> ## Documentation Index
> Fetch the complete documentation index at: https://docs.flows.super.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Changelog

> New features, improvements, and fixes across the SuperAI Flows platform and API.

> [!NOTE]
>
> Filter by **API** or **UI** using the tags on each entry. Subscribe via RSS at `/changelog/rss.xml`.

**October 23, 2026**

Flow icons and safer flow updates

## Give each flow its own icon on the Hub, and update flows safely over the API

Each flow can now carry its own icon, badge colour and label on its Hub card. Over the API, every flow response includes a revision you can pass back with an update so you never overwrite someone else's changes, and documents with very large images or many pages extract more reliably.

**New**

* Choose an icon, badge colour and label for each flow, shown on its card in the Hub instead of the initials generated from the flow's name. On the Hub, open the menu at the top right of the flow's card: type a Badge label, pick a colour from the swatches, or select Change icon. People browsing the Hub can then tell your flows apart at a glance; the menu items are shown only to people who can edit flows.

**Improved**

* Demanding documents extract more reliably. Gemini models no longer refuse requests over our former 20 MB per-image limit; the model provider's own limits apply, and if the provider does reject a request you see the reason instead of a generic error. OCR with word bounding boxes now reads long documents one page at a time, so it finishes within a predictable time and reports an error if it can't. In the review screen, very high-resolution pages such as phone photos now display with their bounding boxes in place. → [Reviewing work in my work ](https://docs.flows.super.ai/reviewing-work-in-my-work.md)

**Fixed**

* Choosing Discard in the Unsaved Changes dialog now also removes input tasks added since your last save, such as Receive Email or a plugin input task, and you are asked before leaving the editor while such a task is pending.
* In the Data tab, choosing "Show in table" for a second search result after opening a run's task details now updates the table to that run.

**API**

* Every flow response now includes a revision. Pass it as expected\_revision when you replace a flow's definition, update a task, validate the flow or restore a version, and your change is applied only if the flow hasn't been edited since you read it, so concurrent edits are no longer overwritten. (`GET /api/flows`, `GET /api/flows/{flow_id}`, `GET /api/flows/{flow_id}/versions`, and 7 more)
* Flow settings carry an optional appearance, the icon, badge colour and label shown on the flow's Hub card. Set it when creating or updating a flow; it is returned with every flow and flow version. (`GET /api/flows`, `GET /api/flows/{flow_id}`, `GET /api/flows/{flow_id}/versions`, and 6 more)
* Service-account API keys accept an optional list of scopes when created, and the scopes are returned when you create, list or update a key. (`POST /api/service-accounts/{service_account_id}/api-keys`, `GET /api/service-accounts/{service_account_id}/api-keys`, `PATCH /api/service-accounts/{service_account_id}/api-keys/{api_key_id}`) → [Authentication](https://docs.flows.super.ai/api-reference/authentication.md)

→ [API reference](https://docs.flows.super.ai/api-reference/introduction.md)

**October 21, 2026**

Hub organization and more document formats

## Favorite and hide flows in the Hub, and process Word and PowerPoint files

The Hub now shows error and awaiting-input counts for every flow and lets you mark favorites or hide flows for your whole organization. Document tasks accept Word, PowerPoint, OpenDocument and RTF files, and the Document Classifier can run on decision models.

**New**

* Mark the flows you use most as favorites in the Hub: open the ⋯ menu on a flow card or list row and choose Mark as favorite (or Remove from favorite). The Hub lists your favorites first, most recently marked on top, with a filled star before the name. Favorites are personal and available to every role. Admins can also hide a flow for the whole organization from the same menu; everyone can then switch the Hub between Show visible only (the default), Show hidden and Show all, hidden flows show a muted eye-off icon after their name, and the sidebar lists visible flows only.
* Doc to Text, Doc to Structured and Document Classifier tasks now accept Word (.docx), PowerPoint (.pptx), OpenDocument (.odt, .odp) and RTF files. They are converted to PDF for you whether they arrive by upload, by email or through a file drop, so RFPs, tenders, contracts, SOPs and briefing decks no longer need converting by hand before a flow can process them.
* The Document Classifier can now run on decision models: Jev 1.13 and Jev 1.14, available globally, and Clef, available in the EU and globally. Choose one as the classifier task's model and it decides which of your configured document types each slice is and whether the next page continues it. Its probability for the chosen option fills the existing confidence output, so the classifier's output shape is unchanged and downstream tasks need no changes. Organizations pinned to the EU get Clef.

**Improved**

* Hub cards and the list view now show Runs, Pending HITL, Awaiting Input, Errors and Last activity for every flow, and you can sort by any of them. Errors counts runs that failed, were cancelled or have a failed task, matching the red run icon on the Data tab; Awaiting Input counts runs waiting on a webhook or another non-review input and is not shown to publishers. Sort icons stay visible in the list header and highlight the sorted column, and the actions column stays pinned to the right when the table scrolls.
* When you review Doc to Structured or Text to Structured work in My Work, in Work mode or in task details, the mouse wheel now moves through the pages of a multi-page document instead of only scrolling within the current page, so you no longer need the page pill or the thumbnails to get around. Selecting a field moves to the page of its value and keeps the connector line to its bounding box visible, also in the expanded view. In tables, a blank number cell is saved as empty, and if a value can't be accepted when you complete the item, the review stays open with your edits in place. → [Reviewing work in my work ](https://docs.flows.super.ai/reviewing-work-in-my-work.md)
* Searching inside a multi-select filter, such as Task Status in the Data tab or My Work, or Document Types when you review a document classification, now matches the option names you see: typing Err in the Task Status filter finds Error. In the Data tab and in My Work, the Run ID filter flags an ID that isn't valid so you can correct it. → [Reviewing work in my work ](https://docs.flows.super.ai/reviewing-work-in-my-work.md)
* Tags you remove from a run stay available in the Add tag suggestions, so you can apply them again later without retyping. A tag only leaves the suggestions when you delete it flow-wide under Settings → Tags on the flow's page.

**Fixed**

* Runs whose document is larger than the selected model can read, or whose model call stops responding, now end promptly with a clear error.
* Review tasks in My Work are grouped under a flow's current name after the flow is renamed. → [Reviewing work in my work ](https://docs.flows.super.ai/reviewing-work-in-my-work.md)

> [!WARNING]
>
> **Breaking API changes**
>
> * settings.hub\_sort in the profile response can now return four new values: error\_executions\_count\_asc, error\_executions\_count\_desc, awaiting\_input\_executions\_count\_asc and awaiting\_input\_executions\_count\_desc. (`GET /api/profile/me`) **Migration:** Update any client that validates or maps settings.hub\_sort to accept these four values, or treat unknown values as a sort it doesn't recognize instead of failing.

**API**

* GET /api/flows/\{flow\_id}/tag-suggestions returns the tags suggested for a flow's runs, including tags removed from individual runs that haven't been deleted flow-wide. (`GET /api/flows/{flow_id}/tag-suggestions`)
* GET /api/flows takes three new optional query parameters: favorites\_first lists the caller's favorite flows first, include\_error\_executions\_count adds error\_executions\_count to each flow (it isn't computed unless you ask for it), and visibility filters hidden flows. visibility defaults to all, so existing calls are unchanged. (`GET /api/flows`)
* Flow objects returned by these endpoints carry three new optional properties: error\_executions\_count, awaiting\_input\_executions\_count and is\_hidden. (`GET /api/flows`, `GET /api/flows/{flow_id}`, `GET /api/flows/{flow_id}/versions`, and 6 more)
* PATCH /api/profile/me accepts settings.hub\_favorite\_flow\_ids and the four new hub\_sort values (error\_executions\_count\_asc, error\_executions\_count\_desc, awaiting\_input\_executions\_count\_asc, awaiting\_input\_executions\_count\_desc), and GET /api/profile/me returns settings.hub\_favorite\_flow\_ids. (`PATCH /api/profile/me`, `GET /api/profile/me`)
* Each model in the GET /api/models response now carries an optional mode property, so you can tell decision models apart from the others. (`GET /api/models`)
* The DELETE /api/flows/\{flow\_id}/tags response now reports history\_rows\_purged for each deleted tag and total\_history\_rows\_purged for the whole request. (`DELETE /api/flows/{flow_id}/tags`)

→ [API reference](https://docs.flows.super.ai/api-reference/introduction.md)

**October 19, 2026**

Reliable runs and a sharper Hub

## Run flows exactly once with idempotency keys and stop runs at the first error

You can now retry run creation safely with an Idempotency-Key header, opt in to failing a run as soon as one of its tasks fails, and use the Hub as a review board with last activity and pending-review counts per flow.

**New**

* Send an Idempotency-Key header (a ULID) with POST /api/flow-executions to make run creation safe to retry. Repeating the request with the same key on the same flow returns the original run in its current state instead of creating a new one, even if the input differs; keys are scoped to a flow and never expire, and a malformed key is rejected with 422 validation\_error without creating anything. When you don't send a key, one is generated for you, and every run now returns the key in effect in its idempotency\_key field. (`POST /api/flow-executions`, `GET /api/flow-executions`, `GET /api/flow-executions/{flow_execution_id}`)
* A new opt-in flow setting, on\_task\_error, fails the whole run as soon as one of its tasks fails. Set it in the flow's settings when you create or update a flow through the API. Systems that watch a run's status then see it fail instead of complete when nothing was delivered, so they can react right away. Flows without the setting keep today's behaviour: independent branches continue and the run completes. (`POST /api/flows`, `PATCH /api/flows/{flow_id}`)

**Improved**

* The Hub's Sort and Filter controls now work like the ones on the Data tab, and each flow shows its last processing activity and how many of its runs are waiting for human review. Sort flows by last activity, by number of runs or by runs waiting for review, and filter by tags that must all match, so the Hub works as a review board that shows you where to look first. Users with the Publisher role don't see the pending-review count or its sort option.
* Moving to the next or previous work item in Work mode in My Work is faster. Neighbouring items and their documents are loaded ahead of time, so you can step through a queue of documents without waiting at each one.

**Fixed**

* API and MCP requests made over reused connections now complete consistently.

> [!WARNING]
>
> **Breaking API changes**
>
> * Flow execution responses can return two additional shapes in status\_data, AwaitingInputStatusData and FailedStatusData, and one previous shape was removed. (`GET /api/flow-executions`, `GET /api/flow-executions/{flow_execution_id}`, `POST /api/flow-executions`) **Migration:** Update clients that parse status\_data to accept the AwaitingInputStatusData and FailedStatusData variants, and make them tolerate variants they don't recognise.
> * GET /api/profile/me can return new values: last\_execution\_activity\_at\_asc, last\_execution\_activity\_at\_desc, flow\_executions\_count\_asc, flow\_executions\_count\_desc, awaiting\_hitl\_executions\_count\_asc and awaiting\_hitl\_executions\_count\_desc in settings.hub\_sort, and contains\_all in the mode of settings.hub\_tag\_filters items. (`GET /api/profile/me`) **Migration:** Treat hub\_sort and hub\_tag\_filters mode as open enums: accept the new values and fall back to a default sort or filter mode for any value your client doesn't recognise.
> * The endpoint PATCH /api/flow-executions/\{flow\_execution\_id}/status was removed. (`PATCH /api/flow-executions/{flow_execution_id}/status`) **Migration:** Remove calls to this endpoint; clients can no longer set a run's status. Read the status through GET /api/flow-executions/\{flow\_execution\_id}, and use the on\_task\_error flow setting if you need a run to fail when one of its tasks fails.

**API**

* Flow responses include two new fields, last\_execution\_activity\_at and awaiting\_hitl\_executions\_count, and the flow settings object accepts and returns on\_task\_error. (`GET /api/flows`, `GET /api/flows/{flow_id}`, `GET /api/flows/{flow_id}/versions`, and 6 more)
* PATCH /api/profile/me accepts the new hub\_sort values and the contains\_all tag filter mode, so you can save the new Hub sort and filter preferences through the API. (`PATCH /api/profile/me`)
* POST /api/flow-executions accepts the optional Idempotency-Key header, returns the key in effect as idempotency\_key, and can respond with 409 in the rare case that a concurrent duplicate request can't be read back. Flow execution responses include the new idempotency\_key field. (`POST /api/flow-executions`, `GET /api/flow-executions`, `GET /api/flow-executions/{flow_execution_id}`) → [Error Codes](https://docs.flows.super.ai/api-reference/errors.md)

→ [API reference](https://docs.flows.super.ai/api-reference/introduction.md)

**October 16, 2026**

New models and smoother review

## Gemini 4 Argon, automatic model failover and a faster My Work

Gemini 4 Argon joins the model list in both regions, My Work, the review screens and the Data tab are quicker and easier to work in.

**New**

* Gemini 4 Argon (gemini-4-argon) is available as a selectable worker in both the EU and global regions. It supports bounding boxes, processes large documents without splitting them, and offers the reasoning efforts low, medium and high.

**Improved**

* In My Work's multi-task review, each task header now opens or closes that task on its own, so you can keep any number of tasks open at once. One open task fills the stage; with several open, drag the grip on a row's bottom border to change its height, or double-click it to reset. Your heights are remembered until you reset the layout.
* If a document doesn't load while you review a task, a Retry button reloads it on the spot, in structured extraction tasks, the document viewer's PDF and image previews, and the classification page tiles. A document that failed to load also recovers by itself when you come back to the window, and the PDF preview keeps your page.
* On a flow's Data tab you can now jump straight to any page instead of stepping through with first, previous, next and last, and the table stays responsive while runs update live. If the table can't load, you see an error with Try again instead of waiting.
* My Work, its sidebar badge and work mode load noticeably faster, especially for organizations with many runs. The selection count always matches the selected rows, changing Items per page shows a loading state while the table reloads, and returning from work mode no longer briefly shows an empty queue.
* In the flow editor, a task added from the Task library lands centred in view above the existing tasks with nothing dimmed, and the canvas is no longer left dimmed after you delete the selected task. The parameter editor keeps focus and its # reference menu while you type, and reference chips update in place when a task is renamed.

**Fixed**

* Text to Structured tasks that use the model-based bounding-box method now return bounding boxes for longer multi-page documents.

**API**

* Saving a flow definition, updating a task and restoring a flow version now respond with 409 when the save conflicts with a newer version of the flow. Concurrent saves to the same flow are applied one at a time, so re-fetch the flow and retry on a 409. (`PUT /api/flows/{flow_id}/definition`, `PUT /api/flows/{flow_id}/task`, `POST /api/flows/{flow_id}/versions/{version}/restore`) → [Error Codes](https://docs.flows.super.ai/api-reference/errors.md)

→ [API reference](https://docs.flows.super.ai/api-reference/introduction.md)

**October 13, 2026**

Chat with long documents

## Chat reads long documents reliably without repeat charges

Chat now handles long documents: reading them no longer fails, paging through them no longer bills the whole document again, and the assistant stops repeating an action that cannot succeed. Gemini 3.8 Flash is available as a model and is the new default for chat document processing.

**New**

* Gemini 3.8 Flash is available as a model in both the EU and global regions, with low, medium and high reasoning levels. It is now the default model when chat reads, extracts from or classifies documents.

**Improved**

* Tasks you add to a flow no longer appear automatically in your existing custom views. The default Runs view still shows them, so you choose where new columns belong.

**Fixed**

* Chat no longer fails when you upload a long document such as a PDF with more than a hundred pages. When a step can't succeed, the assistant now explains why instead of repeating the same attempt and charging you for each one.
* Paging through a large document in chat is no longer charged as a full read of the document each time. Once a document has been read, later pages come from the text already retrieved.

**Week of October 12, 2026**

In-app support and bulk review

## Contact support from inside the app and edit review tables in bulk

You can now reach support from the sidebar or right from a chat or task that needs attention, with the details our team needs attached. Review tables let you select many cells or rows and clear or delete them at once.

**New**

* Contact support without leaving the app. Use Contact support in the sidebar, or the link that appears beside a chat reply, document task or error message that needs attention. Pick a category, write your message, and the details our team needs are attached automatically. An Attached for our team box shows exactly what is sent.
* Select cells and rows in review tables, then right-click to clear or delete them in one step. When an extraction picks up extra line items, you can remove them together instead of one row at a time.

**Improved**

* The Data context in the search panel now searches the last 30 days by default, so results come back faster and stay focused on recent work. Switch to Last 7 days, Last 90 days or All time when you need a different range.
* When a document can't be processed, the task now tells you exactly why, for example a password-protected PDF, a corrupt file or an image that's too large, so you know right away what to change. Doc to Structured, Classifier and Image Rotation also accept more formats, including HEIC, AVIF, WebP, TIFF, SVG and EPUB, and images keep their full resolution.
* The MCP server now offers your AI agent exactly the operations in the public API reference. The tools an agent discovers match the documentation, which makes agent setups more predictable. → [MCP Server](https://docs.flows.super.ai/api-reference/mcp-server.md)

**Fixed**

* My Work now displays review tasks whose target is another review step.
* Deleting a flow from the hub keeps you on the hub and removes the flow from the list.

**Week of October 5, 2026**

Faster review and run control

## Review tables inline, see flagged validations, and cancel runs from My Work

Human Review gets inline tables with flexible split layouts and can highlight exactly which field, row or page a validation flagged. You can also cancel and re-run work straight from My Work, reference stored credentials securely in Python tasks, and choose Claude Opus 5.5.

**New**

* Table extractions now appear inline in the review extraction list, in the same order as your flow's fields, with a preview of the first rows and a See all option to expand in place. Split icons on each table open it in a resizable bottom dock or a side column next to the document, so you can work on wide tables without losing your place. Your layout, dock height and column width are remembered for you.
* Human Review now highlights the exact field, table row or cell, document split, document type or page that a validation flagged, with the validation's message shown right next to it. Add the validations you want surfaced to your Human Review task and the highlight appears on the reviewed data, then clears once the reviewer commits a change, so it's clear at a glance what still needs attention.
* Select runs in My Work to re-run or cancel them together, and cancel a single run from the task screen next to Re-run. Cancelling asks for confirmation first, and after a cancel, work mode moves on to your next item.
* Reference your organization's stored credentials directly in Python task code: type # in the code editor and pick a credential from Resources to insert it. The code only stores a reference, never the secret itself, values are masked in error messages, and an edited credential applies from the next run.
* Claude Opus 5.5 is now available as a model for your tasks, with EU and global processing options and all reasoning effort levels.

**Improved**

* The model list adds Kimi K3, Qwen 3.8 Max, GLM 5.3 and GLM 5.3 Flash, replacing models the provider has retired; flows that used a retired model keep running on its replacement. Image tasks now offer only models that can read images, and saving a text-only model on one prompts you to choose a vision model.
* The document viewer now fits its controls into a single toolbar row, and page navigation moves to a compact control at the bottom of the document with an optional thumbnail strip below the page. Your document gets noticeably more room while you review, and your strip preference is remembered.
* The Data table now remembers your column widths, and default widths fit the column headers. The Input and Output divider in task details keeps the position you set on each device.
* Clicking an Error, Cancelled, Skipped or pending cell in the Data table now opens the cell card, the same as cells with output. Cells for tasks that were added after a run finished now read Task not part of current flow version.
* Task errors are easier to read and act on: long errors scroll within their panel and keep their line breaks, Python task errors show a short, relevant traceback, and tasks explain clearly when a file can't be downloaded or a model's response reached its output limit.
* Document classification is more accurate where one document ends and the next begins in a multi-document file, so pages at those boundaries are assigned to the right document type more reliably.
* Flow canvas groups now have their own floating-tab style with a task count and a menu to rename, recolour or remove the group, so they're easy to tell apart from notes. Note colours now live in each note's menu and apply instantly, and tasks stay in their groups when you save.
* Users with the worker role can now delete runs from the selection bar.

**Fixed**

* Emails are now routed to a task when its address is in Cc, the same as To.
* Each click on a download now starts exactly one file, and the confirmation says Download started or File downloaded depending on what your browser reports.

**Week of September 28, 2026**

Clearer runs and reviews

## Shape your My Work table and get alerted when reviews are waiting

You can now customize the My Work table column by column for each flow, and get notified when a run is waiting on a human review. Run statuses also explain themselves, so you can see what a run is waiting for and why it stopped.

**New**

* Customize the My Work table for each flow. Hide and reorder columns from the new Customize columns drawer, freeze a column from its header menu, and drag a column's edge to resize it. Your layout is saved for you per flow, so each queue shows the fields you work with most.
* Flow notifications have a new Human review pending trigger. Turn it on in a flow's Notifications settings to get an email or Inbox alert when a run is waiting on a human review; the email links straight to My Work. If the review has already been opened or completed, or the run has finished, the alert is skipped, so you only hear about reviews that still need you.

**Improved**

* Run statuses now tell you more. A waiting run shows which task it is waiting for, with consistent wording across the app, and hovering a failed or cancelled run explains what happened, with quick actions to review or re-run. Failed tasks and task columns show the same detail on hover, and finished task results in stopped runs now show while they load.
* Deleting a flow now also stops its runs in progress and removes its pending reviews from your Inbox. If you restore a flow from the trash, its scheduled, email and file-drop triggers are set up again, so it picks up new work just as before.
* Connecting lines on the flow canvas now stay visible above groups and sticky notes, so you can follow a flow at a glance. Sticky notes have a refreshed look that matches the rest of the app.
* Every download in the app now shows a confirmation once the file is saved, so you know right away that it worked.
* References in the flow editor share one consistent look. References to a whole collection carry a database icon, and hovering any reference shows which kind it is.

**Fixed**

* A task condition that checks a sibling task is now evaluated for each item on its own, so every item runs or skips based on its own result.
* Filling an extraction field by clicking a word on the document in a review now saves as expected.

**Week of September 21, 2026**

Shareable views, faster review

## Share filtered views by link and read every sheet of your Excel files

Filters and sort on any view now live in the link, so you can bookmark or share exactly what you see, and save them to a view when you choose. Excel workbooks are now read sheet by sheet, and the canvas, review screen and document classification all get faster and smoother.

**New**

* Filters and sort on every view, including your saved views, are now part of the page link, so you can bookmark a filtered table or send it to a teammate. Saving filters and sort to a view is now an explicit action, and every member except publishers can create, rename, duplicate and delete custom views and set their own default view. Your runs-per-page setting also stays put when you switch views or come back from a task.

**Improved**

* Excel workbooks are now read sheet by sheet, so data on the second or later sheets, common in bid forms, rate cards and volume files, reaches your extractions. In review, the Excel preview lets you move between every sheet, and spreadsheets process faster.
* Very long tables, such as multi-page statements, now extract reliably: when a table is too large to extract in one pass, it is automatically split by page range and the rows are merged back in page order. This works for both text-based and image-based extraction.
* The flow canvas opens faster on large flows, starting on the flow's input task. Jumping to a task from search always lands at the same comfortable zoom, travels at a steady pace, and selects the task you landed on.
* The review screen is cleaner: task fields are always expanded, every AI hint now tells you what it shows on hover and opens on click, and the controls to hide or move extractions now sit in the extraction panel. Field instructions open from the info icon, so clicking a table field's name opens the table, and tags and Edited or Stale badges appear as soon as you save.
* When you work through a review queue, items that are already completed, cancelled or being worked on by someone else are now skipped automatically, so you go straight to the next item you can work on. A short note tells you how many items were skipped, and a new settings menu next to Exit lets you choose which kinds of items to skip.
* Receive Email now accepts attachments based on their actual content rather than their file name, so a PDF, image or Excel file without the expected extension, or without a name at all, is still processed.
* Document classification that splits a document into parts now checks page continuity in parallel, so long documents are classified several times faster.
* The AI validation rule in Item Validation and Collection Validation now offers the same up-to-date model list as the rest of your flow, so newly available models show up there too.

**Fixed**

* Each item in a run now moves through the rest of the flow independently, so one item that needs attention no longer holds back the others.

**Week of September 14, 2026**

Faster, clearer document review

## Review documents your way, with field instructions one click away

The review screen gets a flexible layout, keyboard navigation and full field instructions inside Task details, and bounding boxes now work on rotated pages. Doc-to-Structured adds Mistral Document AI, extraction results trace back to the page of the original file, and Python code tasks can read the flow's own documents.

**New**

* Recent flows now appear in the side panel under Flows, with the flow you have open highlighted. Click one to jump straight to its data, or to the designer if it has no runs yet, without going back to the hub. The collapsed sidebar offers the same list from a menu.
* Mistral Document AI is now available as a model in Doc-to-Structured tasks, billed per page. Doc-to-Text in markdown mode can now return bounding boxes and confidence scores for each block of text, and new Doc-to-Text tasks start in markdown mode with bounding boxes switched on. If a Doc-to-Structured task's model can't extract bounding boxes, the designer tells you when you save the flow.
* Python code tasks can now read the flow's own documents through declared file inputs. You can handle what the built-in tasks don't cover, such as reading every sheet of a workbook or the text of a Word file, without preprocessing outside the platform.
* When a document is split by a classifier, extraction results now record the page number in the original file and a link to that file. You can trace every extracted value back to the exact page of the document you uploaded.

**Improved**

* The review screen adapts to how you work. Move the extraction list to the left or right of the document, collapse the header to give the document more room, and use the arrow keys to move through your work queue and between tasks. Long breadcrumbs now stay on one line and fold gracefully.
* Field instructions are easier to reach while labelling. Click a field's label to see its instructions, and open long instructions in full in the new Field instructions section of Task details, which scrolls straight to that field. Instructions load only when you open them, so work items with large inputs open quickly.
* Abstracted values now stand out at a glance in the labelling view, and the AI badges inside a value open their explanation with a click instead of a hover, so the panel stays calm while you scan it.
* Bounding boxes stay visible and line up correctly when you rotate a document in review, and you can draw new boxes at any angle, including on touch screens. Rotated pages also fit the viewer width properly.
* You can set a split's document type directly from the classifier's expanded page preview, where the pages are large enough to read. The preview stays open even when a type filter is active.
* My Work can now be sorted by Run created at, so items keep their place in the queue by the date the run started, even after a re-run.
* The Flows hub is tidier. Select flows in the list view and Delete now sits in the list header, and with one flow selected an Actions menu offers Rename, Duplicate and Copy flow ID. Template cards now share the look of the flow cards below them.
* The Data view table feels steadier. A new run's row keeps its height while results fill in, an expanded cell scrolls smoothly into view, and the running indicator and sort button are easier to read.
* Searching flow data is faster, including on flows with many runs. The search panel now asks for at least 3 characters before it searches, and it tells you when a search is taking longer than usual.
* Search in the flow designer now zooms to the exact parameter that matched and names list items by their label, for example Document Type > Pay Stub > Instructions. Long field labels truncate with a tooltip, and expanded list items close with a clear Done button.
* The flow designer catches more setup problems before a run. A Write to SFTP task without a file is flagged at design time, and the Use Logprobs Confidence setting can be switched off when the selected model doesn't support it.
* Large page images are now sized to fit each model's limits automatically, so high-resolution scans process smoothly in document and chat tasks.
* Gemini 3.1 Flash Lite is now available in its generally available version, and GPT-5.4, GPT-5.4 Mini and GPT-5.4 Nano show clear names in model dropdowns.
* Re-running a run now happens in the background and is retried automatically until it completes, so you can carry on working while it restarts.
* You can now delete notes you've written on a run.
* Chat, the chats list, My Profile and Organization settings now fill the whole screen on phones.
* The SharePoint receive task now lists every site in your organization when you choose where to receive documents from.

**Fixed**

* Doc-to-Structured tasks now use your Additional instructions when extracting.
* Tasks connected to another task's Execution Error output now run as configured, so your error notifications are always sent.

> [!WARNING]
>
> **Breaking API changes**
>
> * Re-executing a flow execution now returns 202 Accepted instead of 204 and runs in the background, with a Location header pointing at the flow execution. A re-run requested while one is already in progress joins it instead of returning 409, and a 503 means nothing was changed and the request can be retried. **Migration:** Treat 202 as success and follow the flow execution's status, which moves to queued, then running. Retry on 503.

→ [API reference](https://docs.flows.super.ai/api-reference/introduction.md)

**Week of September 7, 2026**

Faster, clearer document review

## See the AI's reasoning behind extracted values and add notes to runs

Extraction review can now show why the AI chose each value and the instruction each field was given, and you can keep notes on runs and find them, along with run metadata, in search. Reviewing documents, Classify tasks and My Work queues also takes fewer clicks.

**New**

* Extraction tasks can now explain themselves. Turn on reasoning for a Doc to Structured or Text to Structured task and each extracted value comes with a sentence or two on where it was found and how it was derived, shown as a badge inside the field during review. Fields that have an instruction also show an info icon with that instruction, so reviewers no longer need the instruction document open in another tab.
* Add notes to a task run from the new Notes tab in task details. Notes form a running log that only their author can edit. The new Metadata search context finds runs by these notes and by job metadata, and data search result counts now match the results shown.
* Page through a Classify task without leaving the document preview. Arrows and the left and right keys move between pages within a split, the up and down keys move between splits, and a split selector and Actions menu sit above the page. Your zoom level stays the same as you turn pages. Splits are grouped by document class and sorted within each class, so you can work through one class at a time, and the Group Splits setting is remembered.

**Improved**

* Document review is easier to navigate. Drag to pan the inline document viewer, and documents now open at a readable fit-to-width zoom. Clicking a bounding box scrolls its field into view, and clicking a field scrolls the document to its box in both directions. Task execution sections in the extraction list can be collapsed, the document type dropdown opens on the current selection, and wide tables scroll inside their panel instead of pushing it off screen.
* My Work keeps your place. When you leave a work item, the list comes back with the filters and sort you had set. After you complete work, the All done dialog takes you back to My Work while new items load in. The Inbox My Work tab adds Collapse all and Expand all for work groups.
* Write to SFTP can now write task output values as file content. Text, JSON objects and lists become the file's contents, large values are written in full, and file names are generated for you unless you set one. Templated destination paths now resolve to the folder shown in review, and review details show the actual values and paths instead of template text.
* A Wait for Webhook task that runs once per item, for example once per page, can now receive data for each item. Set the new optional task\_execution\_idx field in the webhook data request to address a specific run; each run waits for its own data and continues as soon as it arrives.
* The flow detail pages and work mode now share a cleaner single-card layout.

**Fixed**

* Re-running only stale tasks now also re-runs tasks whose settings changed in the flow since the original run.
* Doc to Structured tasks now handle long documents by processing their pages in batches, for documents of up to 150 pages.

**Week of August 31, 2026**

Faster, clearer document review

## Mark abstracted values, fill down extractions and zoom documents in chat

Reviewing extracted data gets faster: flag values that were deduced rather than copied, fill a whole column in one drag, and zoom, rotate and pan PDFs and images in chat and review previews. Review screens, sharing and the app's tabs and buttons also got a cleaner, more consistent design.

**New**

* You can now mark an extracted field value as abstracted, meaning it was deduced or calculated from the document (such as a total summed from line items) rather than copied word for word. Use the Abstracted option in the value's ⋯ menu during review, and a small A badge shows which values are flagged. Doc to Structured and Text to Structured tasks can also pre-fill this flag for you when you turn on their new Extract Abstracted setting.
* The extraction table now supports fill-down and drag-to-fill. Copy a value into the cells below it in one action instead of editing each row, which saves time when many rows share the same answer.
* Admins can now resend an invitation to a member who hasn't joined yet, straight from the members list. If an invitation link can't be verified, the member is now shown how to get a fresh one instead of a dead end.

**Improved**

* PDFs and images opened in chat now have full viewer controls: zoom with buttons, keyboard or pinch, rotate and tilt, and reset to fit. When you're zoomed in, you can drag to pan around the page in chat and in the review document previews, and you can type a page number to jump straight to it.
* The expanded review screen is easier to work in: controls are clearly labelled, and the document preview opens in place instead of in a second window. Validation notes for reviewers now appear in a clear banner you can step through, with the full list in the task details panel, and the setting that produces them has a readable name in the flow editor.
* When you filter or sort the Data table, the previous and next run buttons in task detail now follow the same filters and order. You step only through the runs you're actually reviewing.
* Sharing now uses one type-ahead picker for both members and worker groups, so you find and add the right people faster. The wording says Add rather than Invite, and the option to create a new user sits right under the search box.
* Tabs and buttons look and behave the same across the app. Main navigation tabs use one pill style with clear count badges and can be used from the keyboard, and buttons follow a clear primary, secondary and tertiary hierarchy, so it's easier to see which action matters most on each screen.
* Notices take up less room: the upgrade prompt now sits in a card in the sidebar below your credits, and privacy and consent notices appear as compact cards in the bottom-right corner. The Current credit usage card always shows a button to your plan options.
* Changing your password now asks for your current password first, which keeps your account safe on shared or unattended devices.

**Fixed**

* Downloading files from chat cards works smoothly on iPhone and iPad, and you'll see a message if a download can't be completed.
* Bulk actions on many selected items, such as re-running, cancelling or deleting runs, now run more smoothly. If anything needs attention you'll see one clear message with a Retry option for just those items.

**Week of August 24, 2026**

Faster, roomier document review

## Review documents faster with near-instant page loads and a focused review view

Multi-page documents now open almost instantly, and reviewing several task outputs is easier with a near-fullscreen view, reachable zoom and rotation, and steadier scrolling. This entry also adds choice fields for extraction, HEIC photo support, a barcode reader task and new AI models.

**New**

* Extraction fields can now be single choice or multiple choice. In document-to-structured and text-to-structured tasks, you define the allowed options, the AI picks from them, and reviewers choose from the same list, so values stay consistent across documents.
* HEIC and HEIF images, the default photo format on iPhones, can now be uploaded and processed like any other image. Phone photos of documents go straight into your flows without converting them first.
* A new Barcode reader task decodes the barcode on the back of US driver's licenses into structured fields such as name, date of birth, address, license number and issue and expiration dates. Documents without a readable barcode simply return no result, so the task never stops your flow.
* Claude Fable 5, GPT 5.6 (Sol, Terra and Luna) and Gemini 3.6 Flash are now available as AI workers in the global region. Pick them in the flow builder wherever you choose an AI worker for a task.

**Improved**

* Multi-page documents in review now show their first page almost instantly, however many pages they have. Zoom and rotation are always visible in the review toolbar, even in a side-by-side pane, and you can scroll a zoomed or rotated page all the way to its edges.
* When a review in My Work has several task outputs, you can expand any one of them into a near-fullscreen view and step between outputs with the arrows. On the page itself, one output is shown at full height while the others collapse to headers, and you can pin more than one to view them together.
* The classify screen in My Work shows document types in full, lets you zoom and rotate the page preview, and puts the best match first when you search for a document type.
* Sort the My Work queue by source task name, A to Z or Z to A, to work through reviews of the same kind together. Your sort choice is kept in the link, so it survives a reload and can be shared.
* On a flow's Data tab, clicking a cell opens its details right below the highlighted cell, so neighboring values stay visible for comparison. The whole cell is now one click target, and you can drag the card's edges or corner to resize it and read longer values.
* Tables in Organization settings and Resources share one cleaner layout: search, sort, filters and actions sit in a single row at every screen width, and primary actions are easier to spot. Organization settings tables are now paginated, with 10, 25 or 50 rows per page.
* Signing in is simpler: after you enter your email, a single continue screen shows how you can sign in, and the reset password link takes you straight to setting a new password. Enterprise single sign-on now explains why it could not start instead of showing a generic error.
* The flow builder now uses the same names on every AI task: the model setting is labeled "AI worker" and prompts are labeled "Instructions" (or "Additional instructions" where a task also has per-field instructions), whether you configure OCR, extraction, classification, an agent or validation.
* The web app and the API respond faster and more consistently. Search and database integration checks no longer slow down other requests, and fewer requests wait on new database connections.

**Fixed**

* Completed reviews keep their status when a flow is re-run. If a run is re-run while you have its review open, My Work tells you and lets you check again or move on to the next item.
* MCP clients now stay connected across requests, and the MCP endpoint follows the MCP transport specification for streaming requests. → [MCP Server](https://docs.flows.super.ai/api-reference/mcp-server.md)

**Week of August 17, 2026**

Organized chat and filters

## Find files and tasks in chat, and manage tags across a whole flow

Chat conversations get an Objects panel that gathers every file and task in one place, and flows gain a Tags tab for reviewing and cleaning up tags. Filtering in the Data view and My Work is more precise, and chat task review now works well on phones. Enterprise admins can also claim their email domains so new colleagues join the right organization.

**New**

* Open the Objects panel in a chat conversation to see every file and task in one docked side panel, split into Files and Tasks tabs with search and counts. Preview or download files one at a time or all at once, and jump straight to a related task, so you no longer have to scroll back through the conversation to find what you shared or produced.
* A new Tags tab in flow settings lists every tag used in the flow, with how many runs carry it, a per-task breakdown and when it was first applied. Search and sort the list, open the matching runs in the Data tab, and delete tags you no longer need across the whole flow, one at a time or in bulk.
* Document and text classification tasks now give a confidence score for each page and each split when you turn on the confidence option. Use these scores to send low-confidence classifications to review. The option is available for models that support confidence scoring.
* Every trigger task now exposes the run's execution ID as an output, so you can reference it in any later task in the flow. This lets you include the run ID in custom payloads, such as a request your flow sends to another system.
* Link straight to a review task in My Work using values your webhook already delivers: the flow execution ID, the task name and the task execution index. Your own systems can now send reviewers directly to the right work item.
* Organization admins can now restrict AI processing to the EU from the new Data residency section in organization settings. When it is on, chat and flow model calls use EU-hosted models only. Your profile page also has a new Privacy section where you can manage product news emails, product analytics and automated analysis of your usage.
* Enterprise organization admins can claim their company's email domains under Organization settings › Members › Settings › Email domains. New sign-ups from a claimed domain arrive as join requests in the new Join requests tab, and admins are notified so they can add the person to the organization.

**Improved**

* Chat task cards and the task review window now fit phone screens. Panels stack vertically, task actions become compact icons, and extraction review opens on the fields first with a tap to show the document and a simple zoom menu.
* The Data view tag filter can now match runs that contain all of the selected tags, and you can combine it with a do not contain filter. The Metadata filter is now available in the Data view too, and expanding the input task's cell shows the run's metadata under the content, so you can see what you are filtering on.
* My Work keeps your filters, sort and page in the URL, so reloading or sharing a link brings back the same view. Tag filters in My Work now support both contain all of and do not contain.
* When you select rows in the runs table, My Work, Flow Hub or Worker Groups, the action bar now appears in the same place on every screen, just above the pagination, and never covers your last rows.
* The task card's Actions menu is now called Export, and flow sharing and review task assignment now use the term worker groups, matching organization settings. The chat assistant uses the same names as the interface, so its directions match what you see on screen.
* Signing up is smoother. If you are already signed in, starting from the website takes you straight into the app, and if a sign-in provider such as Microsoft does not share your email address, you now see a clear explanation of what to do.

**Fixed**

* When you draw a box around a value during review, you can always type the value yourself, and you are prompted to do so if the text in that area can't be read automatically.
* Importing a flow definition from JSON now opens the updated flow as expected.

**Week of August 10, 2026**

Free trial and run alerts

## Start with free trial credits and sign up with any email address

New accounts now start with free trial credits and can sign up with personal or academic email addresses. Runs are more dependable too: error notifications now cover individual task failures, runs started while you're out of credits are saved and start automatically, and Wait for Webhook tasks can time out on your terms.

**New**

* New accounts now start on a free trial with one-time trial credits, and you can sign up with a personal or academic email address, not only a work address. When your trial credits run out, everything you've built stays available and the app shows you how to upgrade or move to the free tier.
* Error notifications now cover individual tasks too: with the Error state notification turned on for a flow, you're alerted when any task in a run fails, even if the run itself completes. Each run alerts you once, and the alert clears when a successful re-run resolves it.
* When your organization is out of credits, new runs are now saved as Pending — credit required instead of being declined. They start automatically once credits are available again, so runs triggered by SFTP, email, webhooks or SharePoint are picked up without any action from you.
* Wait for Webhook tasks can now have a timeout between 60 seconds and 48 hours, and you choose what happens when it expires: fail the task or skip it and continue. Without a timeout, the task keeps waiting as before.

**Improved**

* When the assistant in chat changes your settings, the changes now appear inline as clear, key-by-key rows instead of in a separate dialog. Nested values, lists, additions, removals and renames are labelled and indented, and long values are shortened with the full text available on hover.
* The chat composer now shows a character counter as you approach the 8,000-character message limit and disables Send once you go over, so you can shorten your message before sending. If a message can't be sent, your draft stays in the composer.
* Invitation links from email now open a Continue step, so you're signed in only when you click, and links checked by email security tools stay valid for you. Expired or invalid links show a clear explanation.
* Organization settings, your profile and billing now work well on a phone. Members, service accounts and worker groups appear as easy-to-read cards, Compare plans shows one plan at a time with a switcher, and the settings tabs keep the active tab in view.
* GPT-5.2 is now available when you choose the EU region, so you can use it while keeping processing in the EU.
* In organizations that use Chat only, invited members, not just admins, now land in Chat when they sign in, with Chat navigation available across the app.

**Fixed**

* When a request refers to a document you haven't attached, such as a starter template prompt, the assistant now asks you to attach it.
* The document pane now displays correctly while you draw bounding boxes during review, and the drawing controls appear when you hover over or focus a table cell.

**Week of August 3, 2026**

Smarter chat and review

## Fine-tune task models and settings right from chat

You can now fine-tune the AI model, region, reasoning effort and other advanced options for tasks in chat, and review Wait for Webhook tasks on a readable details screen. Runs are easier to find and share with copyable Run IDs, and tags now open the exact task execution.

**New**

* Tasks in chat now have an Advanced settings section where you can choose the AI model, region and reasoning effort, plus confidence scores and bounding boxes for tasks that support them. Options the selected model doesn't support are switched off automatically. You can also switch a task's model from a dropdown on its tile, and the tile shows your saved settings straight away.
* Wait for Webhook tasks have a human-readable details screen. You see the payload you expected next to the one you received, field by field with types and missing values, in resizable panels, plus a clear waiting state while the task is still waiting for data.
* When you start signing up or signing in on super.ai, you can now open the emailed link on another device, such as your phone. Check that the short code matches, confirm, and your session continues in the window where you started, with the prompt and files you added still there.

**Improved**

* Runs are now identified by a short Run ID that you can copy in full with one click, in flow tables, navigation and task review. This makes it easy to share an exact run with a colleague or look it up later.
* Clicking a tag in the Runs view now opens the exact task execution the tag belongs to, so you land right on the result you tagged.
* In review, locating a table row on the document now uses a map-pin icon that tells you which page the row is on, and rows are tinted to show whether they're on the page you're viewing. Locating a row scrolls only vertically, so the document stays where you panned it. If a task's file type setting is invalid, the review screen explains why the preview is unavailable and still lets you download the file.
* Settings changes and output edits in a chat conversation now show as clear before-and-after rows with readable model names. For long or structured values, select View changes to see a detailed diff, and click the document name to preview it in place.
* When a file can't be added to chat, the message now tells you exactly why. An oversized file shows its size next to the 8 MB limit, and an unsupported file lists the formats chat accepts.
* Long structured results that reach the model's output limit now continue automatically instead of starting over. Extractions from large documents complete more often and faster.
* Requesting a fresh invitation from an expired invitation link is now a single click. You no longer need to type your email address.

**Fixed**

* Task references used in the instructions of extraction fields, table columns, classifier document types and agent outputs are now replaced with their values when the task runs.
* Email-triggered flows that stay paused for more than a day, for example while waiting for human review, now resume as expected.

> [!WARNING]
>
> **Breaking API changes**
>
> * The run\_number field and the include\_run\_number query parameter are removed from the list and get flow execution endpoints. Requests that still send the parameter succeed, without the field. **Migration:** Identify runs by their id, which the app now shows as a copyable Run ID.
> * Resending an invitation now takes the invite\_token from the invitation link instead of an email address. The endpoint always returns the same acknowledgement and can return 429 when called too often. **Migration:** Send the invite\_token from the invitation link in the request body, and retry after a pause on 429.

→ [API reference](https://docs.flows.super.ai/api-reference/introduction.md)

**Week of July 27, 2026**

Clearer document review

## See extracted fields highlighted right on the source document

Check extracted data against the source document with highlights, fill repeated values in merged results in one drag, and stop a chat reply mid-turn. The chat app now also works comfortably on your phone.

**New**

* In chat, select View highlights on an extraction result to see the extracted fields and table rows outlined on the original document, in a full-screen view with page navigation. When you review an extracted table, choose Show in document on a row (or its locate icon) to go to its page and highlight it. You can also click a highlight in the document to select the matching row.
* When you review merged results in the spreadsheet grid, right-click a cell and choose Fill down or Fill up to copy its value into the empty cells below or above. You can also drag the fill handle to copy a value across a range. Pressing Enter now confirms an edit in review table cells and fields.
* Use the new Stop control in chat to end a reply that is still in progress without leaving the conversation. Whatever was already written is kept and marked as stopped, and you can send your next message right away.
* Use the chat app, navigation and sign-in comfortably from your phone. A slide-out menu and a New chat button in the header make it easy to move around. Starter prompts and industry tabs scroll sideways, so the message box stays in focus on small screens.

**Improved**

* When you change a task's settings or correct its output in chat, the assistant now shows what changed and asks what you'd like to do next: re-run this document, re-run all documents, or keep the change. Several edits made together become a single question, and corrected outputs appear in the conversation as Output edited with the changed fields.
* Searching a flow now highlights matching fields inside task builders, including nested fields. Open a builder from a search result and the matching row is brought into focus, so you find the setting you were looking for straight away.
* The Data table of a flow now opens noticeably faster on flows with many runs, and filtering runs by tag, task name or status responds much more quickly.
* Flow notification settings now have an explicit Save button, so changes apply only when you confirm them. The Notifications tab in Flow Settings is shown only to people who are allowed to edit it.
* The redesigned billing page groups credit volumes by plan, marks your current and recommended volume, and shows the discount on annual billing. Plan cards compare Free, Growth and Enterprise side by side, with a Current plan tag and clear actions such as Upgrade, Reactivate plan or Downgrade to Free.
* SFTP tasks now retry automatically when the server is briefly busy or slow to accept a connection, and give slower servers more time to respond. Sign-in errors are reported right away so you can correct the credentials.
* The sidebar and chat have a cleaner look, with clearer hover and selected states and refreshed starter prompt cards. New chats get short, label-style titles of a few words.

**Week of July 20, 2026**

Chat for everyone

## Chat is now available to every eligible user, with a new sign-in

Chat is now generally available, with a redesigned sidebar, starter prompts, automatic conversation names and support for text, CSV, Markdown and JSON files. Signing in is simpler with one email-first screen, and a new Usage tab shows your credit consumption at a glance.

**New**

* Chat is now available to everyone whose role and plan include it, with no extra setup. If your account qualifies, Chat appears in the navigation and is where you land after signing in.
* Sign in and sign up from a single screen. Enter your email and the app offers the right option for your account: single sign-on, password, a magic link, or creating a password. If you cancel a sign-in with an external provider, you return to the screen you started from.
* Upload plain text, CSV, Markdown and JSON files in chat. They are read directly without scanning, previewed with formatted JSON and CSV tables, and recognised by their file extension even when the file type is missing.
* A new Usage tab in your organization details shows your monthly and additional credit usage, and you can buy more credits or upgrade your plan right from there. The Credits link in the sidebar now opens it, and the Billing page focuses on plans and subscription details.
* New conversations name themselves from your first message and any attached file names, so your chat list is easy to scan. The name appears in the header and the sidebar while the first answer is still being prepared.
* Start on super.ai: describe your task and attach files on our website, and after you sign up your first chat opens with them already sent, so you pick up exactly where you left off.

**Improved**

* The sidebar has a cleaner layout with quick buttons to start a new chat or browse all your chats, and a collapsible Recent list that opens on hover when the sidebar is collapsed. Long chat, flow and account names now show in full in a tooltip, and the chat header stays hidden on a new chat and floats over the conversation as you scroll.
* Starting a new chat is quicker: pick your industry to browse starter prompts shown as cards, and select one to fill the message box. Previous conversations open at the latest message with their history in order.
* Data extraction in chat now reads a document's text once and reuses it, so you can ask for more fields without the document being processed again. Text extraction appears as a Data Extraction step, and corrections you make to extracted values stay consistent with the summary.
* Classification results in chat can now be copied or downloaded as JSON or CSV, including the document type, confidence, split details and page ranges. When a document is split into parts, each part's results are grouped under the original document so you can follow them at a glance.

**Fixed**

* CSV files generated by the Format Data step can be downloaded, previewed and delivered to SFTP. → [Downloading Files](https://docs.flows.super.ai/guides/downloading-files.md)

**Week of July 13, 2026**

Webhooks, run insights and annual billing

## Pause flows for external data and see why tasks ran

Flows gain a Wait for Webhook task that pauses a run until an external system sends data, and the Run Graph now explains why each task did or didn't run. Monthly Growth subscribers can switch to annual billing in-app, My Work updates live, Python code tasks can process images and PDFs, and voice input works in the agent chat.

**New**

* Dictate messages to the AI agent: press the microphone button in the agent chat on the Hub or in the flow designer to start and stop recording. Your speech is transcribed into the message box so you can review it before sending.
* Add a Wait for Webhook task to pause a flow until an external system sends data. The flow editor shows the URL to submit to and lets you define the expected payload, and the run resumes automatically once the data arrives.
* Monthly Growth subscribers can switch to an annual plan directly from the billing page. A confirmation shows what you will be charged before you upgrade, and a move to a plan with fewer credits is scheduled for the end of your current period instead of charged right away.

**Improved**

* The Run Graph now explains why each task did or didn't run. Edges show how many items flowed down each path, tasks that routed no data are marked 'No path taken', and a skipped task names the task that caused it, with a button to jump straight to it. A new View Task Design button opens any task in the flow editor.
* My Work and review screens update live as human review assignments are created, started, completed or cancelled, so you always see the current list without refreshing. Assignments made to a worker group appear for every member right away.
* Full filenames are now visible across Runs, My Work, data tables and file previews: click a truncated name to see and copy the whole name, and runs with several files list one per row. Data table cells open in a click-to-expand preview for easier reading.
* Search now finds matches hidden inside collapsed fields. Collapsed list items that contain a match are highlighted and expand when you jump to them, and the Field Extractor and Output Builder dialogs open with every row expanded. Jumping to a match in the data table scrolls to and highlights the right cell.
* The flow editor flags invalid human review assignees and unavailable AI models while you design, so you can fix them before running the flow.
* Python code tasks can now use Pillow, OpenCV and PyMuPDF for image and PDF processing, and more of the standard library works as expected, such as parsing dates. Unsupported modules return a clear error.
* The SSO setup dialog shows the service provider details your identity provider needs, including the entity ID, metadata URL, ACS URL and logout URL, as copyable fields, even before SSO is configured.
* The billing page puts upgrading front and center, with a plan configurator beside the plan cards and annual pricing shown by default. Page estimates now reflect typical usage (the Free plan shows 300 pages per month), and the upgrade banner can be dismissed, leaving a compact upgrade card in the sidebar.

**Week of July 6, 2026**

Search for everyone

## Search across your flows and data, now available to everyone

Search is now open to every user, with highlighted matches that take you straight to the right table cell or field. Admins can also top up credits from the Billing page, and the sign-in screen now offers social login alongside SSO.

**New**

* Search is now available to everyone in your organization. Find flows, runs and data from the search panel, then jump to the match: matching table cells and form fields are highlighted, and clear actions let you open details, show items in tables, or filter a table by run.
* Admins can now buy more credits directly from the Billing page: pick a preset or custom amount, review what you're getting and continue to checkout. Other members see that credit purchases are handled by organization admins. The credit volume slider is easier to use too, with clickable tick labels.
* The sign-in screen now offers Google, Microsoft, LinkedIn, GitHub and GitLab sign-in alongside SSO, right above the email form. Password fields on sign-in, password reset and account settings now have a show/hide toggle so you can check what you typed.
* Table extraction can optionally record where each row and cell sits on the page. In review, hovering a table row highlights it on the document, switches to the right page and scrolls it into view, so you can check extracted tables against the source quickly.

**Improved**

* Review screens give your documents more room. Receive File shows each document larger, Receive Email reviews scroll in one place with the PDF fitted to its column, and integration tasks such as SharePoint and Azure Blob now have a details view with their fields and documents side by side.
* Confidence scores can only be turned on for models that support them, and you get a clear message when a model doesn't. When confidence scores are off, results show N/A instead of 0%.
* Edit saved credentials without re-entering your secrets. Secret fields show a masked value with Change and Cancel actions, and any secret you leave untouched keeps its saved value.
* Flows run faster: each task now starts sooner after the previous one finishes, cutting the waiting time between steps in multi-step flows.

**Fixed**

* The CSV upload preview now keeps quoted values that contain commas in a single column, and database matching against CSV databases keeps leading zeros in codes such as 000.

**Week of June 29, 2026**

Azure Blob file handling and smoother review

## Choose what happens to Azure Blob files after pickup

Azure Blob Storage inputs can now leave, move or delete files once they've been picked up. Reviewing executions, managing saved table views and reading credit balances also got smoother.

**New**

* Azure Blob Storage inputs now let you choose what happens to each file once it's picked up: leave it where it is (the default), move it to another container or prefix, or delete it. The action runs only after the file has started a flow, so files are only moved or removed for runs that began.

**Improved**

* Step through executions directly from the review pane. If you have unsaved changes, you are asked before switching, and task details, tags and the selected execution stay in sync and load smoothly as you move from run to run.
* Saved views in the flow executions table update instantly when you create, rename, duplicate or delete them, and a newly created or duplicated view is selected for you. Showing, hiding and reordering columns also feels smoother, with your changes saved as soon as you pause.
* Credit amounts read more clearly: the sidebar credit meter and the credit limit dialog now use the same compact notation as the billing page, such as 1.5M instead of 1500k. If your organization has used more credits than it has available, the Additional credits card on the billing page now shows the amount owed.

**Fixed**

* When you re-run an execution after removing a task from the flow, the removed task no longer appears in that execution.
* Task status in the executions table now matches what task details show.

> [!WARNING]
>
> **Breaking API changes**
>
> * Credit usage is now opt-in on the flow and task execution read endpoints. total\_credits on flow executions and credits on task executions are returned only when you pass include\_credits=true, which keeps standard responses lighter; previously, getting a single flow execution always included total\_credits. **Migration:** Add include\_credits=true to requests that read total\_credits or credits.

→ [API reference](https://docs.flows.super.ai/api-reference/introduction.md)

**Week of June 22, 2026**

Growth plan and billing

## Upgrade to the Growth plan and compare plans right in the app

Admins can now upgrade to the Growth plan from inside the app, compare plans side by side and size their monthly credits before subscribing. Task references in the flow editor are also easier to read and work with.

**New**

* Organization admins can now upgrade to the Growth plan from inside the app, monthly or annually. The new pricing section compares the Free, Growth and Enterprise plans side by side, lets you size your monthly credits with a slider, and includes a contact form for billing questions. Your subscription summary shows your current plan separately from any scheduled change, and your monthly credits now renew on your billing date.
* Python tasks can now use the exiftool library to read metadata from PDFs and images, such as capture details, camera information and embedded document properties. This lets you sort, check or enrich files based on their metadata without an extra step outside your flow.

**Improved**

* Task references in the flow editor are easier to work with. Collection and Single references now have their own colours, and right-clicking a reference lets you jump to the referenced task or switch its mode. The reference mode settings now pre-select and label the default mode. In Python and Calculation tasks, references behave as single blocks that the cursor skips over and that copy and paste cleanly.
* If an uploaded document can't be read, for example because it is empty or isn't a valid PDF or image, the run now shows a clear message, so you know to replace the file rather than retry the flow.

**Fixed**

* Improved compatibility with SFTP servers that use older ssh-rsa host keys.

**Week of June 15, 2026**

Smarter flow building

## Build API calls visually, add task conditions and read from Azure Blob Storage

Building flows gets easier: you can construct API request bodies visually, run tasks only when a condition holds, and start flows from files dropped into Azure Blob Storage. Your inbox is now split into Notifications and My Work, and search results take you straight to the data you are looking for.

**New**

* Start flows automatically from Azure Blob Storage. Add the Azure Blob Storage input task, connect your storage account, choose a container and optionally a folder prefix, then point an Azure Event Grid subscription at the webhook URL shown in the task form. Only files inside the container and prefix you chose are processed.
* Build the request body of an API call task visually: switch the body to Builder and add fields and values, or stay on Raw JSON if you prefer. If a raw body, headers or query parameters aren't a valid JSON object, the task run shows a clear message explaining what to fix.
* Give a task a condition expression so it runs only when the condition is met. Use it to skip steps that don't apply to a document or to branch your flow without extra tasks.
* Find Similar Examples tasks now have their own details view in the Review tab. It shows the search parameters, such as database, table, strategy and column matches, next to the matched examples in a table. Resizing a column in review tables now keeps neighboring columns in place.

**Improved**

* Your inbox is now split into Notifications and My Work tabs, each with its own count, so updates and the review work assigned to you no longer compete for attention. My Work groups items by flow, and you can mark notifications read or unread one at a time or all at once. Work you've already completed shows as completed when you return to it, and review queues refresh right after you act.
* Search results show an icon for each task type and display full run IDs. Use Show in table to jump into the data view filtered to that run, scrolled to the matching task's column, or Go to task to open the flow with that task centered.
* Human review tasks that review items one by one now accept simple values, such as a single number or text, as the reviewed output for each item, not only structured objects. This lets you put a person in the loop on more kinds of task output.
* Python tasks can declare a typed output, and each run's result is checked against it. Mismatches are caught at the task itself instead of surfacing in later steps of your flow.
* The confidence score toggle in extraction tasks is now disabled for models that don't return confidence scores, and it switches off automatically if you change to one of them. Gemini 2.5 models now support confidence scores.
* In member and worker group management, the Teams, Flows and Workers columns expand into a searchable list. Open the full list in a side panel to see everyone or everything assigned at a glance.
* Filter the list of flow executions by date: the List flow executions endpoint accepts created\_after, created\_before, updated\_after and updated\_before (ISO 8601, inclusive), so you can fetch just the runs from a time window.

**Fixed**

* Attribute mapping in SSO settings is easier: paste your identity provider's metadata URL or XML and pick from the claims it lists, or type a custom attribute name. Long claim names scroll fully into view.

**Week of June 8, 2026**

Credentials, signup and review

## Store API credentials once and reuse them across your flows

A new Credentials tab in Resources lets you save API keys and tokens once and use them in API and webhook tasks, with values kept masked. New teams can now sign up on their own and start on the free tier, review screens are easier to edit, save and re-run, and merge tasks gain inner, left, right and outer joins.

**New**

* Manage API credentials in the new Credentials tab under Resources and reuse them across your flows. API tasks can map a stored credential into request headers, and webhook tasks can authenticate with one. Credential values stay masked in task results, logs and edit forms, so you no longer paste secrets into individual tasks.
* Sign up on your own: create an account and organization from the signup page with a supported sign-in provider or a magic link, and start on the free tier without waiting for an invitation.
* Re-run a task's run straight from the review screen in My Work, if your role allows it. Choose to re-run every task or only the tasks affected by changes, and you move on to the next item in your work queue automatically.

**Improved**

* Review screens are easier to work in: collapse or expand individual task panes, save source edits with a Save button, add or remove tags, and confirm before leaving with unsaved changes. Reviews use the exact flow version each run executed with, shown as a copyable flow definition version in task details, and merge reviews highlight invalid cells as soon as they load, with tables that fit the available width.
* Merge tasks now support inner, left, right and outer joins on a shared join key, so you can combine data sets the way you would in a spreadsheet or database. The join key appears once in the output and other columns are labeled with their collection's alias, so joined results are easy to read and use in later tasks.
* The runs table now shows a Created At column, so you can see when each run started and audit your history at a glance.
* SharePoint integrations now notice when the monitored folder has been deleted. The integration editor clears the missing folder so you can pick a new one and keep documents flowing in.

**Fixed**

* Date filters in flow details now cover the same days for everyone, whatever their time zone, including when you share a filtered link.

**Week of June 1, 2026**

Smoother review and run tables

## Resize and collapse review panels, and filter My Work by run

Review screens now adapt to your workspace with collapsible, resizable panels and resizable side-by-side task cards. Run tables are easier to navigate, with clickable inputs, a Run ID filter in My Work and clear sort indicators.

**New**

* Filter My Work by Run ID. Add the new Run ID filter, enter a run's ID, and the list narrows to the review tasks from that run, so you can focus on one execution without scrolling through everything else.

**Improved**

* Document and text review screens adapt better to your space. You can hide or resize the extractions panel, and hide the page overview, which also tucks away automatically when space is tight. Your panel choices are remembered. Each section now fits the screen and scrolls on its own. When two task cards sit side by side in a review, drag the divider between them to resize, and the widths snap to tidy positions.
* Run tables are quicker to navigate. Click a metadata badge or a file in a run's inputs to jump straight to the task details, and the table is less cluttered with the Created At column removed.
* It's now clear how a table is sorted. The Sort button shows the active sort (Created at or Updated at) with an arrow for its direction, and the Created At column header shows a matching arrow, including for the default newest-first order. The sort button also has a descriptive label for screen readers.
* Edited tasks are easier to spot. An Edited badge marks tasks whose output has been edited, and the re-run notice now appears only while downstream tasks still need a re-run, with a See details link to open the task and a Re-run downstream button.

**Week of May 25, 2026**

Organization security and review

## Require SSO and MFA for your organization, and review tasks side by side

Organization admins can now require single sign-on and multi-factor authentication for everyone, and restore deleted items from the new Trash bin. Multi-input human review gets a card-based, side-by-side layout, and a new Review Flow page lets you compare several tasks of a run at once.

**New**

* Admins can now require SSO sign-in, multi-factor authentication, or both for their whole organization from the Security section of the organization settings (SSO requires a configured SSO provider). Members who haven't set up MFA are guided through enrollment with a QR code or secret key, and members of an SSO-only organization are directed to Continue with SSO when they sign in.
* A new Trash bin, available to admins from the profile menu, lists deleted flows, members, resources and databases, sorted newest or oldest first, and lets you restore any of them with one click. Restored items reappear in their lists right away, so an accidental delete is easy to undo.
* The new Review Flow page lets you compare the results of several tasks from the same run side by side. Hover over a row in the data table and click Review, then use Add Task to Review to add, remove or reset the tasks you want to inspect, and your layout is remembered.

**Improved**

* The multi-input Human In The Loop task now uses a card-based review layout that can show several task results side by side. A Layout menu lets you reorder, hide, or switch each source between full and half width and toggle its inputs, and your layout is remembered in your browser. Reviews can also target individual items inside an output collection, and if someone else is already reviewing a task, an inline banner tells you so and the task opens read-only.
* Merge review keeps columns in a predictable order that follows your field mapping, shows the expected type in a tooltip when a cell value is invalid, and keeps your cursor in place while you edit. Saving is only enabled once every cell is valid, so you catch mistakes before submitting.
* The flow data view has a Created At column and a From/To date filter, so you can find items created in a specific period. Open-ended ranges work too, for example everything created since a given day.
* The real-time data table stays responsive on larger flows. Task results show their final status right away, columns keep their order when you show or hide them, and the filter menu always reflects the view you switch to.
* Browser tabs now show the name of the flow or work item you have open, so you can tell several open tabs apart at a glance.
* When a task's output configuration needs attention, the flow's validation now shows a clear message explaining what to fix.

**Fixed**

* The agent prompt editor now keeps your cursor position while you type.
* When you cancel a flow run, any tasks still running or waiting are now marked as cancelled too.

**Week of May 18, 2026**

Smoother human review

## Review tasks that lock to one reviewer and track what changed

Human review gets a clearer screen, a rebuilt merge table, one-reviewer-at-a-time task claims and a new output showing what reviewers changed. Table extraction is more reliable, review screens handle image documents smoothly, and flows with large outputs open faster.

**New**

* When you open a review task that someone else is already working on, you now see who has it and can move straight on with Go to next. Tasks another reviewer has started no longer appear in your queue, so two people don't end up reviewing the same item at once.
* Human review tasks now have a was\_changed output that shows, for each editable input, whether the reviewer changed it. Use it in later tasks to route corrected items differently or to see how often reviewers adjust results.
* Text and number extraction fields can now be set to return a single value instead of a list. The output is one value or empty, which gives later tasks a simpler shape to work with, and the review screen marks these fields as single value.
* Choose your own default view for each flow. The view you mark as Default opens first for you, without changing what your teammates see.

**Improved**

* The review screen in My Work and Task Details is easier to work through. Each source has a header with its task icon, and when a review has several sources you get a counter and expand or collapse per source; every item shows whether it is editable or read-only. You can edit and save individual items in a completed review, and re-running a flow refreshes its open reviews so reviewers always work on the latest results.
* Human review tasks now ask for at least one assigned user or worker group when you configure them, so every review reaches someone.
* Merge reviews now use the same table as text-to-structured reviews. Edit cells in place, resize columns, add or delete rows from the right-click menu, and scroll sideways through wide tables.
* Review screens for document extraction and classification now display image files such as JPGs as smoothly as PDFs, choosing the right viewer from the file type. In the document-to-text view, large images keep an even split with the output panel.
* Table extraction is more reliable. It returns clean rows keyed by your configured columns, captures every row in the document, and handles values with currency symbols, thousand separators or units. If a first pass comes back empty, it automatically tries again.
* The Find Similar Examples task now returns matching rows keyed by the columns of the table you select, so you can reference individual columns in later tasks.
* Flows with large task outputs open faster. Heavy outputs load on demand, their cells show Loading until they arrive, and row details show the full output. The Review tab loads only the task you pick, which keeps live updates light.
* Your profile page now has separate sections for account details and for changing your password, with checks that a new password has at least 8 characters and matches its confirmation. The Credits link in the sidebar now takes you to your organization's billing page.

**Fixed**

* Email attachments with encoded filenames now keep their full original name.

**Week of May 4, 2026**

Organization and review upgrades

## Manage your whole organization from one place, and review PDFs page by page

A new Organization area brings members, worker groups, service accounts and SSO together in one place. Review pages get a paged PDF viewer, and your inbox now tells you when you're added to a team or invited to a flow.

**New**

* A new Organization area brings your organization's settings together in one place: organization profile, members and their roles, worker groups, service accounts with their API keys, and SSO settings. Admins can add, edit and remove members and change roles without going through their personal profile.
* Your inbox now tells you when you're added to a team or invited to work on a flow, with the flow's name and a direct link to open it. When you join a worker group that is already assigned to active flows, you'll get an inbox notification for each of those flows so you can start working right away.

**Improved**

* Review pages now show PDFs in a paged viewer with page-by-page navigation and a page indicator, so long documents load quickly and stay easy to move through. The document viewer has a download action that saves the file directly, and review screens scroll as a single card with sticky section headers.
* The runs table on a flow loads faster and remembers your view, filters, sort order and page in the address bar, so you can bookmark or share exactly what you're looking at. You can also search for columns when customizing the table.
* The Write to Database and Find Similar Examples tasks now work with your table's real columns. Pick columns from a selector, map values to them directly, and choose whether all clauses or any clause must match when searching for similar examples.
* The Multi-input task's output fields can now be of type Object, with their own nested sub-fields, alongside Collections, so you can describe structured results such as an address more precisely. Metadata fields you configure on a file input can now be mapped one by one, and Excel attachments on incoming emails are recognized as spreadsheets.

**Fixed**

* When a review is submitted while your organization is at its concurrent-run limit, the run now waits for a free slot and then continues automatically.
* Notify tasks now add or remove their tag on the selected task, including when that task's output is referenced as a whole.

**Week of April 27, 2026**

Smarter review and learning

## Review several tasks at once and teach flows from past examples

A new multi-input Human In The Loop task lets one reviewer check several earlier tasks in a single review, and your organization now has its own database: create tables in Resources, then save results and reuse past examples from your flows. Webhook tasks gain configurable retries with a clear view of every delivery attempt.

**New**

* A new Human In The Loop (multi-input) task lets one reviewer check the results of several earlier tasks in a single review. Builders choose which items reviewers can edit and which they only see for context, and reviewers can edit merged collections directly in the table. Every reviewed item becomes its own output, so later steps can use the corrected data right away.
* Two new tasks help your flows learn from earlier work: Write to Database saves results to a table you pick, and Find Similar Examples looks up comparable past records to give the next step useful context. The table picker warns you about missing columns before you run, so setup mistakes show up early.
* Webhook tasks now let you decide how deliveries are retried. Open Retry Settings on the task to set the number of retries, the wait between them and which status codes should be retried. In the run details, pick any attempt from a dropdown to see when it was sent and the response it got, with the latest attempt selected by default.
* Your organization now has its own database. In Resources, create tables and columns, choose column types, set primary keys and relationships between tables, and browse the data in each table, ready for your flows to read and write.

**Improved**

* Flows that accept several kinds of input now get a tailored review view that shows which source started the run and the exact data it received. Multi-input flows that accept files can now be started with the Run action and file upload, and their files and details appear in the Runs and My Work tables.
* The flow list in your hub is now a full table with resizable columns that remember their widths, and a header that stays in view as you scroll.
* Long column headers and task titles are easier to read: hover over them to see the full text, and headers show their icon where one is available. Selected rows are now highlighted across the whole row, including pinned columns.
* Python tasks can now use the jsondiff library, so you can compare two JSON documents and work with the differences directly in your code.

**Fixed**

* Run tables and the review tab now stay up to date as runs progress and when you re-run tasks.
* Flows triggered by email now read the sender and date correctly across more email formats.

**Week of April 20, 2026**

Teams and faster review

## Invite whole teams to flows and keep track of what's unread

You can now invite teams to a flow, mark inbox notifications read or unread, and preview files right from your runs. My Work, work mode and the runs table are also easier to navigate.

**New**

* Invite whole teams to a flow, not just individual people. The flow's sharing panel now lists invited members and invited teams separately, and team members get access based on their role, along with an email invitation. When you assign people or teams to a human review task, anyone not yet invited to the flow is flagged with a quick way to invite them, and the search box clears after each pick so you can add several quickly.
* Mark inbox notifications as read or unread, one at a time or all at once from the inbox header menu. Your inbox now shows both read and unread items, and the badge counts only what you haven't read yet, so it's easier to see what still needs your attention.
* Preview files and attachments from a run in a full-height side panel, step through multiple attachments, and jump straight to the task that produced them. The Files and Metadata columns in My Work and Runs now show files from every input type, including email, Google Drive, SharePoint and SFTP.
* Download the source document with one click while reviewing a document or text extraction task. Downloaded files keep a sensible name and extension, so they're easy to file away.
* Turn confidence scores and bounding boxes on or off per task when you configure document extraction. The settings detect what the selected model supports, enable only the options it can use, and pick a suitable default reasoning effort when you switch models.

**Improved**

* My Work now loads one page of tasks at a time, with a page-size selector (10, 25 or 50), first and last page controls, and the total number of tasks shown. Large queues open quickly and stay easy to move through.
* In work mode you can switch to another run from the sidebar to look at it. Runs outside your work queue are clearly marked and read-only, and if you have unsaved changes you're asked whether to discard them before switching.
* The runs table keeps queued runs out of the list and shows separate queued and running counts in the footer instead. Task start and completion times now show to the second, time to complete appears for finished tasks, and cancelled tasks are marked individually in flow review.
* Setting up a Multi-input task is clearer. Edit its output schema in a dedicated dialog where you can add and manage fields and sub-fields, then see a compact summary on the task. Output mappings have separate edit and read-only modes, and each trigger highlights any fields that are not mapped yet.
* Team pages are easier to manage: edit a team's name and description in place, search its members in a table showing role, email and date added, and remove members or delete the team from the same page.
* The flow editor handles groups and sticky notes more smoothly. Importing a flow brings its groups and notes along with its tasks, auto-layout leaves them where you put them, and dragging tasks into and out of groups is smoother, even in large flows. Re-running tasks or runs now saves your unsaved changes first.

**Week of April 13, 2026**

Multi-input flows and review tools

## Start one flow from several inputs and rotate pages while you review

Flows can now accept more than one input source and map them into a single set of fields, and review screens let you rotate document pages. Excel files preview inline, tags can be added at upload, and notification settings have a cleaner layout.

**New**

* Run a flow from more than one input source, such as an email, a file upload or a form. The new Multi-input task lets you add the input triggers you want, define one output schema and map each trigger's fields into it, so the rest of your flow works the same no matter how the work arrived. Files you upload from the Run button go straight to the right input automatically.
* Rotate document pages in 90-degree steps while reviewing Classify, Doc to Structured and Text to Structured tasks. Rotate a single page from its context menu, a whole document in one click, or use the rotate button in the document toolbar. Bounding boxes step aside while a page is rotated and can be switched back on. Rotation only changes your view, so sideways scans are quick to read and the original file stays untouched.
* Excel files now preview inline in task details and hover cards. You see the first sheet as a formatted table with its total row count and a download button; files larger than 5 MB offer a download instead.

**Improved**

* Add tags while you upload files or URLs to run a flow, so runs are organized from the start. In task details and My Work, the tag bar now shows the tags for the item you're viewing and follows your selection as you move between items.
* The Notifications tab in flow settings has a cleaner layout with separate Email and Inbox cards. Set an hourly or daily email digest in a single line, add recipients with a searchable picker, and preview the email subject and body with your flow's actual name.
* Paging through runs in the Table view is easier: jump to the first or last page and choose 10, 25 or 50 runs per page, and your choice is remembered. A single icon on the run number highlights runs that need attention in the columns you're viewing, and columns you reveal return to their natural position.
* The From JSON task can now extract specific fields from items in an array, including arrays nested inside objects, and can target keys that contain dots. This makes it easier to pull exactly the values you need from complex JSON, and the builder stays in sync as you edit or remove fields.
* When a human review task is assigned to a worker group, everyone in that group automatically gets access to the flow. People who join the group later are added to flows where the group still has open review tasks, so reviewers can start right away without waiting for an invitation.
* When your organization has many runs in progress, new runs wait in a queue and start automatically as capacity frees up. Queued work is balanced fairly across your flows, so one busy flow doesn't hold up the others.
* Organization administrators can now delete any database in their organization, while other users can delete the databases they own.
* Long flow names show in full when you hover over them in card and list views, and card titles have more room. Worker tasks are now listed as Processing tasks, model settings are labeled AI model, and the My Work badge turns grey when you have nothing waiting.

**Week of April 6, 2026**

Run alerts and smarter documents

## Get run alerts by email and in your Inbox, and process Excel files

You can now set per-flow alerts that notify you by email digest or in your in-app Inbox when runs need attention. Document processing also gains Excel support, image attachments from email, auto-rotation and whole-document classification.

**New**

* Set up run alerts for each flow from the new Notifications tab in flow settings. Get alerted when a run errors or keeps running longer than a time you choose, pick the people or teams to notify, and receive hourly or daily email digests, so your team hears about runs that need attention without watching the dashboard.
* Flow notifications now also arrive in your in-app Inbox, grouped by flow alongside your review tasks, with an unread count that updates live. Jump straight to the flow from a notification, or copy its run ID to share it.
* Receive File, Receive Email and SharePoint tasks now accept Excel spreadsheets (.xlsx and .xls), and Receive Email also picks up JPEG and PNG attachments alongside PDFs. Use Doc to Text to turn the first sheet of a spreadsheet into text, and its data then shows as a table when you review Text to Structured results. Tasks that don't support a file type tell you so clearly as soon as they start.
* Document Classifier, Doc to Structured and Doc to Text tasks have a new auto-rotate setting that straightens scanned or sideways pages before they are read. It is off by default. With Doc to Structured, the rotated document is shown with the extraction results.
* Classification tasks can now classify a whole document at once instead of page by page, with an optional page range to focus on the pages that matter. The task output also lists which pages were used for classification.
* Edit your templates in a full-screen editor with live validation, update their name, description, tags and visibility, or delete templates you no longer need. If you try to leave with unsaved changes, you're asked first, so you won't lose work by accident.
* GPT-5.4, GPT-5.4 mini and GPT-5.4 nano are now available in the model selector for AI tasks that use the Global processing region, giving you more choice between quality, speed and cost.

**Improved**

* Drawing a box on a document during review now fills in the text right away, reading it from the PDF directly or recognizing text in scanned pages when needed. Overlapping highlights stay easy to tell apart, and when extracting with Gemini models, each value of a multi-value field gets its own highlight.
* Task details now show which flow you're in, plus a Flow Execution section with the flow and run IDs, when the run was created and last updated, and who last modified it. Human review tasks placed after a Branch or Validation task now show the original document and results from the task that produced them.
* In the Data view, filter run status, task status and tags by "contains one of" or "does not contain", and hover truncated filenames to read them in full. In My Work, click anywhere on a row to open it and see the new Run ID column. You can also delete a flow straight from its header, and documents load more smoothly in the viewer.
* Creating a flow in the Hub now starts with naming it, including when you start from a template, and a new Created By filter helps you find your own flows or a colleague's. The Hub also points out empty flows older than 30 days so you can clean them up in one step.
* Configuring tasks is faster: dropdowns throughout the app are now searchable, the Notify task's tag field lets you pick an existing tag or create a new one, and task settings have clearer tooltips explaining what they do. Selected tasks and their connections stand out more clearly in the flow diagram.
* The review screen now shows a clear explanation, with options to re-run or open the flow setup, whenever a task needs attention, and tasks automatically retry on temporary issues with connected services.
* Google Drive and SharePoint triggers now work through large batches of files page by page, so even very large folders are handled smoothly.
* The Write to SFTP task can now rename files by adding a prefix or suffix to the original filename, including values referenced from earlier tasks, so output files follow your naming conventions.
* Admins can search, sort and page through teams, create and edit them in dialogs, manage members, and delete several teams at once. The Users tables now show each user's teams.
* Your profile page now shows your organization's details, with one-click copy buttons for IDs and other read-only fields.

**Fixed**

* Task statuses now always reflect the final outcome of a run, including runs that end early or are cancelled.

**Week of March 30, 2026**

More Gemini models

## Choose from the latest Gemini models, with European and global options

You can now pick Gemini 2.5 Flash Lite and the newest Gemini 3 and 3.1 preview models for your tasks, and bounding-box extraction works with more Gemini 2.5 models. Reviewing work is smoother too, with clear unsaved-changes handling and edits that show up right away.

**New**

* Gemini 2.5 Flash Lite and the Gemini 3 Flash, Gemini 3.1 Pro and Gemini 3.1 Flash Lite preview models are now available when you choose a model for a task. Each Gemini model offers both a European and a global region option. Model names in the selector are easier to read, and flows that use an older model name keep working with no changes needed.

**Improved**

* Bounding-box extraction now also works with Gemini 2.5 Flash and Gemini 2.5 Flash Lite, so you have more model choices for locating extracted values on the page. Extraction also handles more model response formats, including a separate location for each value in multi-value fields, so results are more consistent.
* Reviewing work is smoother. In the work queue, an Unsaved changes indicator appears next to Complete work, you're asked to confirm before leaving an item with unsaved edits, and Complete work takes you straight to the next item. Key-value pairs you add show up in the task output right away, and the Review tab's task selector now shows full task names.

**Week of March 23, 2026**

Smarter review and navigation

## Save edits, auto-rerun stale tasks, and find any task fast in Flow View

Review now shows clearly which results were edited and which are out of date, and can re-run affected tasks for you automatically. Flow View, templates, flow versions and run filters also get easier to work with.

**New**

* A task result can now be marked as both edited and stale, so you can see at a glance which outputs were changed by hand and which are out of date because something upstream changed. You can also filter the data table by Edited. On the review page you save corrections with a Save button, you're asked before leaving with unsaved edits, and hovering the Edited banner shows when the task was edited.
* Turn on the new Auto-rerun downstream stale tasks setting for a flow and, when you update a task's output, the tasks that depend on it re-run on their own. You no longer need to track down and restart each affected task after a correction.
* In Flow View, select any task to highlight its connections while the rest of the flow fades back, and use task search to jump straight to the task you need. Results are ordered by where tasks sit in the flow, and large flows with many tasks stay smooth and responsive as you work.
* Give flow versions your own names, star the ones you use most and switch on the Favourites filter to see only those, and search versions by name to find the right one quickly. The version list now shows the email of the person who last modified each version.
* Add your own tags to flows and use them to search and filter your flow list, so you can organise flows by team, customer process or project.
* The template library has a My Templates tab, tag filters and tags on every template card. You can edit the name, description, tags and visibility of templates you created, and delete your own templates when you no longer need them.

**Improved**

* Filter runs by the status of individual tasks and exclude runs with specific tags, making it easier to find exactly the runs you need to look at. Saved views keep their filters, the status column in the runs table is wider and resizable, and the app now consistently uses the word Run.
* Data table cells are cleaner and easier to read: hovering a cell shows a redesigned card with type icons and a table view for lists of values, and tags, file attachments and status badges show their details in tooltips. Run tags appear on hover in review, all tags share one consistent look, and every task status cell opens its details with a click.
* My Work now has its own dedicated pages: a work queue for each flow and a focused page for each work item, and inbox links take you straight there. When you open a task that is still waiting for a worker, you see a clear Waiting for Worker status. Workers and reviewers with no flows assigned yet see a dedicated empty state with a quick refresh link.
* Classifying and splitting multi-page documents is faster because pages are now classified in parallel. Page rotation is more reliable at turning sideways or upside-down pages upright, and it can reuse the page angles detected during text extraction to skip extra processing.
* Your layout choices, such as cards or list view for your flows, the sidebar state and collapsed templates, are now remembered across sessions and devices.
* Every task type now has its own distinct icon, and integration cards show a matching icon when no logo is available. You can also ask for a missing integration directly from the integration catalog with the Request Integration button.
* Large flows with many tasks now run noticeably faster, thanks to smarter reuse of intermediate results while a run is in progress.

**Fixed**

* File URLs for split documents and split text in flow outputs can now be downloaded through the file download endpoints. → [Downloading Files](https://docs.flows.super.ai/guides/downloading-files.md)

**Week of March 16, 2026**

Version history and team review

## Save flows explicitly, restore past versions, and route review to teams

Flows now have explicit saving with a full version history you can restore from, and Human In The Loop tasks can be assigned to whole teams. Run tables, filters and review pages also got a round of upgrades.

**New**

* Flows now have version history. Open it from the flow menu to view past versions and restore any of them. Changes are saved when you choose to save, with live validation as you edit, and if you leave with unsaved edits you can recover them when you come back. Moving tasks around on the canvas doesn't create a new version.
* Admins can now create teams and assign Human In The Loop tasks to a whole team instead of a single person. When a team member starts a task it becomes theirs, and if it is left unfinished it goes back to the team for someone else to pick up. Teams and their members are managed from the new Teams tab in Admin settings.
* When you duplicate a flow that receives input from email, cloud storage, SFTP or an integration, you now see which tasks are affected and choose whether to clone the flow as-is or reset their input settings. This keeps a copy from accidentally picking up documents meant for the original flow.
* Image rotation tasks have their own review page showing the source and output documents side by side. You can rotate PDFs and images by 90 degrees right from the review screen and the corrected document is saved as the task output.
* Tasks and their parameters now show built-in help tooltips in the flow builder, so you can see what each setting does without leaving the page.

**Improved**

* Filter flow runs by task name or look one up by its exact execution ID. When a task is selected, the tag list narrows to the tags used on that task, which makes it quicker to find the runs you need.
* The table view of flow runs shows more at a glance. Each run number carries a count of failed tasks and a marker for cancelled runs, tags appear on each task execution with a count badge, and edited task outputs are marked as edited. From the hover card you can add or remove tags, confirm or unconfirm a task output, and see the details of whichever execution you are viewing.
* Human Review tasks are now called Human In The Loop tasks, and task search matches display names as well as task types. Your work list no longer shows tasks from cancelled runs, and if a task becomes unavailable while you're on it, you're taken straight to the next item.
* Flow views can now be removed as the default, and default views are listed first so your main view is always easy to find.
* The integrations page has a new sidebar with categories and per-category counts for easier browsing. You can also open the integrations catalog straight from the Add Task panel to explore what's available.

**Fixed**

* If an invitation link has expired, requesting a new one from the sign-in screen now sends a fresh invitation that works right away.

**Week of March 9, 2026**

Smarter review and runs

## Give reviewers context, build custom Runs views and draw to fill table cells

Review tasks now carry messages that point reviewers to what matters, and a Work completed badge shows when work is done. You can also draw directly on the document to fill table cells, build your own Runs views, see at a glance which runs are awaiting a reviewer, and save any flow as a template.

**New**

* Give every task in your flow a description: choose Edit description from the task's menu and type it inline. In a Human Review task, you can pick the validations that matter most, and if one of them doesn't pass, its description appears as a message on the reviewer's screen, so reviewers know exactly what to look at.
* Create custom Runs views alongside your table views, with a column set built for runs and one-click copying of run IDs. The Files column in the runs table is now called Input, and setting a view as the default is labelled Set as Default for Everybody so it is clear who it affects.
* When reviewing extracted tables, click into a cell and draw a box around text in the document to fill that cell. Click a table's name to open it side by side with the document; your selected table and column widths are remembered as you move between tasks and runs, and tables with only a few columns now fill the panel.
* Set vectorization parameters on a task: choose a source task and decide for each output whether it is used as a single value or as a collection. This gives you control over how a task fans out across items and how results are gathered for the tasks that follow.
* Save any flow as a template from the new Template tab in the Share dialog, with a name, description, tags and who can see it. Start new flows from your saved templates in the Flow hub, where you can browse and search them.

**Improved**

* Code tasks use a new code editor with Python syntax highlighting, and typing # suggests references to other tasks in your flow. Your code keeps its indentation and formatting exactly as you write it, and Tab and Shift+Tab indent and outdent lines.
* Completed review tasks show a Work completed badge in the task header. Click Info on the badge to open the task's details, where you can see who completed the work.
* My Work is always available to reviewers and shows a clear message when nothing is assigned yet. Work mode makes it obvious when you're viewing a task outside your queue, takes you back to My Work after you finish a single item, and keeps your edits and cursor in place while you review.
* Runs that are waiting on a person now show an amber Awaiting Worker status and return to Processing as soon as the review is done, so you can see at a glance which runs need someone's attention.
* SFTP tasks let you give folder mappings a display name, fill in the name automatically when you pick a folder, and rename items in place by clicking their title. Folder browsing shows each folder once as you navigate.
* When building JSON in a task, you can choose a Reference value type that pulls in the referenced value from earlier tasks, and fields save smoothly as you type.
* The MCP server now exposes the full public API as tools, so AI agents connected over MCP can do everything the public API supports, authenticated with your API key. → [MCP Server](https://docs.flows.super.ai/api-reference/mcp-server.md)

**Fixed**

* Flow invitation emails now link each invitee to the right place for their role.

**Week of March 2, 2026**

Smoother work for workers

## Filter My Work, pick worker landing pages, and connect agents over MCP

Workers get a clearer daily flow, with filtering and sorting in My Work, a landing page chosen by their admin, and guidance as they work through their queue. Runs show their status at a glance, and AI agents can now build and edit flows in SuperAI Flows through the new MCP server.

**New**

* Connect Claude Code, Claude Desktop or any other MCP-compatible client to SuperAI Flows through the new MCP server. Using a service account API key, your agent can browse the available task types, create flows, read and update flow definitions, and generate file download links, so you can build workflows from the assistant you already use. → [MCP Server](https://docs.flows.super.ai/api-reference/mcp-server.md)
* Admins can choose where workers land after signing in: the Hub, or the Data or Work tab of their most recently active assigned flow. Set it once under Admin Tools > Worker Entry Customization, and workers go straight to their work (they land on the Hub when nothing is waiting).
* My Work now has filters and sorting. Narrow your tasks by tags or metadata, sort by created or updated time, and by default the longest-waiting tasks come up first.
* Task Details now shows a clear explanation when a task needs attention: the task name and the exact message, with one-click options to re-run it, open the flow setup, or, in Work mode, skip to the next task.

**Improved**

* Work mode guides workers through their queue. A banner tells you when a task isn't part of your work queue or who already completed it, Confirm can now be undone with Unconfirm, and an All done! dialog takes you back to My Work when you finish. Results submitted for a task are checked against the output fields your flow defines, so mismatched values are flagged right away.
* The Reviewer role is now called Worker across the app, invitations and emails, and what each role can do is clearer: creating and deleting tags follows the user's role permissions.
* Runs now show their status in a dedicated column, in both the data view and the runs view. In the data view, clicking any task cell, even one still running or skipped, opens its details. In the runs view, tags take you to the related task, with a short list to pick from when a tag appears on several tasks. The selection bar now offers View Task Details and shows Cancel only for runs that can still be cancelled.
* Tables are easier to work with. Run number and status stay pinned on the left as you scroll, an optional Execution ID column in the data view lets you copy IDs with one click, and Flow ID, Execution ID and other details in Task Details can be copied from the side panel.
* Re-running flows with human review steps is smoother. Each re-run starts with fresh review tasks, you'll see a notice if a review task you had open was replaced by a re-run, and tasks waiting on people now read Waiting for worker.
* Python code tasks can do more. You can now import requests to call external APIs, magic to detect file types, and io to work with in-memory files, so a single code step can fetch, inspect and send data.
* The JSON builder handles arrays of structured items. Define an item schema, nest arrays and objects, choose a type per item, and expand or collapse nested fields while you build.

**Week of February 23, 2026**

Work mode and run control

## Work through review items in My Work and cancel runs you no longer need

A new My Work tab and work mode let reviewers move through their assigned items one after another, and you can now cancel runs that are in progress. Data views get pinned and resizable columns, runs can be filtered by metadata, and your credit balance now shows your monthly allowance and additional credits separately.

**New**

* Cancel runs that are still in progress with the new Cancel button, available to users with the right permission. Cancelled runs are clearly marked with their own status and a neutral row style in your tables, so you can tell them apart from completed or failed runs at a glance.
* The new My Work tab lists the review items assigned to you. Select items and enter work mode to step through them one by one: when you finish an item you move straight to the next open one, completed items stay done, and a banner shows the email of any other reviewer working on the same item so you don't duplicate effort.
* Filter runs by the metadata collected with your uploaded files. Pick from the metadata keys and values already used in the flow and select several values per filter to narrow the runs view to exactly the runs you need.
* Usage is now measured in credits. Your balance shows your monthly plan allowance, with its reset date, separately from any additional credits, so it's easy to track what you have left. Run and task details returned by the API now include the credits each one consumed.
* You can now reference the entire output of a previous task in a task's configuration, not just individual fields, right from the reference picker. A new Transform task passes referenced data on to the next steps of your flow.

**Improved**

* Data views are easier to shape around your work. Pin task columns to the left or right, drag to resize columns, and have both saved with the view. The Runs view can now be chosen as your default, and switching between views keeps filters, counts and layout in step with the view you picked.
* Bulk actions in the runs table now show how many rows you have selected, for example Re-run (4) or Cancel (4), so you always know how many runs an action will affect.
* Reviewing is smoother. A flow can open its review in a custom view you choose, review screens share a consistent set of actions such as Re-run and Go to flow editor, and each state has its own clear empty screen. The inbox now says New work item and links directly to the item or your work queue.
* Tags in the review view have a new design: search the tags available in your flow, create a new one inline, and get a confirmation each time a tag is added or removed.
* Setting up SFTP tasks is easier. Browse and pick folders from a folder tree that loads as you expand it, see how many files a folder contains before you process it, and choose from dropdowns with clear labels.
* SFTP, Google Drive and SharePoint tasks now show their recognizable logos on the canvas, and SharePoint receive results can be opened in the review screen.
* When the AI builds a flow for you, it now gives each task a short title that describes its role, so your flow is easier to read in the editor and in your results.
* Starting a run with input that doesn't match the flow now returns a clear validation error right away, listing what to fix, and no run is created. → [Error Codes](https://docs.flows.super.ai/api-reference/errors.md)

**Week of February 9, 2026**

JSON tasks and SFTP

## Build and parse JSON in flows, and pull files from SFTP folders

Two new task types let you extract fields from JSON and build structured JSON inside a flow, and a new SFTP integration watches your folders and starts a flow for each file. Flow tables also get saved filters and sorting, tag filters, and faster browsing of long flow lists.

**New**

* Two new JSON tasks are available in the flow builder. The extraction task reads JSON input and pulls out the fields you choose, including nested paths, with type conversion and a strict or lenient mode. The construction task builds structured JSON from a schema you define, with nested objects and arrays and a live preview, so you can shape output for downstream systems without custom code.
* A new SFTP integration watches one or more folders on your SFTP server and starts a flow for each file that arrives. Once a file is processed, it's moved to a folder you choose or deleted. When you add SFTP credentials, pick password or SSH key under Authentication Method.
* You can now re-run a run's tasks without leaving the Review tab. The task selector shows which tasks need attention and updates as the run progresses, and a Re-run all tasks option sits in the same menu. When the task you're reviewing is out of date, a Re-run link refreshes it in place.
* Custom views on a flow now save their filters and sort order along with the columns. Use the new Sort and Filter controls on the flow table, including filtering runs by any task tag used in the flow, and the view opens exactly as you left it next time.
* Manage service accounts and their API keys from the web app. You can rename a service account, edit an API key's name and description, and delete individual keys or the whole account from per-row Edit and Delete actions. Deleting a service account also revokes its API keys. → [Authentication](https://docs.flows.super.ai/api-reference/authentication.md)

**Improved**

* The flows list now loads 20 flows at a time with a See More button, and search and sorting by created or updated date now work across all of your flows. Workspaces with many flows open noticeably faster.
* You can now edit saved integration credentials. The edit form opens with the credential's name and authentication type already selected, so you can update a password, key or secret without recreating the credential.
* The runs table has a new Metadata column that shows the metadata attached to each run's incoming file. Hover over it to see every value and who last changed it, and tags now appear as consistent badges.

**Fixed**

* Renaming and reordering tasks in a flow now saves as expected.

**Week of February 2, 2026**

Faster human review

## Review runs in batches and connect SharePoint to your flows

Human review gets a dedicated flow: select runs and review them in a batch, with task details and instructions alongside. SharePoint is now available as an integration, and the inbox groups your review tasks by flow.

**New**

* Review runs in batches. When a flow includes a human review step, the runs list shows each run's review status, and you can select runs and choose Review to work through the ones waiting for you. A bar at the top shows that a review is in progress and lets you exit at any time, and Finish Review submits each review and takes you to the next run in your selection.
* Connect SharePoint to your flows. Add a SharePoint integration from the integration catalog, either by signing in with Microsoft or with your app's tenant ID, client ID and client secret, and give it a name of your choice. Flows can then receive files from the SharePoint site, library and folder you pick.

**Improved**

* Select See task details during a review to open a side panel with two tabs, Notes and Instructions and Metadata. Metadata shows when the task ran and was reviewed and how long it took, so you have the context for a decision without leaving the review.
* Your inbox now groups tasks by flow in collapsible sections with a count for each flow. Each task shows which step needs review, a status indicator and when it was created, and you can jump straight to it with Review item or open the flow's runs with Review overview.
* Copy values with one click. Webhook payloads and text form fields now have copy buttons, joining API keys and task outputs, and every copy button confirms when the value is on your clipboard.
* Zoom documents up to 500 percent while you review them, and remove tags you no longer need. Viewer controls now show clearer icons and labels that adapt to your screen size.
* Reviews now keep themselves tidy when you step away. After 15 minutes in a review you're asked whether to continue or exit, and after 30 minutes the review closes and returns you to the flow's runs so the item is free for someone else.
* Integrations show clear, readable names throughout the catalog, forms and dialogs, and Google Drive now appears with its own logo.
* Text fields marked read-only in a task's schema, such as the auto-generated address of a Receive Email task, now appear as read-only in forms. The flow menu now reads Copy flow definition and Import flow definition, and each page of the web app follows the permissions of your role.
* Flow execution lists now sort the same way everywhere, with the most recently started runs first by default, and live updates keep your chosen sort order, filters and page size as new runs arrive. The List flow executions API now also sorts by creation time by default.
* The interactive API reference is now at /api/docs and /api/redoc, and the OpenAPI specification at /api/openapi.json, covering the full public API. If you load the specification from /openapi.json, switch to the new path.

**Fixed**

* Text files saved with a UTF-8 byte order mark (BOM) are now fully supported.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.