How to Share Hex Dashboards Without Giving Away Warehouse Access

November 13, 2026

Hex ships two ways to share work: a published app (anyone with the link, no login on some plans) and a shared project (workspace members, editable). Apps run queries with the credentials of the person configured as owner or with a service account, which is why you can give a stakeholder a filterable dashboard without a warehouse password.

Most teams get this wrong by sharing the project link and then discovering that half the company can edit the SQL. Hex makes the app view a separate artifact for exactly this reason.

What is the difference between an app and a project?

An app is a published view with selected cells and inputs. A project is the full notebook with code, cells and comments.

App Project
Viewer needs a Hex account Depends on plan Yes
Can see SQL No Yes
Can edit No Yes, with permissions
Updates when the project changes Yes, on republish Live

For executives and operations staff, publish the app. For the two other analysts who maintain the logic, share the project. Send both to everyone and you have created a maintenance problem disguised as convenience.

How do you parameterize an app so it is actually useful?

Add input cells and reference them in your SQL. Hex supports text, number, date range, dropdown and multi-select inputs.

A date range input called date_range turns into a filter in the query:

select date_trunc('month', created_at) as month,
       sum(amount) as revenue
from orders
where created_at >= {{ date_range.start }}::date
  and created_at <  {{ date_range.end }}::date
  and status != 'refunded'
group by 1
order by 1

Two or three inputs is the ceiling for a good app. Every additional dropdown is one more way for a reader to produce a number you will have to explain later. Start with the date range and one segment filter.

How do row-level permissions work?

Hex can restrict what a viewer sees by mapping the Hex user to a warehouse role, or by filtering at the query level using a variable that resolves per viewer.

Three approaches, in order of robustness:

  1. Warehouse role mapping — viewers query with their own role, so Snowflake or BigQuery permissions do the work. Strongest, but requires a Hex seat per viewer and matching warehouse accounts.
  2. App-level row filter — a query parameter like {{ current_user_region }} filters rows per viewer. Requires Hex to know who the viewer is, so this rules out fully public links.
  3. Separate apps per audience — one app per region or per client. Crude, but the easiest to audit and the hardest to leak.

If your dashboard contains client-level data and the link might be forwarded, use option 1 or 3. Option 2 fails quietly: a misconfigured variable returns everything instead of nothing.

Four checks, five minutes:

  • Open the app in an incognito window and confirm what a stranger sees
  • Confirm no cell with raw customer rows is selected for display
  • Confirm the owner account has read-only warehouse credentials, not an admin key
  • Confirm the scheduled refresh time matches when the source data actually lands

That third check is the one that bites. Apps run with the configured owner’s credentials; if the owner is a data engineer with write access and the app has a SQL cell a viewer can edit (some plans allow this), you have handed out write access by accident.

How do you keep the app from going stale?

Publishing is not syncing. The app view keeps the last published state until someone republishes, so the team needs a routine.

A workable one: the owning analyst republishes after any change to the underlying models, and the app shows a “last updated” markdown cell with the publish date. Set the cell text manually rather than trusting anyone’s memory. Add a weekly calendar reminder on the analyst’s side; no built-in mechanism will remind you.

Does the free trial let you test sharing?

Partly. You can build and publish within the trial workspace, but external sharing without viewer accounts depends on plan tier. Test with a real stakeholder account before you promise an open link in a meeting; discovering the limitation during a leadership review is a bad look.

Where does Hex sharing fall short?

Embedding in an external customer portal is limited on standard tiers, so customer-facing dashboards often need Enterprise. The app editor is also code-first: building a polished filter layout takes more clicks than a Drag-and-drop BI tool, and there is no chart-level theming beyond basic colors.

If your audience needs pixel-perfect branded dashboards, this is the wrong tool. If your audience needs accurate numbers with a date filter, it is excellent.

Who is this workflow for?

Teams where analysts own the logic and 10-100 stakeholders need to read numbers. Publishing one app beats answering the same Slack question every Monday.

It is not for solo founders (a saved Look or a Notion share is cheaper) and not for companies without a warehouse, since Hex has nothing to query without one.

Quick Summary

  • Publish the app, not the project, for anyone who should not edit SQL.
  • Two or three inputs is the practical ceiling; date range plus one segment covers most needs.
  • Row filtering needs an identified viewer, so public links cannot enforce it.
  • Verify the owner account’s warehouse role is read-only before sharing.
  • Apps do not auto-update; republish after model changes and stamp the date.
Tags: