeting signals are the second feature. Each imported meeting gets a risk read for leadership, with a verdict and detected flags such as no client present. Every signal is stamped with the prompt version and the model that produced it.

What it needs before it works

An OpenAI API key, entered once and masked after saving. Held in your own database, never shown again in full.

Published policies. Publishing is the step that indexes a policy and makes it answerable.

Meeting signals need Fireflies as well, because a meeting has to be imported before anything can read it.

One prompt exists today, called Meeting Signals. An admin can edit it, version it, and re-run it across recent meetings without a code change. Older signals stay stamped with the version that produced them, so a prompt change does not quietly rewrite history.

That last detail is unusual enough to be worth dwelling on. The prompt text sits on a page, with a version number and a backfill button next to it. A company that wants its meetings read differently changes the words.

Questions people ask

What can HashBot see?

Published policy documents only. Attendance, leave and benefits records sit outside its index.

Can we change how meetings are scored?

Yes. The Meeting Signals prompt is editable by an admin, is versioned, and can be re-run over recent meetings.

How should people use HashBot's answers?

As a fast way to find the right policy. The policy document stays the official source, so anything important is read there.