Sixteen upgrades shipped this cycle, spanning access, workflows, forms, tasks, and mobile. The highlights are below.

Role-scoped access

A workspace with one flat permission level means every new hire either gets too much access or has to wait on an admin for each new tool they need. Permission sets fix that by shaping the whole workspace around what each person actually does. Give an end user Tasks and Analytics and that's exactly what they get: a focused "Let's get started" home with a trimmed rail and no clutter. Grant a builder more — Mergebase, Workflow, and Agents alongside — and their workspace expands to match. People land in an experience sized to their role from day one, and teams get an enterprise-grade layer of control over who can reach what.

Global variables

Copying the same list of statuses, regions, or tax rates into field after field means a single change has to be made in a dozen places — and inevitably, one gets missed. Workflows can now define named values, calculations, and option lists once at the workflow level, available to every form field and node from there on. A variable can change as the workflow runs or be locked as a constant, with support for seven data types and multi-value lists.

Select Options extends the same idea to dropdowns and multi-selects: build a named list from a table, a data collection, or an existing field, then share it across multiple fields on a form, each filtering it independently. Define a value once and reuse it everywhere instead of maintaining copies.

Workflow validation, error handling, and recovery

A workflow that fails silently — or dies with a stack trace nobody but an engineer can read — leaves whoever's watching it with no idea what to do next. Workflows now catch problems earlier and fail more gracefully. Structural issues are flagged before a run even starts, and node-level errors surface right on the node with a clear inspector banner: a plain-language heading, the message, and technical detail on demand. Failures are typed — connection and auth, integration and API, AI node, timeout, permission and quota, and runtime — so each one comes with the right response: transient errors retry automatically, while the rest fail fast with guided recovery actions like Retry, Reconnect, Fix with Agent, Go to setting, or Contact admin. A run no longer stops silently; it tells you exactly what broke and how to fix it.

Run history and inspection

Debugging a run used to mean leaving the builder for a fullscreen dialog, losing your place on the canvas in the process. Run history now lives in a dedicated right pane instead. Every run carries a status badge with a hover tooltip, and live runs stream their steps in real time with a running spinner. Expand any step to inspect its input, output, and error detail, or click it to focus the matching node on the canvas. Nested workflows open in a subflow drill-in, and lifecycle actions — Replay, Refresh, End — appear only when they apply to the run's current state. Watching and debugging a run now happens in one place, without ever leaving the builder.

DocuSign task integration

Getting a document signed used to mean sending someone out of MergePoint entirely, into a separate e-signature tool, and hoping the result made it back into the right task. Tasks can now send DocuSign templates directly. Connect an account via OAuth, choose a template from a picker, and MergePoint auto-fetches its recipient roles, date tabs, and text tabs straight from the DocuSign API. Recipient email, name, and signing mode are editable per role, with chevrons to reorder the signing sequence. Signature approvals stay inside the task instead of moving to a separate tool.

AI field extraction

Turning a sample document into a form schema has always meant reading it line by line and typing every field in by hand. Point the field extractor at a document instead and it reads the file for you. Upload one or more PDFs and MergePoint renders each with zoom and page navigation while it analyzes — the page blurs during analysis and snaps crisp when its result lands. Extracted fields arrive in two groups, already in your schema and new suggestions, each showing a preview of the sample value it found. Select what you want with per-field checkboxes or Select all, then apply them in one step. Building a schema from a real document takes seconds instead of manual transcription.

Signature field

A form that needs a real signature has always meant routing the signer out to a separate e-signature tool and back. Forms can now capture a signature directly, inline. One component covers three ways to sign: draw it by hand with a mouse, stylus, or finger; type a name and render it in a signature-style script; or leave it empty until the signer is ready. Both captured states show the same "Signature captured" confirmation, so the field behaves consistently no matter how someone signs. Signatures export as an image on submission, work fully offline, and the typed mode stays keyboard- and screen-reader-friendly for anyone who can't draw with a pointer.

File upload field redesign

A single rigid upload widget rarely fits every situation — evidence photos on a job site need a different flow than attaching a signed PDF at a desk. File upload fields get a full overhaul to match: file-picker or adaptive capture mode, and list, chips, or gallery display, with configurable height and max file count. Chips show thumbnails and upload progress; gallery view adds a lightbox with touch support, drag-to-reorder, and a live camera capture mode. Mobile splits uploads into separate "Select photos" and "Select files" actions, so capturing evidence from a phone matches how people actually use one.

Rich text fields with inline bindings

Instructions and summaries written as static text go stale the moment the data behind them changes — a hardcoded "Order #1234" never updates once the real order number is known. Rich text fields now support inline data bindings so that text can read real values instead. Trigger the binding picker with @, format text with a floating selection toolbar, and insert bindings as chips that resolve to live values in preview and submission. Broken bindings are flagged automatically, and the inspector adds height, background color, and border controls. Instructions and summaries can now read back real record data instead of a static placeholder.

Date and time fields

Date entry, time entry, and date ranges used to mean reaching for three different components and stitching their validation together by hand. A new date and time field covers all three from a single component. Pick from a calendar popover with an adaptive grid, set time down to the second in 12- or 24-hour context, and jump with Today/Now or clear in a tap. Validation supports future-only, past-only, explicit ranges, and formulas, and a date range can link two fields with cross-field checks. Defaults can be static, bound to data, or computed by formula, and a built-in duration engine formats elapsed time for display.

Overlay display modes

A form or detail view that always opens by navigating away breaks whatever context the person was just looking at. Forms and content can now open in an overlay instead — a centered modal, a side slideout, or an anchored dropdown — each with its own configuration for backdrop, alignment, max height, and gesture behavior. Actions and record views can surface in context without navigating away, so a task or form stays in view while a related detail opens on top of it.

URL parameters for forms

Sharing the "same" form with different teams or partners used to mean maintaining near-duplicate copies just to pre-fill a different default. Forms can now declare URL parameters in Settings and bind their defaults to system or session fields. Values resolve in priority order — URL value, bound default, static default — and flow through shared links, embed codes, and {{param:name|fallback}} tokens in field text, with captured values stored on submission. One form can be pre-filled and reused across many contexts from the link alone.

Formula editor field picker

Writing a formula used to mean knowing — and typing — the exact internal ID behind every field, with no way to double-check you had the right one. The formula editor now gets an inline field picker: type @ or use the binding button to search fields by name, filtered by expected type when relevant. Picked fields insert as {{binding_path}} tokens, or as raw member access in JavaScript mode, and render as chips that flag stale or broken references. Building a formula no longer means remembering exact field IDs.

Task list and panel navigation

Working a queue of tasks one at a time, with no memory of where you left off or what's already open, wastes time re-finding your place after every click. The task list now gets a resizable, collapsible left pane with time-grouped buckets, quick filters, and persistent layout state. Open tasks behave like browser tabs — click a card to open one, reuse the tab if it's already open, and auto-close with advance-to-next on resolution. Working through a queue stays fast and keeps its place.

Offline mobile sync

Fieldwork doesn't pause for a bad signal, but until now MergePoint mobile did. MergePoint mobile now works fully offline. Choose which tables sync via Mergebase, configure auto-insert of form values into Mergebase on submit, and rely on background sync to pull remote records and push pending local changes once reconnected. Fieldwork continues without a signal and catches up on its own the moment connectivity returns.

Speak your customers' language — everywhere in the workspace

A form or app that only speaks one language quietly turns away every respondent who doesn't read it — or forces them through a translation they didn't ask for. MergePoint now supports multiple languages across forms and the mobile app, so the same form can be filled out correctly in the languages your customers actually use. Set the languages your organization supports once, let individual accounts or forms narrow that down further, and translate content on demand with AI instead of managing translation files by hand. Drift detection flags content that's changed since it was last translated, and an author review panel keeps a human in the loop before a translation goes live. Desktop and mobile web platform UI (the app's own chrome, not your form content) stays in English for now; locale pickers flag any language a form doesn't yet support, and content without a translation falls back to its original text rather than showing a blank field.

Fixes

Fixes from this cycle are tracked in their individual feature changelogs.