What's new at Kodexa.

2026.14 brings help and announcements into the product, lets a task action ask the reviewer a question before it runs, and keeps knowledge applying as your organisation grows. It reads more kinds of document, stays readable while it is being upgraded, and ties what people can see to the roles they hold. Here are the highlights.

A review desk that asks first

• An action can ask before it runs. A task action can open a dialog first: choose a reason from a list, fill in a field, add a comment, or complete a short form. The dialog writes to the document as you go, so dependent choices update, and Cancel puts everything back.
• Tabs that do not apply step aside. A form tab can carry a condition that hides it when it has nothing to say, and a hidden tab never holds up Approve.
• Checks can warn without blocking. A validation rule can raise a warning instead of an error. Warnings show in amber with their own icon and never stop a task from completing.
• Overrides stick. An exception you have overridden stays overridden when the task is reopened, and a check the platform itself closed is never shown as an override.
• Nothing slips past the form. An Exceptions button in the task toolbar lists any open exception the form has nowhere to show, and long lists fold into a summary with a count per field.
• A cleared field is empty everywhere. Scripts and background steps read a cleared field as empty rather than as the text first read from the document, and a pick-list answer that is not on the list is kept and flagged instead of erased.
• Fewer surprises while you work. Assign on open explains why a task cannot be assigned to you and offers to open it anyway. Rows removed together stay removed after Approve, a refused row delete puts the row back, and the toolbar says when work is still queued. A lookup that answers late no longer lands on the wrong field or freezes the page.
• Kiosk keeps moving. A document reached while it was still being fetched in the background no longer hangs on “Downloading…”, and loading indicators clear when you leave the task.

Knowledge that keeps applying

• Sets match on what a feature is, not which copy it is. A knowledge set applies when a document carries a feature of the same type and contents as the one the set names, whenever either record was created, and a new document reuses the existing record instead of minting a duplicate.
• Reruns see what you attached. A document run again picks up the features attached to it since its last run, and adding or removing a feature from the document card now saves.
• Search every feature. The Knowledge Features page searches on the server, so a match is found however many features your organisation has, with the type filter and sort working alongside it.
• See the knowledge a run used. Every task has standing Task, Documents, Knowledge and Exceptions tabs, and analysts can see, and change, the knowledge a run applied.

More documents, read sooner

• More formats in. Images, JSON and XML arrive as documents with their own viewers, and HTML and iXBRL filings can be viewed and labelled as filed.
• Large workbooks open in seconds. Spreadsheets with dozens of sheets and tens of thousands of formulas parse in a fraction of the time, and the workbook viewer works from the parsed cells.
• Email attachments behave like files. Each attachment opens in the main view or beside the email, and PDF attachments are parsed and published as documents of their own.
• Steadier page preparation. Pages can be classified from their images, documents embedded in a page can be cut out into pages of their own, and a scan with only blank pages keeps them rather than failing.

Running the platform

• Help and announcements in the product. Documentation lives in the product, with guidance on the field an article governs, and announcements that appear when people open the app or enter an organisation.
• A maintenance page during upgrades. While the platform is being upgraded, people see a page that says what is happening and when Kodexa is expected back, with chat where it is available, and the app returns by itself.
• Permissions, not switches. The organisation show/hide switches are replaced by role permissions that the platform enforces. Every new project gives the organisation’s members an editing role on it, so the person who creates a project can work in it. Access granted through an identity-provider group is removed at the next sign-in once the person leaves the group.
• Schedules and polling in view. A Schedules view shows what is scheduled to run and how it went. Polling jobs check each project on its own cadence, can be asked to check now, and say what they decided.
• A refreshed look, in your colours. Navigation, forms, dialogs and notifications are refreshed, your brand colours and welcome images carry through, and the interface is available in US English, Japanese and Spanish. A project can be retired in favour of the project that replaced it while staying readable.

For teams building on Kodexa

• Ask for only the fields you need. Every list endpoint accepts a fields parameter, so a job that reads a few values per record no longer downloads the whole record.
• Mistakes caught at save time. Saving a data definition warns about a field name a formula refers to that does not exist.
• Plans that wait without being blocked. A step can wait for another step to settle without failing when that step fails, and the step graph shows how many documents are queued, running and done.
• AI calls accounted for. Costs are attributed to the assistant, step or script that made the call, repeated prompt prefixes are cached, and a failed call records what the provider said and where it went.

— The Kodexa Team

2026.10 sharpens how closely the platform follows the rules you configure. Validation reaches across the whole record, calculated figures are stored with the document, and the review desk holds on to every change a reviewer makes. Here are the highlights.

Validation and formulas that follow your rules

• Rules can look at the wider record. A rule can test a value against the record it belongs to, not just the field in front of it.
• Each rule carries its own exception. Rules on the same field are told apart, so correcting a value clears its own flag, and retiring a rule takes effect the next time a document loads.
• You decide which exceptions a reviewer may accept. Where you allow it, an Override action appears in the review workspace, including on documents already processed.
• Calculated fields are stored with the document. Opening a document keeps its figures current, so exports, downstream steps and everyone viewing it work from one set — wherever the definition was built.

A clearer, faster review desk

• A save carries every change a reviewer makes. Values and notes are written together with a record of what was applied, and a task confirms its documents are stored before it completes.
• Audit history and notes open in place. Select a value and its history and notes open beside it, including from a collapsed side panel.
• Review in a second window. Values and notes added in a popped-out document belong to the document itself, change history included.
• Documents reach the desk faster. Sharper judgement about which pages need optical character recognition, and bulk copies and moves that finish in a fraction of the time.

For teams building on Kodexa

• Your configuration deploys exactly as written. Deploying brings the platform in line with your configuration — rules you have retired come out, rules you have added go in.
• A firmer foundation to build on. Kodexa’s published interface is described precisely end to end, with the libraries built from it refreshed to match. Some of those interfaces tighten in 2026.10, so integration teams should read the developer changelog before upgrading.

— The Kodexa Team

2026.9 is about dependability. Knowledge sets apply exactly where you scope them and in the order you prioritise them, and the work you do — form values, project settings, and notes — persists reliably. Here are the highlights.

Knowledge that applies where you scope it

• Project-scoped knowledge applies to that project’s documents. A knowledge set scoped to a project is matched to every document processed in that project, and organisation-level sets continue to apply everywhere.
• Priority orders the knowledge that applies. When several knowledge sets apply to the same document, their items are ordered by the priority you set — highest first — both in the knowledge supplied to extraction and in the guidance reviewers see on the form.
• Feature links and item edits persist. Features linked to a set, the order you arrange its items in, and item deletions all save with the set and survive a reload.

Work that holds

• Your latest edits are protected on save. Typing straight after a save is preserved, and when a document genuinely changes outside your session you get a clear warning before it reloads.
• Project setup validates and persists. Required template options must be filled in before a project is created, and the values you enter are saved on the project and shown after a reload.
• Notes travel with the document. Notes on data values save with the document and appear for everyone viewing it.

— The Kodexa Team

2026.8 opens up your analytics to embed wherever your teams work, gives you a clearer picture of how tasks move and how long work really takes, and lets reviewers drive tasks straight from the keyboard. Knowledge matching gets smarter and more forgiving, the review desk gets a broad round of reliability fixes, and the tools your integrators build on get harder-wearing. Here are the highlights.

Your analytics, embedded where you work

• Bring your Kodexa analytics into your own tools. Mint a short-lived, tenant-scoped token and pull your Kodexa dashboards and datasets straight into your own apps and portals — always scoped server-side to your own tenants’ data, so a token can only ever read what its caller is allowed to see.

A clearer picture of your work

• See when and by whom tasks were locked. Task locks now stream to your data lake as their own dedicated feed, and each task record now shows who locked it and when — so you can report on lock activity without guesswork.
• More accurate time tracking. Work Sessions now measures how long work takes from the browser’s own elapsed time rather than comparing clocks, so durations stay accurate even when a machine’s clock is off. New data-quality signals flag any sessions where a skewed clock would have distorted the numbers, instead of quietly folding them in.

Keyboard-driven tasks

• Bind shortcuts to task actions. Task templates can now assign keyboard shortcuts to their actions and to their Save and Cancel buttons, so reviewers can approve, reject, and move through work without reaching for the mouse. The shortcuts help dialog (Ctrl / ⌘ + /) lists them alongside your form shortcuts.

Smarter Knowledge matching

• Matching that heals itself. A knowledge set whose matching expression had picked up a broken condition used to stop matching every document silently. Broken conditions are now dropped automatically, so an affected set starts matching again on the next document with nothing to re-open or re-save — and deleting a set now reliably stops it matching.
• A more predictable expression builder. Negation can always be turned back off, negating something twice cancels out cleanly, and switching between Simple and Advanced modes now warns you before it can drop detail — so building matching rules is safer and less surprising. Deactivated features stay visible and removable rather than silently vanishing from the palette.

A steadier review desk

• Task state changes stick. Setting a task’s status — in bulk, from the details panel, or from the grid — now persists correctly instead of being dropped or cleared, and Bulk Lock now reliably confirms and locks the tasks you select. Opening an already-locked task no longer throws a spurious error when “Assign on open” is enabled.
• The filters you see are the filters you get. Task, task-group, and activity lists no longer report “nothing here” while a stale, invisible filter quietly hides work — what you see and what you get now always agree.
• Actions wait for the form to settle. Approve, Reject, and other task actions now stay disabled until a form’s background lookups finish, so a task can’t complete on a value that hasn’t been saved yet. Grid and document-viewer fixes round out the desk — cleaner column menus, correct reading order when copying from rotated pages, and no more stuck loading spinner on dense pages.

For teams building on Kodexa

• Stamp properties onto tasks as they’re created. A script can now attach custom properties to a task at the moment it’s created — a natural place to compute per-document grouping or routing keys — and those properties are queryable through the standard task API.
• A harder-wearing CLI. Dry-run pushes now validate against the live server, so CI can gate metadata changes on a real check rather than a client-side guess; credentials are redacted from error and debug output; and JSON output is complete and pipe-safe for clean automation.
• Heads-up for status integrations. If your integration sets a task’s status through the API, the field it sends has been renamed — update it to the new name (the developer changelog has the details); the old name is no longer accepted.

— The Kodexa Team

2026.7 is a polish release — no big new surface, just a faster, steadier day-to-day. Review tasks open more quickly, formulas and validation behave more predictably, and a new view shows how long work actually takes. Here are the highlights.

A quicker review desk

• Tasks open faster. Opening a task now does less redundant work and warms up the next task in a queue in the background, so reviewing — especially advancing through a queue — feels noticeably snappier.

Steadier data entry

• Fewer false formula errors. A formula that references data which isn’t there yet now settles to a sensible value instead of flashing a “calculation failed” error, and lookups that depend on a parent row resolve correctly the moment a new row is added.
• Clearer selection guidance. When a field offers a list of choices, the hint for the chosen option now stays visible beneath the field, so the guidance for the current selection is always in view.
• Roomier AI explanations. AI-generated explanation text now shows in a callout that wraps and grows to fit, instead of being clipped in a single-line box.

See where the time goes

• New Work Sessions view. A new analytics view compares how long a task was open against how long it was actively worked — so you can see where review time really goes.

Smoother activities

• A friendlier New Activity flow, at-a-glance provenance (see which workflow a task came from), and live progress on the activities list that tracks what’s actually happening.

— The Kodexa Team

2026.6 is a consolidation release. Organisation administration finally has a home of its own, the review desk opens and moves faster under real workloads, and the workflow surface can now be reached from the outside. Most of what follows is polish on the Activities model from 2026.4 — fewer clicks, fewer surprises, and a steadier experience for the people working documents all day.

A dedicated Manage area

• Administration in one place. Organisation profile, teams, document tags, secrets, intakes, custom modules, subscriptions, model library, and concurrency now live under a single Manage area, alongside Studio, Workflow, and Knowledge. Settings that used to be scattered now have an obvious home.
• Scoped to administrators. The Manage area is role-gated, so day-to-day reviewers keep a cleaner workspace while admins get one place to run the organisation.

A faster, steadier review desk

• Faster task loads. Review tasks open faster, and advancing through a queue no longer waits on a fresh load — the next task is prepared in the background.
• Large lookups no longer stall a document. A very large reference lookup used to hang a document on open. Those responses are now handled cleanly.
• The pop-out viewer behaves. Detaching the document viewer into its own window no longer freezes or drops keyboard shortcuts.

Keyboard-driven review

• Document shortcuts on the form. The full set of document-viewer keyboard shortcuts is now available on data forms, and resets cleanly each time a form opens.
• Fixed for Mac and recent Chrome. Shortcuts now trigger reliably for Mac users on current Chrome.

Connect your workflow

• Workflow connector. A new secured connector exposes the workflow surface — activities, tasks, and task groups, with their assignment, status, and membership actions — so external tools and assistants can work your queues, always scoped to the calling user’s access.

Steadier data entry

• Rows stay put while you type. Adding a row and entering a value no longer reorders the grid mid-entry.
• Edited-by-you markers on Promote and Copy. Promoting or copying a value into a field now shows the same “you changed this” marker as typing it in by hand.
• Fewer duplicate lookups. Selection-list lookups no longer refetch on unrelated changes, cutting flicker and load.

Task groups

• Build groups right from your task list. A new New Task Group button and an Add to group bulk action let reviewers bundle related tasks under a single assignment, without leaving the Tasks view.
• Assignees show up. The assignee picker on task groups now lists the right people.
• Search finds them. Task-group search now matches partial terms instead of coming back empty.

Behind the scenes

• Reliable automated hand-offs. One activity spawning the next now runs reliably across every environment.
• Line-item analytics populate. Nested line-item detail now flows through to analytics instead of arriving empty.
• See who’s working. Optional, organisation-gated presence gives an accurate view of who is online and how long tasks take to open.

Upgrade notes

• Find your settings under Manage. Access, control, and reporting settings have moved into the new Manage area.
• Workflow connector access. The connector honours each user’s access — enabling it grants no broader visibility.
• Presence is opt-in. Presence and task-timing signals stay off until enabled per organisation.
• Analytics backfill. Line-item analytics populate on newly processed documents; reprocess if you need historical coverage.

— The Kodexa Team

2026.4.1 rounds out the Activities release with an analytics-ready view of your data, sharper navigation through multi-document files, and a batch of review-desk refinements — correct totals the moment a document opens, reliable saves, and keyboard-driven forms. It shipped between 2026.4 and 2026.6; here are the highlights.

Your data, ready for analytics

• Every change, available to query. Kodexa now mirrors metadata changes, activity events, and processed content into a store your analytics tools can query directly. It’s on by default for new deployments and opt-in for existing ones.
• Activities and step detail, queryable. Activity status changes land with rolled-up step errors and full exception detail, so dashboards can surface what failed and why without touching the platform database.
• The post-processing view. Content is exported as reviewers see it, so analytics work from the same picture your team does.
• Traceable to the session. Every change carries its task and work-session context, so you can correlate a change back to the user session that produced it.

Navigate multi-document files

• Page Groups picker. A new button in the document viewer lists each classified section of a multi-document PDF — for example Invoice — Pages 1–3, Delivery Receipt — Pages 4–6 — each with a colour swatch and click-to-jump, so reviewers land on the right section instantly.

Keyboard-driven forms

• Forms declare their own shortcuts. Data forms can now define keyboard shortcuts bound to named actions, scoped per form so they reset cleanly each time a form opens. Reviewers get fast, consistent key bindings without leaving the keyboard.
• Navigate from a shortcut. A shortcut can jump to a page, read the current page, or focus a field in the document viewer.

A clearer “take next”

• Know why the queue is empty. Take-next used to return a bare “no content” in a few different situations. It now tells you the reason — nothing queued, filtered out by team, or someone else took it first — so the interface can respond correctly.
• Work the same document in parallel. Picking up a task no longer locks the whole document, so several reviewers can work different tasks against the same document at once, where the task policy allows.

Comments where they matter

• Require a comment on an action. Task actions can now require a comment before they run, captured in the task timeline — so the “why” is recorded alongside the “what”.

Correct the moment a document opens

• Highlights on open — no edit required. A mismatch highlight (for example, a total that doesn’t match its line items) now shows on first paint, instead of staying hidden until a reviewer touches the document.
• Totals correct on open. Formula values show their computed value the instant a document opens, rather than a stale zero that only corrected after the first edit. Documents that are already correct open clean, with no false “unsaved changes”.
• Re-checks on row delete. Deleting a line item immediately re-checks any total that depends on it.

Reliable saves

• Approve always captures your last edit. Selecting a value and immediately approving no longer risks dropping it — save and approve wait for in-flight edits to settle, and surface a clear error rather than continuing if an edit is stuck.
• No edits lost during a save. An edit made while a save is already running is now included in the next save instead of being dropped.

Find work faster

• Faceted filtering across grids. Tasks, Task Groups, and Activities grids now filter by document feature and the document itself, with filter state that survives navigation.

Upgrade notes

• Take-next response changed. Integrations that relied on the old bare “no content” response should move to the new result with reason codes.
• Analytics export. New deployments get the analytics export by default; existing deployments opt in through storage configuration.
• Knowledge dependencies. Some knowledge sets need their dependency re-selected and saved once to pick up corrected path resolution.

— The Kodexa Team

This is the release where the platform’s vocabulary lands. Document-heavy work on Kodexa is now composed of two things: Activities for the runs of automated work the platform takes off your team’s plate, and Tasks for the work your people do where judgment matters. Activity Plans are the reusable definitions, set once at the organisation level and bound to whichever projects need them. Triggers wire the choreography between them. The result is one workflow, one audit trail, and patterns that get cheaper to roll out as more of the business adopts them.

Almost every surface in the platform has been updated to follow this model.

Activities are now the unit of automated work

• A noun for the run. Each run of automated work — extraction, validation, posting, classification, anything else the platform does on your behalf — is now an Activity. Activities have inputs, a status, a sequence of steps, and results you can monitor, retry, and audit.
• One creation flow. A single New Activity action is now the entry point across Studio, Workflow, and the project Activities tab. The previous mix of “New Task”, “New Plan”, and “New Job Run” has been consolidated.
• Steps are explicit. An Activity now reads as a typed sequence of steps — a script, an approval, an AI step, a service call, a task — rather than an opaque run.

A debug experience for every step

• Step debug panel. Open any step inside a running or completed Activity to see its inputs, outputs, and any errors raised — without leaving the Activity view.
• See what the model was asked. For AI steps, the prompt and its result now surface in the Activity status view, so an analyst reviewing an exception can see exactly what the model was given.
• Per-step detail. Each step carries its own execution detail and error trail, visible in the UI and through the API.

An operations view for admins

• Live queue and health. A dashboard shows the live execution queue, Activity health, and failure analytics across the organisation.
• Trend windows. Switch between 6-hour, 48-hour, 5-day, and 30-day views to spot regressions and capacity issues over time.

Fair scheduling at the organisation level

• Per-organisation concurrency. Admins can set how many Activities run concurrently for their organisation. The scheduler enforces those limits so no one organisation saturates shared capacity.
• Deterministic dispatch. When several organisations are waiting for capacity, slots are handed out fairly and predictably.

Exceptions you can act on

• Override inline. Override an exception right in the data form, with the override persisted and its detail visible alongside the field. Overrides survive reload.
• Link to a runbook. Attach a support-article link to an exception so the reviewer gets a one-click jump to the right guidance.
• Conditional actions. Actions can be gated so an approval button knows when an exception is allowed to be overridden and when it isn’t.
• Open/closed filtering. Exception lists filter cleanly between open and closed, with smoother transitions on resolve.

Studio resources panel

• Task Templates and Activity Plans, side by side. The resources panel surfaces both for the project, so authors can see what’s available to bind without hunting through nested menus.

Analytics, expanded

• More dimensions. Task analytics now slice by Task Template, Activity Plan, and time window.
• Trace a document. See exactly which Activities and steps touched a given document.
• Safe bulk updates. Two reviewers updating the same queue at once no longer silently lose each other’s edits.

Permissions and team isolation

• Consistent access checks. Fine-grained access now applies consistently across analytics, execution history, and scheduling.
• Team names on task cards. Task cards now show the owning team, so cross-team queues read at a glance.
• Shared links stay scoped. Direct form links respect list conditions, keeping multi-team workflows isolated even when a link is shared.

Validation and formulas

• Batch validation, fixed. Validation now walks all relevant data rather than stopping at the first match, and formulas recalculate consistently after a document opens.
• Reactive formats catch more. Conditional formats now pick up the edge cases they used to miss.
• React to upstream changes. Formulas react to changes further up the chain without a manual reload.

Upgrade notes

• Vocabulary in your runbooks. If your internal docs talk about “plans” or “running a plan”, refresh them to use Activity Plan (the definition) and Activity (the run). The platform now uses these consistently.
• Concurrency limits. Admins should review the new per-organisation concurrency setting before peak load — the default is conservative and may need lifting for high-volume organisations.
• Existing plans carry over. Existing project plans bind cleanly into the new Activity Plan model; no migration action is required.

— The Kodexa Team

2026.02 is a broad reliability release. Task management gets a cleaner creation flow and stronger guardrails, data forms behave more predictably, large files upload without hitting a wall, and dozens of smaller fixes make the day-to-day steadier. Here are the customer-facing highlights.

Task management

• A cleaner way to create tasks. A new Create New Task button with a redesigned dialog makes task creation straightforward, and project settings let you show or hide the “Take Next” and “New Task” buttons to match how each team works.
• Require an explanation. You can now require a comment when a reviewer changes or adds a value in a data form, so the “why” is captured for the audit trail.
• Document lock/unlock. Lock a document to prevent concurrent edits during review, and unlock it when you’re done.
• Reliable status changes. Accept and Reject now reliably save task status, task templates keep their status when updated separately, and reprocessing a document cancels the old tasks that no longer apply.

Workflow

• Skip the extra step. Reviewers who belong to a single organisation go straight to their workspace instead of a selection screen, and the move dialog skips steps when there’s only one sensible option.
• Clearer task actions. The Accept button now shows a disabled state while a form still has exceptions, task rows respond to a click anywhere in the row, and Studio shows when a file has related tasks with a quick jump to them.

Data forms and line items

• Collapsible line-item summaries. Invoices with many line items grouped under sub-accounts now show a collapsible summary header — account, total, page, and exception count — so long documents are easier to scan.
• No accidental reordering. Column sorting was removed from the line-item area, where an accidental sort could put data integrity at risk.
• Click exceptions to jump to them. Clicking an exception message scrolls straight to the field it refers to, and overridden exceptions stay visible where you expect.
• Click to select from the PDF. Click a single value in the document instead of dragging to highlight it.
• Currency, the way you want it. A “prefer two decimal places” option for currency fields interprets amounts consistently.
• Clearer required fields. Blank required fields show a red outline, and required fields across the app are marked with an asterisk.

Large file support

• Upload big files. Large documents now upload cleanly, well past the previous limit, so oversized files no longer fail on the way in.

Data and knowledge

• Deletes preserve history. Deleting data now keeps it out of everyday views while preserving the full audit history behind it.
• Knowledge applies itself. The platform evaluates a document’s features and applies the matching knowledge sets automatically, with richer filtering to find the documents you mean.

Access and teams

• Team management. Create, edit, search, and manage teams within an organisation.
• Group sync from your identity provider. Memberships can sync automatically from your identity provider’s groups, including pattern matching for flexible rules.

Performance and stability

• Faster, steadier processing. Processing functions start faster and scale to load without redeployment, so busy periods stay responsive.
• A steadier interface. Task status saves correctly, page navigation is consistent, false “unsaved changes” warnings are gone, and a batch of save-conflict and blank-screen issues are resolved.

Upgrade notes

• Mandatory comments. Turn on “require an explanation on save” per task template where you want changes justified.
• Large uploads. Large-file uploads use a two-phase flow; no action is needed for end users.
• Group sync. New group synchronisation runs on a schedule, so membership changes may take a short while to apply.
• Deletes. Data deletes are now reversible-by-history rather than immediate; plan for retention accordingly.

— The Kodexa Team

2025.12 focuses on the interface and on data-form reliability — steadier navigation, forms that behave predictably, and a batch of fixes that make review work smoother day to day.

Interface and navigation

• Steadier pages. Pages no longer flash the wrong content during transitions, new organisations appear in the navigation immediately, and project headers match the page you’re actually on.
• Clearer workspace. Workflow no longer flashes documents that aren’t there, pipeline diagrams keep their connecting lines, and Task Mode and Kiosk Mode now look distinct so it’s obvious which one you’re in.
• Document interactions. Value highlighting works in the document viewer, copying text from a document gives a clear confirmation, and full filenames are visible on data-form tabs.

Data forms and fields

• Better keyboard flow. Tab moves between fields with a single press, and focusing a field at the bottom of a form no longer scrolls you back to the top.
• Clearer exceptions. Exception markers show correctly on customised forms, override messaging is clearer, and confusing highlights on date fields are gone.
• Field values you can trust. Original values display correctly, linked data forms look consistent in dropdowns, and form configurations save and reflect reliably.
• Dates and metadata. Date formats are consistent across forms, created and modified times display in a consistent timezone, and metadata tables show document and task status.

Line items

• Stable while you edit. The line-items table holds its position when you insert a row, adding several rows at once works cleanly, and deleting a row no longer risks losing newly entered values.

Tasks and exceptions

• Smoother task flow. Accept and Reject can auto-assign the next task, action buttons set their values reliably on save, and the basic task creator finds the right template.
• Exception overview. The task overview shows open and closed exceptions correctly, and override buttons disable properly after a task is reopened.

Documents and processing

• Bigger uploads. Documents that previously failed for size now upload successfully.
• Reliable extraction. Missing extracted data after reprocessing is fixed, and the AI assistant and document pre-processing run reliably.

Configuration and automation

• Scheduled jobs. Schedule options save and persist, custom intervals update to the minute, and jobs show the correct schedule status.
• Assistants and forms. Model steps can be reordered again, form configurations copy cleanly across projects, and the platform prevents duplicate product names.
• Form state. The Save button enables after a change, Cancel clears initial values correctly, and forms without data groups clear their exceptions when groups are added.

AI and data movement

• Steadier AI processing. Better handling of rate limiting and long inputs, and cleaner recovery when a template can’t be processed.
• Reliable data movement and publishing. Service-to-service data movement works, and publishing reliably detects the documents to send.

System stability

• Fewer dead ends. No more unresponsive pages when switching data groups or saving from Studio, better modal and overlay behaviour, and clearer error feedback when something unexpected happens.

— The Kodexa Team

Kodexa Platform Release 2025.11 focuses on data accuracy improvements, workflow reliability enhancements, and user experience refinements. This release delivers enhanced currency field handling, improved form interactions, and system stability optimizations that ensure data integrity and operational efficiency.

This comprehensive update introduces enhanced currency and numeric field precision, improved line item management for smoother data entry, smart date field auto-completion, and workflow reliability improvements including automatic document assignment and better exception handling.

Improvements

Enhanced Data Accuracy

• Reliable currency handling with improved decimal precision
• Seamless numeric field interactions

Workflow Reliability

• Automatic document assignment
• Better exception handling
• Reduced false change warnings

Improved Forms and Interface

• Content visibility tooltips
• Consistent edit indicators
• Better override tracking

System Performance

• AI processing timeout protection
• Bulk operation improvements
• Faster Kiosk mode loading

Configuration Management

• Prevention of duplicate products
• Ability to delete assistants

Improved Line Item and Date Handling

• Multi-row addition support
• Smart date auto-completion

Enhanced User Experience

• Auto-assignment on document open
• Stable form layouts
• Proper notification display

We’re excited to announce the Early Access release of the KDX CLI - a modern, kubectl-style command-line interface for the Kodexa AI Platform.

What is KDX CLI?

KDX CLI brings powerful command-line management to the Kodexa platform, designed to feel instantly familiar to developers and DevOps engineers who work with Kubernetes and other modern cloud-native tools.

Key Features

• Dynamic Discovery automatically syncs with your platform’s OpenAPI specification, ensuring no deprecated commands as your platform evolves
• Profile Management seamlessly switch between environments with stored authentication profiles in ~/.kodexa/config.yaml
• Schema Validation catch configuration errors before they reach the API, saving time and reducing deployment failures
• GitOps Ready enable version-controlled infrastructure workflows across all your environments
• Flexible Output choose from interactive tables, JSON, or YAML formats to fit your workflow
• Extensible Plugins resource-specific operations like workspace execution and store reindexing

Who Should Use It?

KDX CLI is perfect for:

• Developers who prefer CLI workflows over web interfaces
• DevOps teams implementing GitOps practices
• Administrators managing multiple Kodexa environments
• Anyone scripting or automating Kodexa platform operations

Get Started

Visit our KDX CLI documentation to learn more and start using the CLI today.

We’re actively developing additional features and would love your feedback during this early access period. Let us know what workflows you’d like to see supported!

This release focuses on improving data accuracy, workflow reliability, and user experience across the Kodexa platform.

Data Accuracy Improvements

• Improved extraction accuracy for invoice line items and totals
• More reliable table structure recognition in complex documents
• Enhanced text recognition for low-quality scans and difficult documents

Workflow Reliability

• Better error recovery in document processing pipelines
• More reliable job scheduling and execution
• Enhanced monitoring with better visibility into workflow execution status

User Interface Updates

• Streamlined navigation with simplified access to key platform features
• Enhanced dashboard metrics and visualization of processing results
• Optimized UI for tablet and mobile devices

A major stability release with 36 bug fixes and significant performance improvements across the platform.

Performance Improvements

• Up to 95% faster processing for document processing pipelines
• Reduced memory footprint for large document batches
• Improved database query performance for knowledge retrieval

Bug Fixes & Stability

• 36 bug fixes addressing issues reported by customers across all platform components
• Better handling of edge cases in document processing
• More detailed diagnostic information and logging for troubleshooting

Platform Updates

• Updated core libraries to latest versions for security and performance
• Improved handling of high-volume document processing
• New metrics and dashboards for system health monitoring

Introducing Workflow Kiosk Mode, AI-enhanced processing capabilities, and major performance gains.

Workflow Kiosk Mode

• Streamlined UI designed for dedicated document processing stations
• Enhanced support for touchscreen devices with touch-optimized interface
• Role-based access with restricted access to specific workflow operations
• Fast document submission and real-time status tracking

AI-Enhanced Processing

• Better document type detection using advanced machine learning models
• More accurate field extraction with contextual understanding
• AI-powered validation of extracted data with smart suggestions
• Adaptive learning system that improves accuracy from user corrections

Performance Gains

• 30-50% faster processing across all document types
• Lower response times for API requests and operations
• More efficient document storage and retrieval mechanisms
• Improved handling of parallel processing jobs

Additional Features

• Updated authentication and authorization mechanisms for enhanced security
• New API endpoints for advanced workflow control and automation
• Comprehensive documentation and guides for all new features