Every change that reached customers from May 2025 to October 2026, compiled from the code that went to production. Pick a month, browse what changed, open any item for the details.
147 new · 204 improved · 362 fixes · across 18 months · updated Oct 26, 2026
The CLI keeps a record on your machine of every change it makes and how to reverse it. mammoth history list and show say what was done, history undo takes a change back, from restoring what went to the trash to putting a view's pipeline back as it was, and history resume picks up a multi-step run where a failure or an interruption stopped it.
mammoth recipe run does every step of a common task in one command: summarize or pivot a view, keep its top rows, find and replace, split or rename columns, fill blanks, remove duplicates, add a calculated column, look up a column from another view, export to CSV, Excel or S3, email a view, schedule a refresh, or build and share a dashboard. It asks once before it changes anything, --dry-run previews every step, mammoth recipe list shows them all, and mammoth how answers a goal with the recipe to run.
Repeated reads download only what changed since the last time, a write can be retried after a dropped connection without running twice, and a multi-step run can pass one step's result to the next. Every command describes the shape of its result, so scripts and assistants know what comes back before they run it.
From URL is back as its own item in the New dataset window: it reads the file first, shows its first rows, then asks whether the dataset should follow the link. A linked dataset gets Get the latest and can refresh on a schedule like a connector's, a copy behaves like an uploaded file, and either can switch later. The Datasets page marks linked datasets with a link badge, and Connections lists a project's linked datasets with when each was last fetched and any failure.
A PDF with several tables can now be automated: one rule sends each table to the dataset it fits, and a rule can watch files added at the project's top level, outside any folder.
A PDF that looks like one already read, by its name or by its tables' columns, offers Read it the same way, which adds its tables to the same datasets with the same instruction instead of making new ones, and the save step offers to add tables to the datasets they fit.
Send into a dataset checks columns before sending, lets you merge look-alike columns or choose where one goes, and skips sources that already send there; the result is pinned and opened beside the list, and rows say what they send to or are fed by. A dataset's summary card draws its workflow, joins included, and the panel's actions sit on a strip beside its main button.
While AI builds a connector, its question cards ask only what applies, mark optional answers and suggest a value, and an id a table's address needs is filled from your answers. A table listed under each row of another, such as invoices under each invoice run, is read row by row up to two levels deep, an empty table shows as empty instead of an error, and a scheduled pull refuses to save when part of the parent list failed.
The Source Interpreter preview shows up to 200 rows instead of 50 and says when you are seeing only the first rows; every row is still imported.
The Math step's Insert Parameter adds a number parameter to a formula or to SQL mode, and the step uses the parameter's current value each time it runs.
A chart that would have nothing to show is left off when a board is made from a template or a prompt, or a page is added from another table, and the reply names what was left off; the editor chat doesn't add one either. Templates whose charts would come out empty on your data are not suggested as a match, a table too small to chart starts its new page blank, and page tabs name only the page.
An automation's schedule sets a first run and an optional end date and lists the next runs, a monthly schedule can repeat every few months, and editing an automation keeps its start. A live link export can be edited.
The CLI and SDK can list every connection in a project and test one, cancel a step preview, see or cancel a pending plan change, and attach a file to a dashboard's chat.
The CLI and SDK can return the rows behind a dashboard figure and drill into it, on a draft or a published board, turn a chart description into a figure spec without saving anything, and list, create and delete your workspace's chart types. They can also read watched folders, their run history and where named files would go, change the order of a folder's rules, and add a held file anyway.
mammoth run reads one command per line from a file or standard input and runs them all in one process, with one sign-in and one connection, so a batch of reads takes about half as long as running each command on its own. Every line gets its own result, and a failing line stops the run unless you ask it to continue.
Commands start faster, a project command no longer looks up the workspace every time, the CLI remembers which dataset holds a view from one command to the next, and each view transform makes fewer requests to the server.
mammoth view export download saves a view as CSV, Excel, JSON, NDJSON, XML, YAML or Parquet. Reading a view's data can return only the columns you name, a column's display format can be set from the CLI, conditional formatting rules can be replaced in one call, and date conversion takes several input formats, with a strict mode that leaves an ambiguous day or month blank instead of guessing.
Create dataset works on every saved connection, including ones AI built, and opens the connector page with that connection already in use and each of its tables read. On a connector AI built, Refresh list has the AI look for more tables, optionally ones you name, and shows the new ones under Just found.
Through the Mammoth MCP server, an assistant can change, pause or restore one step of a view's pipeline and is told which later steps a change breaks. It can change, pause, restore or run an export and see its recent runs and their errors, list the tables of a database the project is already connected to, and work in every project you granted it when connecting. Its list of connected connectors now shows only the connected ones.
Tables opened from the free PDF to CSV tool land as ready datasets, with section headings and repeated header rows kept as data and dates that could be day- or month-first read in your project's order. With only one project to add them to they go straight in, and the Datasets page shows them arriving.
On the Datasets page, a folder an automation watches shows where its files go, and each file dropped in says whether it was added, with how many rows, or why not, such as new columns, a duplicate, a password or sheets to pick, with the next step. A duplicate can be added anyway, a file goes to the first matching rule in an order you set, the bell rings only when files don't go in, Monitor and each automation's History list the runs, and workbook passwords are not kept.
New dataset → Connect is one search box over every source, and a source Mammoth doesn't list yet can be built by AI from a name, a sentence or a link: the AI confirms what you meant, builds the connector, and its page shows each table it can read with sample rows. Tick the columns you want into Your table, and Save makes a dataset that refreshes daily.
Click a chart mark to filter the pages it reaches, see up to 500 rows behind it when signed in, or drill down one level with a breadcrumb back. A board can read a different table on each page, up to five, all from its own project, and new chart kinds include Icon stat, Pictograph and charts you describe, which can be saved as your workspace's own chart types.
A motion story can start at 0.5×, 1×, 2× or 4×, Ask about this moment stays on the moment you asked about, and title edits are saved. In the Dashboards library you can tick several boards, or shift-click a range, and add them all to a collection, and a collection that isn't shared keeps drafts as drafts until you share it.
Every page has one back arrow in its top-left corner that returns to where you came from and names that place on hover. Opened from a link or after a reload, it goes up to the page's parent instead.
The Dashboards library opens with its cards sooner, offers list or tile view, and sorts by last updated, most or least viewed, or date created. Cards keep the title readable, and a double-click opens a board directly.
Motion stories gain a weekly grain, a choice of date column and range, a line race chart, following one entity, pauses and pins on events, last year's values shown as ghosts, and Ask about this moment. Video exports in landscape, square or portrait.
The free PDF to CSV tool accepts any email address and sends one Open tables button, which opens the tables in Mammoth; a new address gets an account on that click, with no password to set.
A Data Consolidation automation shows Running, Needs review or Failed beside it while it works, so you can see when it is processing files or has stopped on files that don't match.
mammoth dataset list reports how many datasets the project has, and with names set to true returns every dataset's id and name in one call instead of page by page.
mammoth dataset refresh fetches a URL dataset's link again, and dataset and view lists show a URL dataset's source as its link's host. dashboard delete and dashboard share report what they did, and the SDK's embed settings call changes only the settings you pass.
Paste, drop or attach a picture, PDF, Word file or text in the editor chat of a dashboard, doc or deck. A screenshot can restyle the board to match its layout, chart types, colors and font once you approve a card showing what carries over, and a brief can steer a rewrite, and where a figure in it matches one in your data the two are shown side by side, while a figure your data cannot support is named as such; spreadsheets are pointed to Use my data instead.
Turning Explore filters into a pipeline task or a new view now opens from a menu beside the filters chip, shown whenever filters are on, and a column's value list says how many distinct values it has. Long names wrap inside delete dialogs.
Someone invited to a workspace, or added to one by an admin, signs in with a code emailed to them instead of setting a password first, and the email opens the workspace they were added to.
The Unpublished changes pill shows a count and opens a list of what changed since the last publish, who changed it and when, with Publish and Discard; Discard names colleagues whose edits would be lost and offers Undo. The embed panel asks one question, all pages, one page or one chart, and offers only what is published, which is what an embed shows.
The published dashboard viewer and Your dashboards show which account you are signed in as, note when the numbers you see are limited to your rows, and let you sign out or use a different account, which brings you back to the same dashboard.
mammoth how followed by a goal returns the commands to run, in order. An unknown command, option or column gets a did-you-mean suggestion, help lists every allowed value, mammoth capabilities lists known limits with their workarounds, and results suggest the next commands to run.
A Connections tab lists every connection a project uses, whether your own, Mammoth-managed or a live link, with what each one imports and exports. Every connection is health-checked daily, so a broken one shows up before a refresh fails, and an export's credentials link to the matching saved connection instead of being stored twice.
Download a view as JSON, NDJSON, XML, YAML or Parquet beside CSV and Excel. Exports to Postgres, MySQL, SQL Server and Redshift can append or upsert instead of replacing the table, Snowflake, Databricks and Google Sheets join the destinations, and a saved export can run on its own schedule, send only rows newer than the last run, and keep a run history with an email when it starts failing. One-off download files are removed after seven days.
On the REST API export form, paste a link to the API's documentation, with notes if you like, and AI fills in the form for you to review, along with what you'll need and any caveats. Nothing is saved or sent until you apply the form, credentials are never filled in, and Undo puts the form back.
Every function in the functions panel previews its result as soon as its settings are complete, in a window over the grid you can move, resize or minimize: rows and columns before and after, the columns it adds or changes highlighted, and tabs for changed values, removed rows, duplicates, groups and join matches. Smart fill previews when you click Run preview, and a new fill, top or bottom rows, or window step uses the sort it shows.
Bring in a JSON, JSON Lines or nested XML file by upload, link, connector or folder automation and Mammoth turns it into a table: the main repeating records become the rows, with nested fields as columns. When a file could be read more than one way, pick the list to use as rows or describe what you want, and the reading is saved and reused for later files.
Through the Mammoth MCP server, an assistant can total, count and average a view by group or by day, week, month, quarter or year, with Mammoth doing the arithmetic instead of the model. It can read each dashboard's KPI and chart values, link to the matching page in Mammoth, list the connectors your plan includes, and send you to Mammoth to connect one, so no password passes through the model.
mammoth auth login signs in to the CLI through your browser, or with a device code on a machine without one, instead of pasting an API token. Signing in again warns when it ends another stored sign-in, and signing out of one that has already ended says so plainly.
The Datasets page is redesigned, with a folder path and toolbar above one-line rows and a side panel you can drag to any width that opens on the dataset you just added or last had open. Its Summary card gives a one-sentence AI summary and what the dataset's tasks did, from rows in to rows out, and where the data goes; Explore opens the editor on suggested cards and Transform on the pipeline. A dataset with several views has a views switcher showing each view's health, and Move to trash can be undone.
Every upgrade button opens one plan picker inside the app, and a workspace on a trial can pay for a plan from it; backing out of checkout leaves the trial running.
Select pending batches in a dataset's batches list and delete them. Column mapping no longer needs a separate Validate step, because Apply runs the check itself, and the unmapped-column count is correct.
A dataset made from a web link keeps that link, and Refresh from URL fetches the file again and runs it through the dataset's recipe. It works on demand for a link to a single CSV file, and the new rows are added or replace the old ones according to the dataset's update method.
A dataset's activity history can include the events of its views, and search matches dataset and project names, so a view called View 1 can be found by its dataset. Exporting one object's history works, and the log pages reliably as you scroll.
The CLI and SDK gain commands for editing explore cards of every kind, exporting a card as an image and adding it to a dashboard, reading raw view data including the time in dates, and working with collections, the engagement report and templates with your own data. A schema diff compares two datasets before an append.
Settings > Connected apps has a Connect an AI assistant card with the MCP server address to copy and setup steps for popular AI assistants.
PDF extraction can run on your workspace's own AI provider key, picked from a list showing each key's provider and model.
When you connect an AI assistant over MCP, tick Read-only on the consent page and it can read your data but change nothing. Connected apps shows each app's access, and reconnecting never widens it.
Dashboard cards sit in full-width rows with no gaps: resize a row from a card's corner or the row's edges, drag a card into another row, reorder KPI tiles, and undo layout changes. A run of hand edits shows as one line in the editor chat, with the details on hover.
A SQL step generated from a description is checked against your actual rows before it is offered, so asking to blank some values no longer drops the column or deletes rows. You can refine a generated query, see how your request was read, and keep your prompt; a SQL step reads only its own view.
The Extract JSON rule can read a key nested inside objects and lists, such as a company's location country, in one rule instead of one rule per level. Suggested keys list every nested value, and key suggestions load again.
The engagement report opens on one sentence saying how many of the people a dashboard is shared with opened it, followed by three headline figures and views per day. The Audience tab is slimmer and the library's board panel uses the full height.
On the dashboard site, readers land on Your dashboards, which lists their collections and the dashboards shared with them, and a collection, a shared board and the no-access screen each link back to it.
Open the activity history of one dataset, view or automation, including events that used it as a source or destination, and search the Activity Log by name.
New CLI commands create one filtered view per value of a column, read and set a view's explore panel, add a chart to a dashboard, overwrite a column with an expression, take the first word of a text column, and return the public URL of a live link. A filter that removes no rows today can still be staged with an opt-in flag, and sending rows to a new dataset no longer needs a name.
Start a dashboard context by typing or dropping files, and review what was read before saving; the name and slots fill themselves in. A slider sets how strongly a background image shows in a title band or cover. Published dashboards open on their first page, then on the page each reader last viewed, and collections gain a Set order control and a collapsible shelf.
See how a shared dashboard is being read: how many of the people it is shared with opened it and came back, engaged time, which pages and tiles were seen, each person's visits, and the days and hours people read it. Send a reminder to people who have not opened it yet.
Connect an AI assistant or any other MCP client to your Mammoth workspace: add the server at mcp.mammoth.io/mcp and sign in with your Mammoth account. Available on every plan.
Upload your own file to a dashboard template and keep working while it is prepared; the board shows progress and a notification says when your version is ready. AI proposes your version of the template, including calculated columns added as pipeline steps and extra tiles for columns the template had no place for, and you review every change before applying it. When a file does not fit, you get options instead: reshape a month-per-column file, pick a template that fits, or build a new dashboard in the template's design.
Settings shows API tokens and external API keys in two lists, newest first, with who created each key, when, and its scope. A new key's secret is shown once when it is created.
New CLI commands and SDK methods cover dashboard embed usage, dashboard format previews and audiences, workspace search, folder paths, resuming a paid plan with a scheduled downgrade, checking what a delete will affect, copying a project, finding a dataset by name across projects, every explore card option, and reading several schemas in one call.
Group several dashboards into a named collection and share it as one; people invited to it can open every dashboard in it. Viewers get a home page of their own with their collections, and published dashboards open at a fixed width by default.
Deleting or restoring a task, view, dataset or folder first shows what changes downstream, including the exports and dashboards that depend on it, with links to open each one. Dashboards now go to trash and come back together with the views they are built on. Restoring from Trash also lists the exports that come back with an item, and a view can be restored without its dashboards.
When an uploaded file holds several sheets or several tables on one sheet, preview each one and create a dataset from every table you want, in one pass.
Apps that support OAuth can connect to your workspace once you approve them on a Mammoth consent page, choosing the workspace and project they may use. See and revoke the apps you have approved under Connected apps in settings, and workspace admins can register their own OAuth apps.
The sign-in pages and the sign-in code email fit a phone screen, with full-size fields and buttons; desktop and tablet are unchanged.
The duplicate check shows sample duplicated rows instead of only a count, variant spellings say how each one differs, a project check can run against a single view, and creating a dataset reports the final name it was given.
Appends, replaces, batch creation and database exports run from the CLI now check the destination first and refuse a mismatched column type, an incomplete column mapping or an unconfirmed table replacement. A dry run names the exports a pipeline rerun will fire and any columns they would leave blank.
Create one conditional-format rule that covers several columns, or the whole row, from the CLI using column display names.
Find datasets by name, column name or a sampled value with a new search command in the CLI and SDK.
Explore from the CLI and SDK can return a running total per bucket, figures are rounded for display, and dataset find can match on columns.
Send a chart from an explore card straight to a dashboard. Explore cards also gain a standard deviation measure and keep banded axes in their natural order, and your explore state is kept per user.
Read a project's Monitor Needs Attention list from the CLI and SDK, including pipeline errors, pending data and checkpoints awaiting approval, with the dataset each item belongs to.
List pending workspace invites, and revoke, resend, delete or change the role of an invite from the CLI and SDK. Removing a user from a project from the CLI works again.
While editing a dashboard you can see which colleagues have it open and where they are working, and a conflicting save names who made the other change. Before you publish, you see whose unpublished changes the publish will include, and they are told when it goes out.
The Python SDK is now asynchronous: calls to the API are awaited and the client works as an async context manager. This is a breaking change for scripts written against earlier versions; the CLI works as before.
AI-built custom connectors run their generated code in an isolated environment, and can now export an entire Elasticsearch or OpenSearch index rather than stopping after the first page.
See who reads your dashboards, documents and decks: where they are, what device they use, how far they get and what they ask. Editors get a summary in the editor header, notices about notable activity and a weekly digest.
A dashboard now belongs to the project of the data it was built on, and that project's roles decide access: Admins and Editors build and edit, Viewers see published boards. The Dashboards page shows each board's project and can be filtered by project; sharing with people outside the project stays view-only.
A dashboard's buttons each do one job: Sync pulls new data into your draft, Publish makes the draft live, Share sets who can see it, and Updates says which of these is waiting. The Share dialog is one column, and changing who can see a board no longer marks unpublished changes as published.
The Source Interpreter, which uses AI to read a messy file, reads messy spreadsheets more reliably, including timesheets, stacked headers and PDF-style numbers. Excel files no longer import as a single cell, and special characters are no longer garbled on import.
One click re-aligns a hand-arranged board so edges line up, widths are even and each row shares a height. Title band edits now stay put after a save, and the preview shows the look you saved.
Title bands gain a soft tone that suits light and dark styles, line and wash textures, and a short label above the title. The Editorial theme's headings use a more contemporary serif.
CLI commands that change data now report the resulting state, flag a change that has not settled or that broke a downstream export, and can hand back a still-running job instead of timing out. Data warnings point out values spelled several ways and condition values are checked against their column type before they are sent.
When a messy upload or Google Sheet is offered for reshaping, you can choose Import as is to keep the data exactly as it was read, and future files for that dataset are imported the same way.
The free PDF to table tool keeps numbers exactly as printed, keeps separate tables apart, gives tables readable column headers and drops footnote markers from labels. Tables are no longer lost when a page is hard to read, and a stalled extraction ends with a message instead of spinning.
Start a dashboard from a ready-made template inside Mammoth. Each template opens on sample data first, then moves onto your own dataset, with a rating that says how well your data fits it.
When you attach or edit the context the AI uses for a dashboard, a review shows exactly which items would change before anything is applied, and accepting it saves a version you can undo. Requests such as "lower is better" now land on the board, and suggested wording cannot introduce terms your board and data do not support.
Every chart's menu opens a panel listing the columns it reads and how each is used, including the filters applied to it. Editors also see the query behind the chart.
Five templates for venture-backed companies and their investors: SaaS Metrics, a startup board report, an investor update, a startup board deck and a portfolio KPI rollup. Each opens on sample data.
Moving a template from sample data onto your own is smoother: you can add new data without leaving the board, the preview draws your real numbers, and the fit rating says plainly when a dataset cannot fill the board. After the swap, insight text names your columns and charts that cannot trend on your dates are left out.
Opening a workspace link you cannot access now explains why and offers a workspace switcher, instead of silently redirecting. If you have no other active workspace, you can create a new one from that page or from the suspended-workspace page.
Summarize a view's data from the command line: group and aggregate, bucket dates and numbers, and explore a single column as a trend, distribution or list of top values.
Each command that changes data reports whether the change landed, with row counts before and after, and flags writes that empty a view or joins that match poorly.
Download a dashboard as a Power BI or Tableau file from the Python SDK or the command line.
The command-line tool can manage dashboard embedding: embed settings, keys, allowed sites and usage.
An API token can cover all projects in a workspace, including ones created later, and changing a token's project takes effect immediately. Members without permission to manage tokens see why the option is unavailable.
The New automation wizard starts with the destination, tracks which steps are complete, and keeps a running summary of your choices alongside, across all automation types.
Paste a list of addresses, such as an email's To: line, into a dashboard's invite field to share it with everyone at once; names and duplicates are cleaned up for you.
Set how dates are shown for a whole project; every new dataset's date columns follow it, whatever their source. New workspaces also start with US or UK date order based on where you sign up.
When a page is too large to extract, the free PDF converter keeps the tables from the other pages and explains which pages were skipped and why.
REST API connections can authenticate with OAuth 2.0 client credentials, using a client ID, client secret, token URL and scopes.
If a PDF page is too large to extract within the plan's limit, that page is skipped and the tables from every other page are kept, with a message saying which pages were skipped.
New projects read dates as day-first or month-first from your browser's time zone, turned on by default, and the setting folds to a one-line summary you can open to change.
API tokens are now a single token issued by Mammoth, shown once when created and sent as a standard bearer credential, replacing the earlier key-and-secret pair.
Dashboards can chart trends by day or week, and charts can show minimum, maximum and median values.
Signing up from the free PDF converter now brings every table you selected into your new workspace, not just the first.
Upload HEIC photos and TIFF images, including multi-page TIFFs, and extract tables from them like any PDF; the file keeps the name you gave it.
A PDF or Word document dropped into a dashboard's context is now read by the AI that builds the board, not refused.
HEIC and TIFF images can be previewed and converted in the browser.
The two-factor sign-in screens have a refreshed design, and sign-up can verify your email address with a one-time code.
When an appended file's columns do not match the dataset, it waits as a pending batch, and you can map its columns to the dataset's and check the result before adding it.
A new comparison page picks two meaningful groups from your data to compare, so it never opens empty. Presentation templates now preview with a sheet of their slides instead of just the cover, and chart slides no longer clip their axis labels.
Put a live dashboard, or a single tile of it, inside another website or app with an embed code; it keeps updating as the data does.
The free PDF converter has a refreshed design. With a personal email address, it now offers to open your extracted tables in a free Mammoth workspace instead of only rejecting the address, and the option to automate the extraction appears before the email step.
A view's Data In / Data Out menu now lists the dashboards it feeds, with a direct link to each one you have access to.
The connection-count limit is gone from the connector picker: you can create as many connections as you need, and every paid plan includes the full connector catalog.
Before you commit an uploaded file, PDF or Google Sheet, the review step lists each column with the type it was read as and its first values, and lets you choose manual or scheduled refresh.
Q&A answers are set at a readable width and can use paragraphs and bullet lists. The Q&A page fills the screen, fits on a phone, and redraws its charts as the window changes size. Saving a context no longer drops text still in the paste box.
New users land directly in their workspace after signing up, with no profile form to fill in first. Someone who chose a template from the public template gallery lands on that template.
The build screen names the format being created, publishing sits behind a single button with its options alongside, Q&A turns show when they were asked, and a data-failure banner can be dismissed.
Upload a messy CSV, Excel file or PDF and AI works out how to reshape it into a clean table, showing a preview before anything is committed. The steps you confirm are saved as a recipe and applied to each new batch of data that arrives.
Plans now price storage separately from the plan itself, so you can add storage without changing plans. Plan limits are applied in the product, with guidance on what to do when a workspace is over a limit, and the Change Plan window lays out the options without clipping.
Q&A understands relative periods such as "last quarter," chooses a sensible time grain for trends, and describes only the data its chart actually shows. Answer charts use the full width of the card in the board's own colors, and can be expanded to full screen.
Project settings now separate how ambiguous dates in uploaded files are read from how date columns are shown, and state the default when nothing is chosen.
Managing columns in a view and editing a dataset's schema now happen in one shared Column Manager, with a confirmation before a change that affects the source data.
Dark mode has been updated across the app.
The dashboard editor now fits the board to the screen the same way preview does, whether side panels are open or closed.
Stack several CSV files or Excel sheets into one table: pick which sheets take part, line up columns that don't match, preview the result, and receive the merged file as CSV or Excel by email.
Browse ready-made dashboard templates on mammoth.io without signing in, each with a preview picture and its own detail page.
Bulk Replace can match values with a dynamic condition, and the matched values are worked out each time the pipeline runs, so the step card shows what was actually replaced.
Set rules that automatically delete or suspend a dataset's older batches by age, by how many to keep, or by a condition you describe, with an optional approval step and email notifications.
Turn a connected, cleaned dataset into a dashboard, a presentation or a document, built by AI from your real columns and kept current as the data refreshes. Change the style, layout and focus by asking in plain language, add a Q&A page readers can question, and share it publicly, behind a password or with signed-in viewers.
Creating an API token no longer requires a description, and the project is picked for you when the workspace has only one.
Import failure emails for SQL Server, PostgreSQL, Redshift and MySQL sources now explain a lost connection in plain language instead of raw driver errors.
When you add a dataset by upload, URL, clipboard, sample file, webhook or connector, a placeholder row appears in the Data Library right away while it is processed.
A dataset pulled from a third-party connector now shows which saved connection it uses, alongside its query, in the dataset's properties.
Numbers written with regional decimal and thousands separators are read as numbers when data is imported or converted.
Pipeline errors in Monitor now show when the failure happened and where the view lives, so you can triage without opening it.
Automation options are grouped with a short description of each, the create wizard marks completed steps, the destination folder is shown from the start, and running an automation manually is easier to follow.
Empty and space-only values show as [blank] and [whitespace x N] in Explore cards, filter details and pills, condition suggestions and Bulk Replace lists, so you can tell them apart.
Datasets and files show which automations act on them and how those automations are doing, so you can see at a glance what is running and what needs attention.
Work with Mammoth from a terminal: install with one command, sign in interactively or from a credentials file, and create folders, load data and run other operations. Commands wait for the jobs they start, ask before destructive actions and explain any usage error.
Creating and editing an automation happens in one guided window, and automation summaries show a short preview you can expand instead of stretching the page.
Select several files by dragging across empty space, then drop them onto a folder or a breadcrumb segment to move them, with an undo right after the move. Folders can also be expanded to show their contents in place.
Rename a dataset or view from edit icons beside its name in the preview panel, or from Rename in the dataset's options menu.
Adding a data check starts by picking the check type from tiles that describe each one in plain language, then setting it up in a single card. The value range check can now require every value to be negative.
Errors and warnings on a pipeline step appear at the top of its card, and stay visible as an icon with a tooltip when the card is collapsed.
Column name changes in dataset settings register as you type, without pressing Enter, so an edit is no longer lost when you close the window.
Checkpoint approval and alert emails show the dataset name alongside the view name in their link to the pipeline.
Search on the Data page finds a dataset or view by its folder path as well as its name and tolerates near-miss spellings, and the dashboard list shows each dashboard's sync status.
A snippet can hold any reusable value, not only SQL. When you add one to a rule, choose to reference it so it stays linked or paste its current value, and hover a snippet in the SQL editor to see its details.
Files waiting to be added to a dataset appear as pending batches on that dataset, marked when they need your action, and several pending batches can be tracked at once.
Checkpoint approvals and data check items in Monitor's Needs Attention tab open the view they belong to.
Adding or using your own AI provider key gives specific messages when a key, model or plan cannot be used, picks a sensible default model for each provider, and sets web search per provider.
The Checkpoint panel has a cleaner layout, with clear cards for notify-only and pause-for-approval, tidier email and webhook rows, and a collapsible webhook payload.
Copy one or more pipeline tasks and paste them into the same view or another view's pipeline; pasted tasks open as a draft, and database export steps keep their saved credentials.
A view with no rows to show now says so plainly, and a view whose source has no data no longer shows the wrong message.
Premium connector cards show how many of your plan's premium connections are in use, and the limit applies across all premium connectors together.
Deleting a dataset, view, dashboard or automation moves it to Trash, where you can restore items one at a time or in bulk, or delete them for good.
Branch Out, crosstab and BigQuery export steps show whether they replace or combine data right on the step card, without reopening the step.
When a pull from a connected source fails, you get an email so you can act on the error.
Bulk Replace places its result column right next to the source column, so you can compare original and replaced values side by side.
Monitor opens as a window over the page you are on, so you can check pipeline status without leaving your work.
The Parameters panel no longer flickers on hover, its edit window sizes to its content, and parameters are always workspace-wide, so there is no scope to choose.
A column you drag to a new position stays highlighted until you drop it, so you can see which one is moving.
Connect a mailbox and bring in the CSV, TSV and Excel files attached to incoming emails, without downloading them by hand.
Every change to a pipeline is saved as a version you can browse with its timestamp and task details, and you can restore an earlier version or copy tasks out of it.
If a workspace is suspended and you belong to others, you can switch to another workspace straight from the suspended screen.
Ask the dashboard chat for several edits in a row: they queue and run one at a time, so one edit no longer undoes another, and a waiting request can be changed or removed before it runs. A refreshing dashboard shows a loader instead of flashing white.
Pull messages from Apache Kafka topics straight into a dataset, connecting with plain or SASL_SSL authentication.
Subscribe to topics on an MQTT broker and collect sensor messages into a dataset, with column types worked out from the message contents.
Send a dataset to any REST API endpoint with POST, PUT or PATCH, using no authentication, an API key, a bearer token, basic auth or OAuth2.
When a subscription or trial ends without renewal, the workspace moves to the Free plan instead of being scheduled for deletion.
Add your own API keys from several AI model providers to use with GenAI steps and other AI features, with clearer validation messages when a key is added or switched.
Write a data check rule as a plain-language description and have it turned into an expression; turning off notifications on an existing check now clears its recipients.
Highlight cells that match rules you define, including color scales, and see a clear notice when a rule refers to a column that is missing or has changed type.
When connecting Dropbox, choose which folders the connection can use instead of the whole account, and pick files without being asked to reselect them.
Have Mammoth collect files from your cloud storage on a schedule, filtered by folder, and bring them in without manual downloads and uploads; sources can also be removed from an existing collection.
When a column holds JSON, its column menu offers JSON extraction as the first action, including in files with only a few rows.
Preview a dataset in a resizable pane, or expand it to a larger view, before selecting it for a join, lookup or other step.
Pipeline step cards for SQL and AI-built filters, metrics and math show both the plain-language intent and the query or condition behind it.
Send a dataset to Tableau Server as an export destination.
A new AI dashboard shows its pages as each one is ready, with progress you can follow and a way to cancel, and starting a dashboard from a suggestion keeps your prompt. This was the earlier dashboard product, replaced in August 2026.
The in-app plan picker and comparison reflect the current self-serve plans, including a Free plan, and features a plan does not include open an upgrade prompt instead.
Dashboards can show values on a filled map of the world, US states or India states, and filters set to zero or false also include blank values. This was the earlier dashboard product, replaced in August 2026.
The Resource Finder opens with its search bar ready for typing, opens automatically when you add a Join or Lookup, and selects the dataset for you when only one matches.
A banner counts down to a scheduled app update so you can finish what you are doing first.
Import data from and export data to Azure Blob Storage.
The condition builder has an updated, easier-to-read layout for setting up conditions.
Connect to Microsoft Dynamics 365 and bring its data into Mammoth.
Set a new member's workspace-level permissions in the redesigned invite dialog, and see invited users more clearly in User Management.
The Parameters & Snippets panel has an updated layout; parameters set to 0 now display correctly and long names no longer overflow.
Notifications arrive as they happen, without waiting for the app to check for them.
Choose which SharePoint document library to import from, for sites that do not use the default one, with a notice when the chosen location has no files.
Move through every section of the transform menu with the Tab key.
A failed checkpoint or data check no longer stops you renaming, reordering, sorting, formatting or hiding columns; only a failed task step does.
Math, Metric and conditional Filter steps accept a SQL expression directly, or generate one from a plain-language description, and the result can be reviewed before it is applied.
Send a dataset straight to Microsoft OneDrive from the export destinations panel.
Dataset schema settings no longer let you remove every column: at least one must remain, or a new one must be added, before you can save.
Connecting to SFTP uses one connection dialog, with password or private-key sign-in, and the file picker leaves out hidden folders.
The Transform menu groups its tools by category, lists SQL beside them, and has one box that searches the tools or suggests steps for what you describe.
Connect the REST API connector to services that require OAuth authorization, and reauthorize a connection when its access expires.
A new finder for choosing datasets and views in Join, Lookup, Branch Out, Group & Pivot and other steps lets you browse folders and search; in Join and Lookup it opens at the current dataset's folder and no longer offers the current view as the other side.
When connecting Google Drive, pick the folders Mammoth may use instead of granting access to your whole drive; imports are limited to files in those folders.
Group & Pivot accepts several value columns when the Columns section is in use, so you can aggregate more than one field in a single pivot.
Import data straight from Microsoft OneDrive by browsing your folders and picking one or more files.
Drag columns into a new order in a dataset's schema settings, alongside renaming, removing and changing column types.
The REST API connector can now pull from an endpoint once for every record returned by another endpoint, passing each ID as a query parameter, so data that is only available per item (such as per project) is collected for all of them.
The Fill Missing Values step shows an illustration for each way of filling gaps, including filling from above or below within groups, to make the right option easier to pick.
Pipeline steps are reordered by dragging the handle on each card, so clicking elsewhere on a card no longer moves it by accident.
The Remove Duplicates step can compare only the columns you choose, instead of whole rows, when deciding which rows are duplicates.
Project Admins can invite people from outside the workspace straight into a project, and project settings are limited to project Admins. Workspace members see API keys as read-only, cannot use premium connectors, and no longer see the Activity Log.
Data Update is now called Auto-Sync throughout the app, with clear on and off states.
Add a checkpoint to a pipeline as an Alert, which notifies recipients by email or webhook and carries on, or as an Approval Gate, which pauses the pipeline until someone approves. Pending checkpoints can be approved without opening the pipeline, and Monitor is now split from a separate Pending Items view for everything awaiting action.
Download a dashboard as a PDF from the dashboard toolbar; each tab becomes its own page. This was the earlier dashboard product, replaced in August 2026.
Add a data check to a pipeline to test the data as it runs, using project parameters if you like. Checks awaiting attention appear in Pending Items and are sent by email.
Adding an export to a pipeline starts with a step that lays out the destinations, from Mammoth datasets to external databases and live links, before you configure one.
Automations can be scheduled to run every hour, and an automation's summary shows the schedule it runs on.
Define parameters and reusable SQL snippets once at project level and use them across pipelines. Changing a value marks the affected pipelines as stale and lets you rerun just those; deleting one keeps its current value in the steps that used it.
A minimized pipeline view keeps more of the grid in sight, pipeline cards are redesigned and expand or collapse one at a time, and the function picker gains search, categories and tooltips on function names.
Import data from SharePoint: browse its folders, select one or several files or a whole folder, and refresh the dataset with the latest file contents.
Columns created by a SQL step are placed next to the columns they were built from instead of at the end of the grid.
Plan & Storage settings show storage used by each dataset and, when you expand it, by each of its views. The tab is available to every workspace, whatever its plan.
Rename columns in bulk from Dataset Settings: standardize names to formats such as snake_case or camelCase, or have AI suggest descriptive names.
Profile Settings have a new layout, and the security section shows when two-factor authentication is already on instead of offering to enable it again.
Automations pick up files in nested subfolders, run as soon as they are created, and check connection details when a data refresh automation is added or edited.
Connect to an SFTP server with your own private key and username, with an optional passphrase for encrypted keys. Password connections no longer try key-based sign-in first.
Monitor has an Automations tab showing running and queued automations, with live progress through validating, appending and finalizing, and a countdown for queued runs. A data consolidation automation now targets a single destination dataset.
When a payment is due, retry it with the card already on file instead of going through checkout again, and plan status updates promptly after payment or a trial converting to paid.
Choose the file name an SFTP export writes, and decide whether each run overwrites that file or creates a new uniquely named one.
Duplicating a view and querying its rows run in the background, so very large datasets no longer freeze the screen while they work.
Pipeline steps run without blocking the screen, so you can keep building the pipeline while earlier steps are still processing.
Upload XML files, including XML files inside a zipped folder, and work with them like any other dataset.
SFTP folders load one level at a time as you expand them, so servers with many nested files open quickly, and folder counts stay correct when you search.
New projects use your browser's local date format by default; setting a project-specific date format is now an opt-in.
Import files from a Google Cloud Storage bucket using a service account: browse the bucket, pick a single file or a whole folder, and pull new files on later refreshes.
Create a new workspace from the sidebar and choose its plan and payment there. Plans, add-ons and billing are managed in settings, and reaching a plan limit or a premium feature shows an upgrade path instead of a request to contact sales.
Send a dataset to a SharePoint location as an export step in the Data Editor.
Importing many files from an SFTP folder at once runs in the background, and automations that append a folder of new files into one dataset handle them together in a single run.
The Data Library, its search and storage settings stay responsive in workspaces holding a large number of datasets and views.
API keys in Settings are shown masked rather than in full.
The Project Settings window has a cleaner layout with search and role management, matches the project creation flow, and lets you set access to only you, everyone in the workspace, or selected users, adding several people at once.
When you describe a transformation in plain language, separate Apply Now and Add to Pipeline actions make it clear whether the change runs right away or joins your pipeline.
Joins and lookups made in another view now show up in the Dependent Views menu of the view they pull from, so you can see what relies on it.
The Edit Cells header shows more clearly that you are in editing mode, with a tidier layout and buttons.
Edit a query directly without extra clicks, and run it straight from the keyboard when creating a dataset from a query.
If the app open in your browser is out of date, Mammoth asks you to refresh instead of leaving screens unresponsive.
Join on numeric or date key columns using an inclusive range instead of an exact match, so rows pair up when a value falls within a set distance of the other.
Set the date format for date columns from the preview panel and file details, and have it apply consistently to previews and future uploads. Edit Data Settings drops the automatic option and only asks about columns whose dates are ambiguous.
Converting text to a date suggests the source format it detects and lets you enter a custom one when none fits; converting a date to text lets you pick the output format. The pipeline step shows "Source format" when a custom format is used, and numeric-to-date conversions skip the source format setting.
Dataset settings and the preview panel have a refreshed design with clearer icons, spacing and navigation, making dataset details easier to scan.
The Functions panel reopens in the state you left it, including an unfinished step, and the pipeline keeps its scroll position and preview when you move between views.
The export menu in the Data Editor groups its options more clearly and has streamlined collapsed and expanded views.
The Data Quality view groups issues, warnings and recommendations more clearly, so you can judge a dataset's health faster. It is always on, with no separate setting to enable it.
The item selected in the preview panel stays selected as you move between the Data Library, Dashboards and Orchestrations pages, and the Data Library reopens where you last were.
When a change you describe in plain words can't be turned into a step, the message now explains why.
A search box in the connector list under Add Data lets you find a source by name instead of scrolling.
An empty Data Library has a new starting view, and dragging files onto it to upload works more smoothly.
Webhooks now have their own entry in the left sidebar of Add Data, so creating a dataset from a webhook is easier to find.
When pulling data from a third-party source into an existing dataset, choose next to the dataset name whether the new data replaces what is there or is combined with it.
A progress widget shows your uploads while they run, and long file names display cleanly in it.
Group and aggregate data and build a crosstab in one Group & Pivot step, including joining text values together within a group; it replaces the separate Crosstab, Group-Aggregate and Concatenation functions.
Share a dashboard publicly by link or invite people by email, who confirm their address before viewing; dashboard owners can see total views, average time spent and who viewed it. This was the earlier dashboard product, replaced in August 2026.
Transform opens a single window where you either describe the change you want in plain language or pick a transformation tool, and it remembers which you used last; the task's details are shown before you add it, and the pipeline opens so you can see pending changes.
Changes to a dashboard appear in place as they finish, without reloading the whole page. This was the earlier dashboard product, replaced in August 2026.
The Data Quality window shows the view's quality score, completeness, critical issues and recommended fixes more clearly.
Add Data has a new layout with redesigned connector cards, new sample datasets, and clearer feedback when you drag and drop a file.
Bulk Replace can reveal spaces and other whitespace in values so replacements match exactly what you expect, and its suggested replacements are more relevant.
Export now sits in the top-right of the Data Editor and opens in the same side panel as other functions, with one consistent layout for every destination.
A new Insights panel in the Data Editor uses AI to point out key observations and possible data issues in the view you are looking at, with no separate setting to switch it on.
Set a date format for a whole project so dates are read the same way across its datasets, and when a file has ambiguous date columns you are asked to confirm the format once and apply it to all of them.
Publish as Dashboard in the Data Editor suggests dashboards to build from your data, or lets you describe the one you want. This was the earlier dashboard product, replaced in August 2026.
Describe the dataset you need in plain language, including the fields you want, and Mammoth creates a structured dataset you can start working with.
Asking the AI Prompt to hide, show or sort columns now changes how the grid is displayed instead of adding a pipeline step.
Paste copied rows straight into Add Data to create a dataset, without saving a file first.
Leave a workspace from User Management or your profile settings; if you are its only owner, you are asked to transfer ownership first.
Start an orchestration right away with Run Now, and existing data refresh schedules are now managed as orchestrations rather than from the dataset preview panel.
Connect to Apache Hadoop with a host, port and credentials, browse its schemas and tables, and bring them into Mammoth.
The side navigation bar, folder breadcrumbs and the Transform and Export menus have an updated design, and breadcrumbs list only folders from the current project.
Describe what you want to analyze and Mammoth builds a dashboard from your data, choosing the charts, summary cards, filters, layout and theme. Preview it for desktop, tablet or mobile, choose how it stays in sync with the data, and share it by link. This was the earlier dashboard product, replaced in August 2026.
Sign in with a one-time code sent to your email and switch freely between that and your password. The sign-in and sign-up pages are redesigned with simpler forms, and bot protection runs in the background, falling back to an email code only when a sign-in looks suspicious.
A new Orchestration panel replaces Automation and brings workflow operations together: refresh third-party datasets manually or on a schedule, combine or replace data when new files arrive, and send automated notifications about workflow updates, with an unsubscribe link in each message. Switching a scheduled orchestration back to manual runs now works.
Open a data quality view from the Data Editor to see a score for your data along with completeness, critical issues and recommendations.
In Bulk Replace, move all or selected values from one group to another, and choose whether searching for values is case-sensitive and whether it uses fuzzy matching.
The Copy Columns function has a new design in line with the rest of the functions panel, and copying into an existing column now shows the result correctly.
The Activity Log records more kinds of events, shows its columns in a clearer order and filters by date more reliably. Entries for projects that have been deleted are no longer clickable and say why on hover.
Datasets brought in from third-party connectors now have data update settings you can review and change, as uploaded datasets do.
Hover over a task in the pipeline to see when it was last changed and by whom.
Switching between views keeps your search and scroll position in explore cards.
A copied or newly added column now appears right next to the column it came from instead of at the far right.
Upload a PDF or an image, including from a URL, and Mammoth pulls the tables out of it into a dataset you can prepare like any other.
The SQL query rule now accepts queries that use UNION to combine results.
Workspace owners can delete their workspace themselves, which also cancels its subscription.
Copy every column name in a dataset to the clipboard in one click from the column browser or dataset settings.
A task created by describing what you want is applied to the pipeline immediately, unless the pipeline is in Draft Mode.
Clear a column's decimal formatting from the number formatter, and the column menu shows when no formatting is applied.
Metric cards and the metrics panel have a cleaner layout that is easier to read, and metrics refresh when the view's data changes.
Delete your own user profile from your profile settings.
Files you upload, including additional files and profile pictures, are scanned for viruses, and if a file is rejected you are told which one.
Choosing a plan uses a redesigned page for comparing the available plans, and you can still switch workspaces from it when you belong to more than one.
See a workspace-wide log of what people did to datasets, views, projects and automations, with search, filters by timeframe and activity type, event details, and export.
Asking for a SQL query in plain language produces more accurate queries that need fewer manual corrections, and you see progress while the query is being written.
Rule and function panels have clearer dropdowns, illustrations and descriptions, better-aligned controls and smoother cursor behavior, so building a rule takes fewer steps to understand.
Profile a dataset to see its structure, completeness, consistency issues and key statistics, with generated charts and recommendations for improving it.
The sign-in, sign-up, password reset and email verification pages have a new, simpler design.
When you switch a join to another table, key and column selections carry over wherever the new table has columns with the same names, so you don't rebuild the join from scratch.
Deleting a workspace cancels its subscription and suspends it, and a suspended workspace can be restored.
Review and manage the column renames, format changes and other display changes you have made to the data grid in one place.
The preview panel shows a resource's options as buttons, including moving it, instead of hiding them behind a menu.
The free PDF converter makes it easier to explore the tables it extracts, and only offers the projects you belong to.
Choose a plan when you sign up, check out on your own, and pick or change your workspace's plan from settings; a workspace owner can also delete a workspace themselves.
When draft mode is on, the pipeline toolbar says so plainly, and the draft banner sits at the bottom of the pipeline out of the way.
Sessions that sit inactive past their timeout are signed out automatically.
Source cards in the pipeline show more information about where the data comes from and its current state.
Explore cards can show values by distinct count, alongside the existing aggregations.
Exports to BigQuery can be limited to the datasets and tables you allow, with more flexible export options and better handling of credentials.
Paste a list of email addresses separated by commas, semicolons, spaces or new lines when inviting users, and it splits into separate invites; other multi-value input boxes split pasted lists the same way.
The Data Library shows how much storage each dataset uses, and the sidebar shows your workspace's storage use at a glance.
Ctrl- or Shift-clicking an item that is already selected in an Explore card deselects just that item and keeps the rest of your selection.
Pin columns to the left from a column's header menu so they stay in view while you scroll; pinned columns are remembered per view, for you.
Publish a view to a BigQuery database that Mammoth manages for you, with access details to connect your own tools, progress messages while it runs, and clear errors when something goes wrong. Column names, including renamed ones, carry over cleanly, and the export contains only your data columns.
Add and remove batch columns directly from the Column Browser.
Workspace owners can turn access to alpha and beta features on or off for their workspace from the workspace settings.
Editing cells is smoother: add rows, see edited rows highlighted, keep your pipeline toggle setting, and read tooltips more easily.
Turn images into a PDF with the free Image to PDF tool, which also works better on mobile screens.
Import data from NetSuite into Mammoth.
Reports now opens from the vertical navigation bar into a workspace-wide reports page, and the explore section inside a report works again.
API tokens and external AI provider keys now live in their own API Keys section of Settings, separate from General.
Settings shows every workspace user with their role and the projects they are assigned to, lets you change a user's role per project, and invites new users from a redesigned invite window.
Add your own AI provider API key in Settings for generative AI features to use. If the key fails validation, runs out of quota or hits a rate limit, you see a clear message and the Gen AI rule cannot be applied until it is fixed.
Describe a filter or transformation condition in plain language and Mammoth builds the matching structured condition for you to review.
Upload or select files and choose whether to add them as a new batch, combine them with existing data, or replace it, with clear batch and column mapping options.
The browser tab icon shows whether the current view's pipeline is processing, has an error, or has finished, so you can follow progress from another tab.
Edit your profile and change your password from a dedicated profile and security window.
Leading, trailing and extra spaces are highlighted in the grid and in Explore cards, so stray whitespace is easy to spot.
Turn on highlighting to see leading, trailing and extra spaces in cell values, so hidden whitespace is easy to spot before you clean it.
The redesigned action bar's Transform menu now includes Lookup, Extract JSON, Extract Text and smallest/largest values, and its Export menu offers SFTP, with several layout issues in the new interface cleaned up.
When new data arrives, the Gen AI rule runs only on the new rows instead of reprocessing everything, and it now handles up to 50K rows instead of 10K.
When converting text to dates, Convert Column checks sample values first and pre-selects the suggested format, including for incomplete dates, and changing the format no longer leaves the pipeline showing empty data.
Add people to projects in the same step as inviting them to a workspace, from redesigned Manage Users and Invite Users dialogs.
The Gen AI rule works out which columns are relevant to your instruction, so you no longer have to select them by hand, though you can still choose them yourself.
The workspace switcher shows pinned and unpinned workspaces separately, and lets you pin, unpin and reorder them.
The batch table wraps long names, shows dates more clearly and lists the most recent batch first.
AI Query can answer questions about a connected database while it is still learning the database's tables, lets you refresh what it knows about them on demand, shows a clear error when a query fails, and handles Xero data more reliably.
In the free PDF and image to CSV tool, add your own instructions to guide how data is extracted from an image, and downloads of results work again.