Web dashboard

Find and verify an assistant answer

Ask the assistant a question in plain words, check where its answer came from, and point it at a specific record.

Sound is off

Star Beam is the assistant built into the dashboard. It opens beside whatever you are working on, and it answers from your organisation's own content — not the open internet. That means knowledge base guides, form responses, client records, shifts and schedules, and library documents that have been indexed for chat.

It answers within your access. It cannot show you a client, a response or a shift you would not be allowed to open yourself, so two people can ask the same question and get different answers — and that is correct behaviour, not a fault.

The way you ask decides where it looks. The assistant picks a tool before it answers, and the phrasing steers that choice: "how do I clock in and out?" is read as a product question and answered from the walkthroughs, while "what does our clocking in and out guide say?" is read as a question about your own records and answered from the library. Both are right answers to different questions — but if you wanted your own policy and got a walkthrough, the fix is the wording, not the indexing.

Every answer carries its sources as pills in the text, each naming what it used. Opening one brings the source up beside the answer. This is the part worth the habit: if the source does not say it, the answer does not stand. An AI answer can be confidently wrong, and the source is how you tell.

Add sources narrows a question to particular records — a client, a person, a facility, a form's responses — so you can ask about one thing rather than everything. And full view gives the conversation the whole window when an answer needs the room.

Step 1
Open Star Beam
It opens beside the page you are on rather than taking you somewhere else, so you keep your place.
Step 2
Read what it offers
The suggested prompts follow the page you are on and what your role can reach — they are a fair guide to what it can do for you here.
Step 3
See what it can help with
The list underneath is the set of things it can actually answer from, rather than a general claim about AI.
Step 4
Ask in your own words
Plain language works — ask the question you would ask a colleague. Name the thing you want it to read if you know it: "what does our clocking in and out guide say…" sends it to your library, where "how do I clock in and out" sends it to the product walkthroughs.
Step 5
Where the answer comes from
Your organisation's own content: knowledge base guides, form responses, client records, shifts, and library documents indexed for chat.
Step 6
The pills are its sources
Each source the answer used appears as a pill in the text, labelled with what it is — a document's filename, a client's name, a guide's title. It is a control, not a footnote: it opens.
Step 7
Check it before you act on it
Opening a source brings it up beside the answer, so you can read the thing itself without losing the conversation — and reading it is the point: find the line the answer rests on and check it says what the answer says. If the source does not say it, the answer does not stand. An answer with no source at all is a summary of nothing you can check.
Step 8
Point it at one record
Add sources attaches particular records to the question. The categories are Clients, Users, Facilities and Form responses — records, not documents, so a library file is something the assistant may cite but not something you can scope a question to. Pick what you mean, press Apply, and ask again: the assistant can only answer from what you pointed it at, which is how you stop a broad question drifting into the rest of your data.
Step 9
An answer from one source
With a source attached, the answer can only come from what you pointed it at. That is worth knowing in both directions: it makes an answer easy to check, and it means anything outside the attachment is not being considered — so a scoped question is the wrong tool for "is there anything I've missed".
Step 10
Give it the whole window
Full view opens the same conversation in the whole window rather than the side drawer. Worth it when an answer is long, or when you are reading it against a source side by side. Nothing about the conversation changes — it is the same chat, with more room.
Step 11
When the answer cites nothing you expected
Work through it in this order. First the wording: a "how do I" question is answered from the product walkthroughs and will not reach your own documents however well indexed they are. Then whether the document is indexed for chat at all — that is a per-file switch in the Library, and an un-indexed document is invisible to the assistant no matter who can read it. Then your access: it reads only what you could read. "It doesn't know about our policy" is nearly always the first or the second of those.
Step 12
It answers within your access
The assistant reads what you could read. A client outside your permissions, a response you cannot open, a shift at a site you do not cover — none of those reach the answer. So "the assistant did not mention it" is never evidence that something is not recorded.
Step 13
What an AI answer is worth
It is a fast way to find the right record and a summary of what it says. It is not a clinical decision, a policy, or a record in itself. Anything that changes what you do for a person is worth reading in the source — which is exactly why the sources are one click away.