Stack replacement

Replacing Confluence and Jira

An Atlassian estate has administrators, permission schemes and an auditor, and none of those are content problems.

What the move costs

Confluence and Jira form a governed estate rather than two products, with shared identity, space and project permissions, admin controls and retention obligations. Neither has a Polaris importer or connector. Confluence exports spaces as XML, HTML or PDF and Jira exports issues as CSV, but page macros that render live Jira data export as dead snapshots. A single-team pilot is the only realistic starting point.

Importers available
Neither product
Biggest hidden cost
Permissions and macros
Recommended scope
One team, not the estate
Polaris auth
Passwordless email code only

This is an estate, not a stack

Every other page in this cluster describes tools a team chose. An Atlassian estate is usually something a company decided, with an administrator who owns it, a permission scheme somebody documented, and at least one process that exists because an auditor asked for it.

That changes what a migration means. The content is the easy part. The hard parts are that spaces have restricted pages whose restrictions are not visible in an export, that identity flows through Atlassian Access rather than through each product, and that somebody has to answer for where the data lives afterwards.

The specific trap is the Jira macro. A Confluence page that renders a live issue table looks like documentation and is actually a query. Export the space and those macros come out as dead snapshots or as nothing at all. Teams discover this after the export, which is the wrong time.

What an Atlassian shop has that this cluster's other readers do not

Constraints to check before starting

  • Space and page-level restrictions that do not survive an export
  • Identity through Atlassian Access, with SSO and directory sync
  • Retention and residency commitments made to customers or regulators
  • Deploy or release gates that read Jira issue transitions
  • An administrator whose job includes saying no to this

What Polaris offers against those

  • Passwordless email-code sign-in, and no password path at all
  • Row-level security in Postgres, with agents and humans in the same members table
  • Connections authorised once, org-wide, stored server-side and unreadable by browsers
  • A per-task work log that itemises what an AI worker did and how long it was billed for
  • No permission-scheme equivalent, which has to be said plainly

A pilot, not a programme

Five stages that stop being reversible only at the last one.

  1. 1

    Pick a team with no compliance surface

    Internal tooling, growth, or a team whose work never appears in an audit. The pilot is about whether people prefer working this way, and running it inside a regulated workflow answers a different question badly.

  2. 2

    Inventory the macros in that team's space

    Search their Confluence space for Jira issue macros, page includes and any blueprint-driven template. Each is a page that will not export as what it looks like. Decide for each whether it becomes a written document, a Polaris task list, or something you keep in Confluence.

  3. 3

    Rewrite rather than export

    For a single team's living documentation, rewriting is usually faster than converting an XML space export, and it removes a decade of pages nobody opens. Export the full space to PDF for the archive and keep Confluence readable.

  4. 4

    Rebuild the current board by hand

    Export the team's live issues from Jira to CSV, then create them as Polaris tasks with owners and dates. Leave every closed issue behind. Anything a release gate transitions stays in Jira until that gate is rewritten.

  5. 5

    Give the pilot team an AI worker and measure one thing

    Whether work comes back finished. Hiring takes about a minute and produces a SKILL.md file the team can read and edit. The machine delivers as a comment with files, ticks its acceptance criteria, and never closes its own task. If that changes the team's output, you have a case to take to the administrator. If it does not, you cancelled nothing.

Questions people ask

+Can we import a Confluence space into Polaris?

No. Confluence has no Polaris importer and is not in the connection catalog. Confluence can export a space as XML, HTML or PDF, which is a fine archive and not an import path. For a single team's living pages, rewriting into Polaris Docs is usually faster than converting an export and produces documentation somebody has actually read this year.

+What happens to Jira macros embedded in Confluence pages?

They are queries rendered as content, so they do not survive an export as living data. A page showing an open-issues table becomes a snapshot at best. Identify these before you export, and decide per page whether the content becomes a document, a Polaris lane, or stays in Confluence.

+How do page restrictions and space permissions map across?

They do not map. Polaris has organisations with members and row-level security in Postgres, not Confluence's space and page restriction model. A team relying on restricted pages for confidentiality needs to check that Polaris's model is sufficient before moving anything, and for some content the honest answer will be that it stays in Confluence.

+Does Polaris support SSO or SCIM?

Authentication is passwordless email codes, with no password path in the product. There is no SAML SSO or directory sync. For an organisation whose identity requirements run through Atlassian Access or an equivalent, that is a blocking difference and it should be raised at the start of the conversation rather than the end.

+Is the cost saving worth this much work?

For a large Atlassian estate the licence saving is real and the migration effort is larger than any figure on the cost pages accounts for. The stronger argument is delivery rather than price: free software plus roughly two dollars per human-equivalent hour of AI-delivered work, itemised and challengeable. If that is not compelling on its own, the licence maths probably will not carry the project.

Your next hire takes 60 seconds.

The software is free — unlimited people, tasks, workstreams and docs. You pay only for work an AI worker actually delivers, itemised by the hour.

Get started free

Last checked .