Jedox 2026.1, what actually shipped and what it changes

2026-09-17 · Jedox

A practitioner's read of Jedox 2026.1 - the two JedoxAI agents and the modelling work they quietly require, the Integrator interface that is still beta, Python in FieldTransform and what it cannot do, plus the breaking changes and the pre-upgrade calls you need to make before booking a date.

Jedox 2026.1 was released on 14 April 2026 and is supported until 13 July 2027. As of today it is the only supported release: 2025.1 reached end of support on 31 July 2026 and now sits in the legacy table. So if you are reading this from a 2025.1 instance, the upgrade conversation is overdue rather than upcoming.

Two framing notes. All of this is cloud, because the last supported on-premises version is 2022.2. And I have stuck to what the platform release notes and the feature documentation actually state, including where two Jedox pages disagree with each other, because a version post that guesses is worse than no post.

First, the thing that makes "we are on 2026.1" an incomplete answer

The 2026.1 release notes are not a snapshot of April. They are a living document, last updated 17 September 2026, with dated sections added through the cycle:

• May 2026 - the launch content, split by product area: Web, Canvas, Charts, Dynatables, JedoxAI, Integration, Models, OLAP
• June 2026 - one item: multiple element selection inside subsets in Dynatables
• July 2026 - one item: query suggestions in the Reporting Agent
• August 2026 - six items, almost all JedoxAI and Integrator
• September 2026 - three items, including the native Microsoft Fabric connector and one flagged breaking change

Two consequences worth carrying into client conversations. Items added after May are not re-sorted into the product-area structure, so the same area appears in several places on the page. And more importantly, the patch level is the real answer, not the version. Two tenants both "on 2026.1" can differ by five months of features, and at least one breaking change lands on a specific patch rather than on the base release. Ask for the build.

A note on labels, since the wording matters. The support lifecycle page defines four availability states: GA (released to all users, on by default), Preview (opt-in, may still change), Early Access (on request through your account team) and Admin Preview (available, but an administrator must enable it). The 2026.1 notes apply almost none of them. Only four items carry any label at all: freeze panes as generally available, Cube Delta Extract as Early Access, and the new Integrator UI plus the new navigation sidebar as beta - and beta is not one of the four defined states. Everywhere else I say "documented as shipped", because that is all the source supports.

JedoxAI is now two agents, and one of them needs modelling work

This is the headline of the release, and the third attempt at the AI story. The progression is real rather than marketing drift:

• 2024.1 shipped a generally available AI Search in the Help Center, explicitly ChatGPT-based and carrying a warning not to share confidential information, plus a separate in-preview natural-language capability over Views.
• 2025.1 replaced that with a Support-gated beta and removed the previous generally available version from the Start page.
• 2026.1 ships two named agents. The JedoxAI page states it plainly: "JedoxAI is currently composed of two Agents: the Knowledge Agent and the Reporting Agent."

Split them by the job they do, because they have very different prerequisites.

The Knowledge Agent is the product and scripting helper. It reads the Knowledge Base, Jedox Academy, Jedox Consultant Best Practices and Jedox4Beginners, and answers on fields, steps, configurations and technical issues. It generates Groovy, JavaScript, SQL and RScript, and reviews scripts for Integrator projects. Since August each response ends with a citation badge linking to the articles behind the answer, which is the difference between an answer you can check and one you have to trust. Worth noting: Python is not in its documented generation list, even though Python arrives in FieldTransform in the same release.

The Reporting Agent is the data analyst. It gives conversational analysis over the OLAP database, builds and updates a Dynatable iteratively without rebuilding the whole context, and offers multidimensional KPI analysis through Data Insights from a cell value or KPI card.

Here is the part that turns into project work. The Reporting Agent page lists configuration that has to be done in the Modeler, per dimension, on every cube you intend to query:

• Dimension type set to "Business dimension" under Modeler > Cube > Dimension > Advanced > Type, which the page says is required to obtain the key drivers behind KPI performance
• A Default Read Element per dimension, called out as especially important for Currency, where the Reporting Currency should be the default
• Name attributes, so people can ask about an account or entity by its readable name
• "Comparison Version 1" and "Comparison Version 2" attributes in the Version dimension

None of that is a switch. It is per-dimension metadata on an existing model, and without it the agent will not produce driver analysis. If a client expects to enable AI and get insights on Monday, this is the conversation to have first.

The documented limitations are equally worth knowing before a demo. The agent works with a single cube context, so one cube per query and nothing spanning a Sales and an HR cube. Charts cannot be edited or saved. Standalone report mode is not supported. Variance results are restricted to narrative output, and ranking on variance queries is not supported yet, which rules out questions like which customers show the largest negative variance to budget.

The monthly additions have mostly been aimed at making the Reporting Agent usable. August lifted version comparison beyond Actual versus Budget to any supported versions, which is the change that matters most in practice, since the old restriction was the reason it could not answer most real planning questions. August also added streaming responses and descriptive names for generated Dynatables and charts. July added proactive query suggestions. September replaced the generic loading spinner with step-by-step progress descriptions.

On availability and hosting, be careful with the wording. Neither the release notes nor the agent pages attach a label; the one explicit statement is in the FAQ, which says JedoxAI is generally available for users within the platform. JedoxAI is hosted on Microsoft Azure within Jedox's own environment and uses the Azure OpenAI Service, and Jedox prints its own caveat that large language models can produce inaccurate or misleading answers. Conversations are not stored after logout, and each user has an isolated session.

Regional defaults are the sentence to know for European clients. The FAQ lists UAE North, Germany West Central and Switzerland North as regions where JedoxAI is disabled by default and needs written approval to your Customer Success Manager, because AI requests may be processed temporarily in a data centre in another region. The JedoxAI page still lists Australia East as well. The two pages disagree, so anyone in Australia East should ask their CSM rather than trust either page.

The new Integrator interface, and why I would pilot rather than migrate

Integrator got a redesigned interface built on survey and interview feedback, with a consistent look across all component editors. You switch between legacy and new with one toolbar button, and the legacy UI is not going away in this release.

What only exists in the new UI: project and component search across all projects with filtering by component type, FlowGraph zoom controls with fit-to-screen, FlowGraph direction set horizontally or vertically, a component count in the project view, on-the-fly variable creation while configuring a component, a mini map, a Duration column for completed and in-progress jobs, and search in the monitor log.

The reason for caution is that the notes say plainly it is not yet feature-complete, and Jedox publishes a feature comparison page that is the real migration document. It splits gaps into two lists. Not carried over: Save all, Sort by type, Pin, the Properties panel, FlowGraph PNG export, runtime details as a flow graph, delete executions, the Help button and Validate. Coming soon: exporting multiple projects, remove unused components, download project documentation, tracepoints, define security in project, dynamic edit mode, adding or renaming nodes, copying and grouping nodes, and showing variable values.

Three of those hit real projects: Tracepoints, Download project documentation and Validate. If your debugging or handover routine depends on any of them, the legacy UI is still your working environment. Pilot the new one in a non-production project and keep the toggle.

One inconsistency to be aware of. The release notes and the User Settings documentation route enablement through self-service: a Beta features panel in the User Settings dialog, effective immediately on OK and easily reversed. The Knowledge Agent page, updated the same month, says the new Integrator UI is an in-preview feature that must be enabled by Jedox Support. Check your own User Settings dialog first.

Integrator beyond the interface

Python in FieldTransform. A new FieldTransform function runs Python 3, using GraalPy. For anyone who has been writing Groovy for row-level logic this is the most welcome change in the release, and the fence around it needs stating in the same breath: only the Python Standard Library is supported. No pandas, no numpy, no requests. Local filesystem and network access are disabled in the sandbox, and external resources go through the Integrator Scripting API instead. print() does not work, so log with LOG.info("message"). The FILE API readBinary and writeBinary methods are not supported. A running Python script cannot be interrupted gracefully, so the load or job has to be stopped twice.

The quiet trap is the Buffered setting, which is on by default: leave it enabled on a script that depends on row number or randomness and you get cached answers rather than an error. That is worth checking before anyone debugs the wrong thing for an afternoon.

Cube Delta Extract. The most requested capability in the release and the one nobody can simply turn on. It extracts only new or changed data since the last extraction, as a delta between two time periods. It is labelled Early Access and available on request only, through your Jedox sales representative. Two modes are documented: Previous delta, which returns the previous delta values and is useful for validation, and Recalculate. Running Recalculate twice returns an empty result by design, not because something failed.

Protected projects, from August. Integrator projects can be protected so business logic and proprietary ETL configuration cannot be viewed or modified without the right permissions, shown by a protection badge on the project icon. Jobs, transforms and data flow logic are hidden while the project stays executable and schedulable. It is aimed at implementation partners, and it is the first documented way to ship ETL logic a customer cannot open. The catch: only projects inside a Jedox model can be protected, so loose projects have to be moved into a model first.

A native Microsoft Fabric connection, from September. It reaches Fabric Data Warehouses and Lakehouse locations through their SQL endpoints, lets you browse workspaces, schemas, tables and views, and supports read and write-back jobs, with four component types (Relational extract, Relational table extract, Relational load, Relational SQL load). An OAuth Token connection is required, so there is no basic-auth path and the work cannot be scoped until someone in the customer's Entra tenant can supply the token endpoint, client credentials and Power BI API scope. I have written this one up separately as a setup template, and it changes the recommendation in my older post on getting Jedox data into Fabric.

JSON Extract gained looping at different levels within a JSON file. Note the two open issues attached to it below.

One behaviour change that is easy to miss, filed under Integration rather than as a breaking change: if a project containing an external job sits in a namespace and a model is named in the job's scope, that model must be in the same namespace. Check cross-namespace external jobs before you upgrade.

The quieter front-end work

Less discussed than the AI, and more likely to be noticed by users on day one.

• Dynatable sorting. Row header columns sort A to Z or Z to A while preserving hierarchy, data columns sort by numeric value, and the active sort shows in the column header. The notes say this works across all contexts including Canvas and the Power BI integration, so it changes what embedded consumers can do too.
• Multiple element selection in POV dimensions, with an aggregation of Sum (default), Average, Minimum or Maximum, shown as tags in the POV area. The prerequisite is the part to write down: multi-select requires cross drill-down and data entry to be disabled, which is only the default on new Dynatables, so existing ones need it enabled by hand. An input-enabled Dynatable cannot offer multi-element POV selection.
• The Select Element window was reworked in both Charts and Dynatables for large hierarchies: four sorting modes (Default, Ascending, Descending, Custom), select and deselect all, flatten and unflatten, and a Shortcuts split button offering Select Branch, Select by Level and Select Base Level. Select Base Level is the real time-saver on deep hierarchies.
• Canvas freeze panes are generally available, reachable in the Inspector on a WSS Frame object with no feature flag, up to 1600 pixels wide and 1200 high. Settings made in 2025 carry over. Delete the leftover flag from Administration Settings, and search for it exactly as the docs spell it, features.rollout.canvas_freze_pane, typo included.
• A redesigned navigation sidebar, in beta. Less frequently used actions moved into the Administration area, which makes this the change most likely to generate "where did that go" tickets, especially for teams with written click-path documentation.
• Report title bars were removed from the main viewing area to free screen space, and the Show title option is gone from report properties. Existing external hyperlinks created with the title bar keep it for backward compatibility, so an embedded portal can end up with a mix until links are regenerated.
• Polish joined the supported Jedox Web languages.
• White cell borders under browser zoom are, in the notes' own wording, significantly improved rather than fixed, with Firefox and Chrome named. Keep the hedge when you tell a client.
• The About dialog gained a Copy button that copies all versions, including feature flags and beta features, as JSON. One click instead of a screenshot when you open a support case.
• Financial Consolidation model licences can now be shared across development, QA and production for the same customer, so complete databases can move between tiers without manual re-licensing.
• Archived databases can be excluded from auto-loading through a context-menu option, lowering server memory use. It is per database, so it needs doing deliberately.

The upgrade checklist

This is the part with the most practical value, and the reason to read the deprecations page rather than a summary of it.

Do these before you book the upgrade date:

• Supervision Server outbound access is now whitelist-only. SVS restricts outbound network access to the internal OLAP and ETL service hosts; anything else is blocked by default. Custom SVS scripts calling other hosts must be whitelisted through Jedox Support before upgrading. This is the one item with a hard third-party dependency and no admin-side fix afterwards, so inventory every outbound call (REST, SMTP relays, file drops, internal APIs) and raise the request early.
• Review who can reach "Clear all" in the Modeler. Announced September 2026 and flagged breaking: Clear all now clears the entire cube in a single operation regardless of the user's rights to view or edit particular elements. Clear selection is the rights-bounded action, and Jedox recommends restricting Clear all to users with full cube access. This is a permissions review and some retraining, not a code change, and it is the item most capable of causing real damage.
• Move any integration off HTTP GET for OLAP login. /server/login no longer accepts GET; use POST. Grep custom scripts, monitoring jobs and homegrown wrappers. Anything that built the login URL with query-string credentials had a second problem worth fixing anyway. My Python pull post uses POST throughout.
• Rework anything that parses login logs. The format now includes the licence serial and user index, the user ID and the login client (Jedox Web, Excel, OData and so on), which breaks existing analysis. The upside is that licence-usage reporting can finally split Web from Excel from OData without guessing.
• Check custom Java and Groovy against the deprecated API methods. The 2026.1 round is mostly the withAttributes overloads on IDimension, IElement.getSubTree() replaced by collectSubElements(), and the four-argument loadCells(...) replaced by a five-argument form that adds boolean useAudit. Most of it is a mechanical rename; the loadCells change is a behavioural decision. Note this is the second round, after a set removed in 2024.2.

Then know about these:

• A maximum HTTP response size in the In-Memory DB, default 10 GB and described as configurable. Requests over the limit return an error instead of letting the server exhaust memory, and responses over 20 percent of the limit are logged automatically. Two precisions matter. It arrives with patch 2026.1.7, not with base 2026.1. And although the note calls the limit configurable, the linked configuration page does not document a parameter for it, so raising it today means a support ticket. Do not go looking for a palo.ini key that has not been published. The 20 percent logging threshold is the useful part: anything over 2 GB starts appearing in the log, which gives you the list of extracts that will hit the wall next.
• Global OLAP connections have required HTTPS on port 443 since 1 December 2025. Date-driven across all versions rather than an upgrade event, covering connections from Excel, Integrator and Jedox Web to external OLAP servers. The internal jedox-olap connection is exempt.
• Salesforce is retiring the OAuth 2.0 username-password flow in 2027. Not a Jedox change, which is exactly why nobody catches it by reading Jedox release notes. Anyone with a Salesforce extract has a deadline set by Salesforce.

Deprecations to plan around:

• The AI Search Help Center is deprecated, announced May 2026, in accordance with the Knowledge Agent deployment. Deprecated, not removed: the page's own policy is that deprecated features keep working until a release names their removal.
• The SAP transport package Z_PALO is deprecated in 2026.1, with a warning at execution until it goes. It only affects SAP Netweaver 7.00 to 7.30, so check with the customer's basis team before treating it as live risk.
• Lucanet connection support is listed against 27.1. Affected customers are pointed at their Customer Success Manager rather than Support.
• Spreadsheet Macros move to PHP 8.3 in 2027.1, automatically on cloud. This is the strongest forward-looking item, because 2026.1 still runs PHP 7.3.33 on the Spreadsheet Server while Apache is already on 8.3.0 and the Supervision Server on 8.3.3. Three PHP versions in one platform is also why macro advice differs by component. Audit macros during the 2026.1 project, while you are already in the code, not in 2027.

The rest of the shipped component list: Apache 2.4.66 and OpenJDK 21.0.5, the latter mattering to anyone shipping custom Groovy or Java.

Known issues, in both directions

Six issues are open on 2026.1 and not on 2025.1, and the pattern is the story: they cluster around what the release added. ETL-1993 and JI-310 both concern the new multi-level JSON looping, and JI-310 is a design decision rather than a bug, stating that the legacy Integrator UI will not be updated for the new configuration and can read only the first looping level. ETL-1993 has a partial workaround, prefixing the second duplicate field name with a dot, which makes the extract run but returns that field as an array. VIT-1055 (a Dynatable from Drill Anywhere cannot be saved) and VIT-1074 (column sorting is unset when a POV variable changes) sit on the new Dynatable work. JEA-172 is an empty Select Elements dialog in Paste View when a dimension is added only to Rows, and WCA-457 is an error when closing an out-of-focus Web Spreadsheet tab in Designer.

Two issues where upgrading is itself the fix. WCA-86, where the ribbon shows the wrong options when switching between a Web Spreadsheet tab and a Canvas tab, is listed as fixed, though the page's "Fixed in" cell contains the text "Jedox Web" instead of a build number, so I would not quote a version for it. TOLP-1613 is fixed in 26.1.0: a 2025.1 Excel COM client against a 2026.1 OLAP server loses sight of an archived database after a restart. That one is really a mixed-version problem, so upgrade the Excel COM clients alongside the server rather than leaving old clients pointed at a new OLAP.

What the platform notes do not cover

Worth saying explicitly, because a reader who assumes the platform notes cover everything will conclude that nothing shipped elsewhere. The 2026.1 platform notes contain no Excel add-in section, no Marketplace section and no dedicated security section. The release notes index lists Add-ins for Microsoft 365, Cloud Console, Mobile App, the Financial Consolidation Model, AIssisted Planning and Jedox Solutions as separate documents.

AIssisted Planning deserves its own paragraph precisely so it is not folded in. It is the prebuilt forecasting family, it runs its own release train with its own numbering, and its 2026 versions shipped on 27 February (26.1.0), 13 April (26.1.1), 28 July (26.1.2) and 14 September (26.1.3). Those are not platform 2026.1 numbers, and it appears nowhere in the platform notes. Conflating the two is the easiest factual error to make in a conversation about this release.

The short of it

2026.1 is an enablement release with a long tail. The two agents are a more credible shape than the earlier attempts, and the Reporting Agent's value depends on modelling metadata that most existing cubes do not have, which makes it a project rather than a setting. The new Integrator interface is genuinely better to work in and genuinely incomplete, so the honest recommendation is a pilot with the toggle kept. Python in FieldTransform is the change most likely to alter how you write transforms, provided nobody expects third-party libraries.

The upgrade itself is not heavy, but it has three items that need a decision rather than a click: the SVS whitelist request, which has to go through Support before the date; who is allowed to press Clear all; and any integration still sending a GET to /server/login. Everything else on the list is a check.

And the habit I would keep from this cycle: record the patch level in your solution documentation, not just the version. With features arriving monthly and at least one breaking change tied to 2026.1.7 rather than to April, the version number tells you less every release.