Ask how to do anything
“Where is the filter button?” “How do I set a reminder before calibration is due?” Answers name the exact buttons and menus, from the product's own guide.
Ask how to do anything in Qualsted, ask about your own tables, get a chart in seconds, and catch up on a long deviation without reading every comment. It works only with what you're allowed to see, suggests instead of deciding, and every use is logged for your admins.
Chart open CAPAs by owner, and are there any sorts on the CAPA Log views?
Here are the open CAPAs by owner. The Open CAPAs view sorts by Due (earliest first); Board has no sort.
Written by AI: check it before relying on it.
Five things people ask for every day, built into the places they already work.
“Where is the filter button?” “How do I set a reminder before calibration is due?” Answers name the exact buttons and menus, from the product's own guide.
“Which views of the CAPA Log have a filter?” “What automations run on Deviations?” It looks up the columns, views, sorts and automations of the tables you can see.
“A bar chart of deviations by area.” “Total hours by status.” The chart appears in the conversation with live numbers; add it to a page as a dashboard.
A deviation with forty comments and a long history? Get what happened, where it stands and what's still open in a few lines. Nothing is saved to the record.
Type “open CAPAs due this month owned by me” and the filter fills itself in, with real dates and options, for you to check and apply.
Say what the table will track (“supplier audits: who, when, outcome, follow-ups”) and tick the suggested columns, with sensible types and status stages.
Open a record and choose Summarize. The assistant reads its fields, comments and history, the same ones you can see, and writes a short brief for a colleague who hasn't read it.
Open points: stability results for 5 boxes; root cause sign-off.
Written by AI: check it against the record before relying on it.
The draft EU GMP Annex 22 keeps generative AI out of GMP-critical decisions. Qualsted's assistant was designed that way from the start.
| Concern | How the assistant handles it |
|---|---|
| Decisions | It never decides. It doesn't approve, sign, reject, close records or assign root causes. People do, with their own signature. |
| Changes | It never changes anything on its own. Filters, columns and charts are applied by the person, through the same checked, audited actions as doing it by hand. |
| Access | It sees only what you can see. Every lookup goes through your own permissions; tables and records you can't open don't exist for it either. |
| Accountability | Every use is logged. Admins see who used it, for what, when, its size, and whether the result was used. The text itself isn't stored. |
| Your data | Not used to train models. Content goes to Anthropic, our listed AI sub-processor, only when someone with a seat asks a question. |
| Who can use it | Only people you give a seat. Admins assign seats; without one, the assistant doesn't appear. Viewers and API keys can't use it. |
The assistant is an add-on to either plan. Buy as many seats as you need, say 4 or 5, and give them to people in Settings → People. Move a seat from one person to another at any time.
$15per user / month, billed annually
Help and questions about your tables, charts, record summaries, filters from words and suggested columns.
$0the Help guide is free for everyone
Every workspace has a searchable help guide. Companies whose policies don't allow AI simply don't buy seats: nothing is sent anywhere.
The assistant only reads and suggests; it can't change, sign or approve anything. Anything you apply goes through the normal actions, with the same validation, permissions and audit trail. That keeps it out of GMP-critical decisions, as the draft Annex 22 expects. Treat its answers like a colleague's notes: helpful, and checked before relying on them.
No. Record content is sent to Anthropic, our AI sub-processor, only when someone with a seat asks a question, and it isn't used to train models.
No. Every lookup runs with your own permissions, so it sees exactly what you can see and nothing more.
Admins have a log of every use (who, which feature, when, size, and whether the result was used) in Settings → Workspace → AI assistant. The text itself isn't kept.
Not today, on purpose. It summarizes what's already in a record and helps you find and chart information, but the investigation and its conclusions stay with your people.
It's only on for people you give a seat. With no seats, nothing is sent and the assistant doesn't appear; the help guide is still there for everyone.
Ask for the assistant in your demo, and we'll try it on a table built from your own process.