Changelog

⌘Z steps back through the edits you just made — a mis-drag that rewired a connection, a step replaced from the catalog, a delete — and ⌘⇧Z steps forward again. It covers step configuration as well as the shape of the graph, and one gesture is one undo step. Run past the start of the session and it carries on into version history rather than stopping dead. Copy and paste work too: ⌘C and ⌘V put a configured step wherever you want it, settings intact, under a name that won't collide with the original.

A branch is where your workflow decides — and until now, deciding meant writing JavaScript. Build it from dropdowns instead: pick a field, pick a test, fill in a value. The field picker offers the same data as the {{...}} picker, the tests are filtered to what that field's type can actually answer, and the value control matches — a toggle for a yes/no field, a number for a number, a date for a date. Underneath, a live verdict says what this condition would decide with the data you have right now, so you can tell a branch is right before you run it. Expressions are still there for the rare thing rules can't say, one click away and reversible.

A new Wapp that runs a content pipeline end to end, replacing Content Calendar in the store. Every Monday it plans the week onto the calendar — real slots, real channels, ideas attached. Move a piece to Drafting and it researches the topic on the web, writes the draft in that channel's voice, and fans out cuts for your other channels, each parked at Review for your verdict. Approved pieces post themselves to X and LinkedIn from your own logged-in accounts through the Companion, capture their permalink, and a daily pass records how each one did — so next Monday's plan learns from what actually landed. Email an idea to the Wapp's own address and it lands in the backlog.

A finished run used to report itself in the language of the machine that executed it: a raw id for a title, collection_change for what started it, and a "step trace" whose steps were named step-22626a2e-1bf8-…. It now reads as what happened. Steps appear under the names you gave them, with their icons, in the order they ran, and nested under the branch that chose them. Each one says what it did — "Amount is at least 5000 — took this path", "Sent "Big win!"" — and the header says how the run ended and what set it off: Deal "Kestrel Robotics" was updated. The raw data is still one click away under each step, for when that is the question you have.

A step that generates an image or a video now shows it — on the step itself, on the canvas, the moment that step finishes rather than at the end of the run. Click it to open it full size. Steps that produce something else say so in a line instead. The same pictures appear on the run page and when you Test Step, where a generated image used to be reported as a file path you had to go and find.

Every screen was built for a desktop pane and merely survived a narrow one. This is the pass that fixes it. Headers no longer stack — one identity bar, at most one tab strip, and nothing titled twice. Chat opens as a full-height sheet from a floating button, so the workspace keeps the whole screen and chat gets three quarters of it when you want it. Collections open on a record list, with the grid one tap away and columns that give way as the screen narrows, so you can finally see a record's name and its fields at once. The calendar gets a readable week instead of 25px day columns. And controls that used to be clipped out of reach — Trash, whole tab groups — now scroll.

⌘F searches every step by name and integration, lighting each result's path on the canvas behind the palette so you can see where a match sits before you jump to it. A chevron on any Branch, Parallel or loop folds its whole body down to a head that says what it's hiding — "2 paths · 15 steps". During a run the canvas follows the step that's actually running and hands control straight back when you pan away. And Auto-arrange puts a graph you've dragged out of shape back the way the editor would have drawn it.

Press ? on the canvas for a sheet listing every shortcut it has, grouped by what you're trying to do. Each button in the control cluster names itself and its shortcut in a tooltip that arrives as soon as you point at it. With a step selected, the arrow keys walk the graph — up and down along the chain, left and right across the arms of a branch — and Escape hands them back to panning.

The Send Email step gains a transport choice. Keep the Cheqq sender, or pick Your SMTP server and mail leaves from your real sending address instead of a cheqq.ai one. Your settings are verified when you save them — a wrong host or password fails in the form, not days later inside a run. Track Replies in Cheqq is the useful middle ground: recipients see your domain, replies route back into Cheqq, and reply-triggered workflows keep working. Send Gmail can now reply into a thread too, so a follow-up lands inside the conversation rather than starting a new one.

Ask for a Wapp in chat and a build card now ticks through the phases as they happen: workspace, collections and views, workflows and triggers, activation. When it finishes it offers to open the finished Wapp with a five-second countdown, and a Stay here if you'd rather keep talking.

A template update now brings the new version's workflows along with its collections and fields, so a feature never arrives without the automation that powers it. What you decided stays yours: whether each workflow is switched on, the Mailhook address people already forward mail to, and the schedule you set. Workflows you wrote yourself are never touched.

The Companion can now upload a file from your workspace — or an attachment on a record — straight into a page's file picker, so an automation that ends in "attach this image" finishes the sentence instead of settling for a text-only post. Content Studio's poster uses it to attach the image to a post, and refuses to publish a stripped version if the approved media doesn't make it.

Free workspaces were capped at two active workflows and one schedule, which meant trying a couple of templates hit an upgrade wall before you had learned anything. That cap is gone: install and keep as many workflows as you like, on any schedule frequency, with the step-run quota as the only thing that meters usage. Mailhook addresses go from 1 to 3 on Free, and 5 to 15 on Pro.

Sales Pipeline gains an Outreach collection — what you sent, when, and whether it came back — with a board grouped by reply status, a calendar of follow-up dates, and dashboard tiles for who owes you a reply. On top of it sits an email workflow that scores inbound enquiries for fit, opens a deal and replies in-thread for the good ones, and two days later chases anything still silent. It re-reads the record before chasing, so someone who replied in the meantime never gets a follow-up they've already answered.

Add the Cheqq bot to your Slack workspace — one Add to Slack click — or link it on Telegram, and a DM or @mention becomes a full agent turn. The bot acts as you: your Wapps, your integrations, your memory, your permissions, your billing. It streams its answer, runs the same tools it has on the web, pauses for approval when an action needs one, and posts real files back into the thread instead of links. Each teammate links their own account once, so a shared channel never blurs who asked. Discord is next.

Turn on Push in Settings → Notifications and workflow alerts, approvals, and reminders arrive as browser notifications — no Cheqq tab required. The settings screen was rebuilt around it: each category shows its channels as plain toggles, one Delivery control chooses instant or a daily, twice-daily, or weekly digest, Do Not Disturb has its own card, and per-Wapp overrides are picked from the Wapps you have installed instead of typed in by hand.

A generated clip is no longer final. Video mode gains a Generate / Extend / Remix control: extend continues a clip past its last frame, remix restyles it into something new. Attach the clip you mean, or just say "extend this video" and the agent edits the latest one in the conversation. Available on models that accept video input, billed at the model's own edit rate, and every edit keeps its lineage back to the original.

While the agent works, its reasoning streams into a live timeline, and tool calls narrate themselves in plain language — "Searching the web for supplier reviews" instead of a bare tool name. The timeline shimmers on the current step while running and folds to a one-line summary when the turn completes. Reopening a thread keeps the whole trail.

Chat opens on a centered composer with six creation modes — Image, Video, Websites, Docs, Sheets, Slides. Each mode has its own page with starter prompts that prefill the composer and wait for you, never auto-send. Hopping between home and a mode keeps your draft and staged attachments, and the active mode shows as a chip you can dismiss.

Manage models curates the model picker: browse the full catalogue grouped by provider — with search, modality icons, and price hints — and tick the models you actually use. The picker shows only those, on every device you sign in from. If you never curate, you keep the featured set, and the model you're using stays listed even when it isn't ticked.

Your monthly grant refills on the 1st; a credit pack doesn't. The meter now draws them as separate tracks with a legend saying what happens to each next, so an emptying bar tells you whether to wait or to top up. The allowances grew too: the free grant triples to 750 for accounts with no paid plan, Pro moves to 7,500 credits per seat, Business to 20,000. Packs hold 10,000 credits at $30. Adding a seat mid-month delivers its credits immediately, and long conversations now cost less: the part of a prompt the model has already processed is billed at a discounted rate.

Footnotes are now first-class: a [^1] reference renders as a superscript chip, its definition as a marked block, and both survive saving byte-for-byte. The "advanced formatting may simplify" warning is gone: if saving would change any of a document's original formatting, your first save keeps an untouched copy as an Original version in history — automatically. Prose also reads like a document now: a proper heading ramp, display headings, inline images, and table and code styling.

A web-enabled Field Agent now researches the web before it writes, so a cell asking for a market summary gets a researched answer with citations. "Make this photo professional" uses the photo: a file field named in the prompt travels with it as the image to edit. And Auto chooses models by quality, considering only ones equipped for the job — a run with web access always lands on a model that can search. Every image and video model available in chat is available to Field Agents too.

A length of time is no longer a number field you agreed to call "Minutes". Duration fields carry their own unit, so you can type 1:30, 90, or 2h 15m and read it back as 1:30. Choose hours-and-minutes or add seconds, sum and average them like any other number, and export them in the format you actually use.

A collection that records when something starts and how long it lasts no longer needs a separate end date. Point a Calendar view at a Duration field and every block is drawn to scale — a 90-minute intensive finally looks different from a 15-minute check-in. Drag the bottom edge and the duration updates; drag the top and the end stays put. The time grid now snaps to 15 minutes rather than 30, and subscribed phone calendars get events with a real end time instead of a bare timestamp.

Installing a store template gives you your own copy, and until now improvements published afterwards never reached it. Cheqq now tells you when a newer version exists — a dot in the sidebar, a card in the Wapp's settings, and a notification — and shows exactly what will change before you agree to it, including anything that changes how your data is stored.

Taking an update keeps your records, the fields and views you added yourself, and your Wapp's name and icon. New collections, fields, and automations from the template are added alongside them.

Typing a filter to find one record used to re-filter that view for everyone else looking at it, seconds later, with no signal at either end. Your filters, sorts, and column widths are now yours alone. When you do want everyone to see a change, Save for everyone publishes it deliberately and offers an undo. A view can also be locked, fixing its shared definition while still letting anyone filter their own copy — and viewers can now keep filters at all, which they never could before.

The onboarding checklist is now a widget you control: a pill in the app bar, a compact "Next" card, or the full guide grouped into goals. Progress is tracked per person, so joining a workspace where the work is already done no longer hands you a checklist that's mysteriously complete — and connecting an integration or inviting a teammate now ticks itself off.

Any collection with a date field can now be a calendar. Month, week, and day views with drag-to-reschedule and edge-resize; up to 20 date fields plotted as separate color-coded series; an Unscheduled panel for records with no date yet; and timed events that convert to all-day (and back) with a drag. A calendar can also be shared as a read-only iCal feed that Google Calendar, Apple Calendar, or Outlook subscribe to directly.

The MCP Servers tab is now a browsable catalog: 31 providers — Linear, Notion, GitHub, Stripe, Atlassian, Figma, Airtable, and more — each connecting with your own account through the provider's own sign-in screen, plus open servers like Hugging Face and DeepWiki that connect in one click. Bring-your-own MCP servers that require sign-in connect too.

Every tool on a connected MCP server now appears in the workflow step picker under its provider, with a form generated from the tool's own schema — a deterministic step, no model in the loop. A generic Call MCP Tool step remains as the escape hatch.

Plans, credits, and files now belong to the workspace, not the person. Buy a plan for the workspace you're upgrading; every workspace mints its own monthly credit grant; work done in a shared workspace bills that workspace instead of quietly draining the member's personal credits; and files, app builds, and generated media are shared per workspace, so two members finally see the same tree.

The Wapp store now carries 30 templates across six categories, and every one arrives working: seed data worth looking at, a dashboard, and at least one automation that runs from install. Each template gets its own store page built from its actual manifest — the collections, fields, and triggers you'll really get — with captured screenshots of the installed workspace.

A notification now opens where the work is — the collection it's about, or the exact record when there is one — instead of the Wapp's front door. Links inside a notification are clickable, long bodies open into properly rendered Markdown, and a run waiting on your approval can be approved or declined from the notification itself. A new Notifications page holds everything at full width, grouped by day and filterable to unread, and alerts from a single run collapse into one entry instead of repeating once per recipient.

Seeded records can now carry real files, so a template arrives looking like something you'd keep. Meal Planner installs with a photo on every one of its ten recipes, each linking to the real published recipe it names, and grids that hold attachments install tall enough to see them.

Deleting a Wapp is no longer forever. A Trash drawer on the Wapps page lists deleted Wapps for 7 days (30 on paid plans), and restoring one brings back its installs, workflows, and data links — workflows return paused so nothing fires on a timer with nobody watching.

The extension now stays signed in on its own, so agents keep their browser tools without a Cheqq tab open. Before an agent trusts a browser, Cheqq now checks it's actually reachable — an unavailable one costs a moment, not a stalled run — and a Check button in Settings answers "is my browser connected?" without starting a task to find out.

Cheqq now produces actual deliverables: PDF and Word documents from what you ask for, PowerPoint decks from a slide plan, and Excel spreadsheets whose numbers stay numeric — sortable, filterable, with frozen headers. Generated images can be embedded straight into a deck, so "make me a product one-pager with a hero image" is one request, not three.

A generated deck opens on a real canvas, not a static preview. Select, drag, resize, and rotate slide objects; insert text and shapes; recolor through the deck's theme so a reskin carries your edits along. Every completed gesture is one undoable version, charts render live from the slide's own data, and the downloaded PPTX matches what the canvas shows — same geometry, same chart, same stacking order.

Uploaded office files used to dead-end at "Preview is not available". DOCX now renders with its real formatting, PPTX pages through every slide, and spreadsheets display as tables with sheet tabs — from the Files page, from file links in chat, and from a record's attachment fields alike.

Attach a Word document, deck, spreadsheet, or PDF and ask about it. Extraction happens server-side, so any model can work with the contents — summarize a contract, pull the numbers out of a deck, or turn an uploaded spreadsheet into a collection.

Collection fields can now generate typed text, selections, numbers, product images, videos, and Deep Match relations. Stable @field references, manual or automatic execution, live cell states, bulk quotes, human-edit protection, web citation badges, and immutable history make high-volume generation reviewable and reversible.

Choose Create video from Chat's attachment menu, select an approved video model, and optionally animate a product image. Results play directly in the conversation and are retained as private workspace files.

Workflow AI Agent steps now support Video output with duration, resolution, aspect-ratio, and audio controls. Generated paths and metadata can feed later workflow steps, while model availability remains controlled by the admin approval registry.

Every accepted turn now creates a recovery point before Cheqq starts working. Restore a conversation to that point, fork an independent path, or undo a restore without losing the abandoned future. Forked threads preserve the correct conversation prefix, open with their messages and surface context intact, and appear in recent conversations.

Wapps and workflows now have a shared, collaborative history with Undo, Redo, comparison, and explicit restoration. Paid workspaces can name and pin snapshots. Supported deleted workflows, Wapp collections, and Wapp records remain recoverable in Trash for the plan's retention period. Wapp comparisons explain record additions, deletions, restorations, and field changes, and restored records update the grid and side editor immediately.

Apps can now live on your own custom domain, carry search and social metadata, and be remixed — any public App can be copied into an independent draft of your own. Previews of unpublished work stay private to you, published thumbnails are more reliable, and generated App URLs now live under cheqq.site.

Consecutive tool calls now roll up into concise activity groups such as “Working with Wapps” and “Worked with Wapps.” Groups stay collapsed until you open them, never close themselves while work streams, and use a subtle active shimmer. Web, browser, Wapp, and workflow actions are classified correctly, while delegated agents continue streaming their own progress and tool calls.

Markdown documents in your workspace now open in a full editor. Autosave captures snapshots as you write, and the new version history lets you browse and restore any previous state of a document.

The chat panel inside the workflow builder now keeps its own thread history, scoped to the workflow you're editing. Jump back into a previous conversation without losing the context of what you were building.

Admins can now choose which AI models are available across the platform, with each model's capabilities and pricing under their control. Workflow AI steps pick a specific model, and every step runs on exactly the model it was configured with.

AI call steps can now generate and pass along images, so loops that produce visual content — thumbnails, social assets, illustrations — run end to end without manual handoffs.

A new home for creating and organizing Wapps: start from a template with an onboarding video, then search, filter, pin, archive, and unarchive as your list grows.

Invite others into a workflow and control exactly what each person can do. Notifications and escalations are now tied to who's responsible for what.

Workflow runs now update live as they execute. Workflows can also carry multiple triggers — manual, webhook, schedule, or collection change — all feeding into the same flow.

Files in your workspace can now be previewed directly in Cheqq, no download required.

Build workflows on a real canvas: drag steps in from the catalog, connect them with edges, and group or multi-select nodes to keep large workflows organized. Auto-layout and zoom keep everything readable as they grow.

Workflows can now call external APIs, run custom code, and kick off on a cron or interval schedule, alongside the existing AI and CLI steps.

Email and password, magic links, or one click with Google or Apple — Cheqq now supports all three, with session management handled for you.

In-app, email, and push notifications keep you posted on what your Wapps and workflows are doing, with digest settings so you control the volume.