Hex vs Deepnote: whose permissions and which version apply when business users run a report?
Hex better suits reporting teams that need drafts separated from live apps. Deepnote’s new apps support per-user runs, but require acceptance of beta and immediate-update boundaries. Evaluate database identity, result sharing, published versions and compute costs together.
A business colleague opens a sales report, selects a region and quarter, and clicks Run. Whose database permissions execute the query, which results appear, and is the formula the analyst just changed already live? For analytics leaders, choosing between Hex and Deepnote should start with these three questions. Prioritize Hex when drafts must be clearly separated from the live app. Consider Deepnote if you accept its new app format, independent runs for each user and the responsibility of managing immediate updates. This analysis draws on official documentation. No user data was connected and no hands-on tests were performed; the scenario concerns ordinary business reporting only.
First establish which app type you are comparing
Both products now offer AI-assisted analysis and app capabilities. Hex AI is integrated into analytical work. Deepnote data app projects use an Agent to build a web page, with a notebook running SQL and Python behind it. The latter is still labeled beta, and the old notebook block apps are deprecated. Comparing an old Deepnote publishing workflow with current Hex can therefore invalidate conclusions about versioning.
Suppose a sales manager may see only their own region’s totals, while analysts maintain the metric definitions. Providing an app and hiding code addresses the interface; a region dropdown cannot replace database authorization. First establish which data each person may access, then decide whether shared credentials and results are acceptable. Parameters need bounded choices, sensible defaults and query-scope validation so a quarterly report cannot turn into an unlimited historical detail export.
Review execution identity separately from result sharing
The Hex OAuth documentation distinguishes personal credentials from embedded publisher credentials: the latter let viewers query as the publisher. An easier detail to miss is that results can still be shared across users by default when personal credentials are used. For individual isolation, check Keep private to each user. The trade-off is that only one editor can be active at a time, app caches are separate for each user, and scheduled runs update only the schedule owner’s cache. Teams must choose among result isolation, collaborative editing and response speed.
Deepnote’s least-privilege guidance also supports per-user OAuth on Enterprise, applying the querying person’s warehouse permissions; shared integrations instead use the integration’s credentials. People who can edit a project can write new queries, and an Editor can attach workspace integrations to projects they can edit. “This integration was originally attached only to the finance project” therefore does not restrict them. Personal OAuth also does not filter notebook outputs already generated by someone else; scheduled runs use the schedule owner’s credentials. Reviews must cover saved results and the scope of project access.
In Deepnote’s new apps, business users can receive App user access without opening the notebook. However, every output from their own runs remains readable through the app API, so undisplayed details must not be treated as inaccessible. The official self-service app guide also says not to put secrets in HTML or scripts sent to the browser. For the regional report above, narrow data in the query and returned results rather than merely hiding charts.
Credential storage has different semantics too. Hex Secrets are stored in an encrypted vault; ordinary environment variables are neither stored encrypted in that vault nor redacted. Do not confuse the two. Deepnote environment-variable integrations store values encrypted and make them readable by code after mounting them into a project. In both products, review executable code, the people who may modify it and credential scope together. “Encrypted storage” does not mean editors cannot use credentials to query data.
Live versions and latest results follow separate timelines
Hex’s publishing workflow runs a preview, shows differences from the previous published version and then publishes. It also permits skipping the preview, so team discipline remains necessary. Its app documentation explicitly says draft edits do not affect the live version. This boundary is useful for teams that must approve formulas before a weekly meeting: publish after review, rather than treating a saved edit as delivery.
Deepnote’s current new apps make a different trade-off. Its self-service guide says changes go live as soon as the Agent saves them; for larger changes, it recommends setting the app to Unpublished so only editors can access it. This makes iteration convenient but adds the operational cost of downtime or a separate validation project when continuous service matters. Do not transfer the old block apps’ staged-publishing promise to the new apps. If the live version must remain unchanged, validate your own release-isolation approach before adopting them.
Result freshness requires a separate check. Hex run settings can refresh expired published results in the background or disable automatic refresh to control update timing. Ordinary runs may reuse SQL caches within the allowed age. Deepnote’s new apps instead display each person’s latest run. Refreshing or submitting parameters reruns the entire notebook; executing blocks in the editor does not update what the app displays. Its execution documentation therefore supports explaining results as “my current run,” rather than assuming everyone sees the same precomputed report.
If every regional summary is already available to everyone, a shared read-only service account and common results can reduce operational complexity. If each person has a different data scope, make individual authentication a purchasing prerequisite instead of adding a filter after launch. Also establish who handles departures, role changes and expired tokens. A report that still opens does not prove its underlying query remains authorized; a failed query should not trigger an automatic switch to a more privileged account.
Version history helps recover code; it does not by itself prove what a business colleague saw yesterday. Leaders should require apps to show the report’s metric definitions, result-generation time and data coverage period, and retain important release records and query parameters. When figures are disputed, first confirm that the comparison uses the same account permissions, parameters and data timestamp, then discuss the formula. Otherwise, mistaking old results for output from a new formula leaves both rollback and reruns without a clear objective.
Let release requirements determine the budget
Hex pricing places scheduled runs and unlimited version history in Team, and OAuth database connections in the custom-priced Enterprise plan. Paid plans include Medium compute, while Team and Enterprise can opt into additional metered compute. If each person must inherit their warehouse permissions, do not estimate the bill from basic editor seats alone. AI credits, extra compute and the warehouse’s own query charges also need separate accounting.
Deepnote Team lists $39 per editor per month, billed annually, with scheduled notebooks, 30-day version history and monthly AI, CPU and GPU allowances. Enterprise is custom-priced. Per-user OAuth still requires Enterprise. New apps being available on all plans does not make individual app-access permissions and enterprise authorization features free. Each submission may start an independent machine and execute the entire notebook, so estimate costs from business users’ actual submission frequency, not just the analyst headcount.
The recommendation is conditional: validate Hex first if reviewed releases and controllable shared-report state matter most. Validate Deepnote first if you want an Agent-generated business interface and accept beta and immediate-update constraints. Use two accounts from different regions to check permissions, one formula change to check the release boundary, and one expired credential to check the error and who takes over. Record parameters, versions and result timestamps too. Only after these three acceptance checks pass should “the analyst sends a link” become a sustainable data-app delivery process.
Content note:This article is based on publicly available information and is intended as a reference for evaluating tools. Features, pricing, and related policies may change at any time; please refer to the latest official information.
Original source:learn.hex.tech