What's new

Cruma now plays back what it notices about your business, with its sources. Open a workspace's company page and choose "What I noticed" for an evidenced read, like whether your site is set up for AI assistants to find you, with each finding paired with something you can act on. Nothing is claimed without a source, and when Cruma has not learned enough yet, it says so plainly.

The app now keeps itself up to date. When a new version ships, your open tab moves to it on its own, so you no longer have to hard-refresh or clear your cache to see the latest. This also clears up a case where a stale page could get stuck pointing at an old sign-in link.

When you are getting set up, you can now hand Cruma what ChatGPT or Claude already knows about your business: just paste it into the getting-started chat, or attach your exported history, and Cruma reads it as a starting point you review before anything becomes standing knowledge. As your brief takes shape it also asks you a real question at a time, so the workspace it builds fits what you actually do.

You can now add Cruma as a connector inside Claude or ChatGPT. In Settings, under "For AI agents," mint a token and paste it into your assistant's connector setup. From then on you can tell your assistant to share what it knows about your business with Cruma, and it can read back what you've already approved. Anything it shares lands in your review first, never as a standing rule, and you can revoke a connector any time.

Once your crew has earned it by doing the same small thing your way, time after time, Cruma now offers to take that slice off your plate, within bounds you set. You can say yes and it just handles it from then on, or keep it asking you. Anything to do with money, or anything unusual, always still comes to you, and you can take it back any time.

When your crew is running several strands of one job at once, you now see the shape of it: each strand named, what it covers, whether it runs on its own or waits on another, and how far along it is. A busy run reads as "here is workstream A, here is workstream B" instead of one long list of steps, on your home, in Tasks, and in the live crew tray.

Your crew can now keep recurring work going on its own, on the schedule you set, and it still checks with you before anything leaves the building. Ask it to run the outreach every Monday or close the books each month and it keeps doing it, files the results in your room, and lets you know, while a would-send step still waits for your approval. You can pause, resume, or stop any recurring task whenever you like.

When your crew makes a set of related files for one goal or campaign, they now land together in one folder instead of scattered across your room, so a project stays in one place.

You can now delete or rename any file in your workspace, and it is really gone from your rooms, not just hidden on your screen. Deleting a file removes it from every room and folder and it can no longer be opened or downloaded; renaming it updates its name everywhere, including the name it downloads under.

Everything your crew makes is now a real file in your workspace, a document, a page, a spreadsheet, sitting in the room where it belongs and ready to download, right next to the files you upload.

Your crew now tells you the truth when it cannot find real information instead of guessing, owns it when a step fails, and can set up live tracking on anything you point it at. When a search comes back empty it says so plainly rather than inventing names or numbers, when a step like building an image does not work it tells you instead of quietly handing you something else, and asking it to "keep an eye on" a web page now sets up a real watch that pings you the moment something changes.

When your crew cannot reach the information it needs, it now tells you plainly and offers a real next step instead of a dead end, and it never makes the data up. If a source is unreachable, your crew says so and asks you to connect a source or offers clearly-labeled sample data, rather than leaving you with a bare "I hit a problem."

A long crew job that gets interrupted or runs out of time now shows you exactly where it stopped instead of a spinner that never ends. You see which steps finished, which one was in progress, and what is still ahead, and every bit of finished work is saved. Reopen the run later and the plan is right there, honestly marked as paused, so you can see what you have and pick up from it.

Your crew now comes with a starter playbook of proven business skills, ready on day one. From pricing strategy and unit economics to funnel diagnosis and meeting follow-through, your crew follows a built-in library of business know-how out of the box, grouped by which crew member runs it, so real expertise is there the moment you start, before it has learned a single thing specific to you.

Ask your crew for a document, deck, or spreadsheet and it makes the real file for you, using the skills it knows. Just say what you want ("a one-page Word doc summarizing these three points") and the crew builds the actual Word document, PowerPoint deck, or Excel model, files it as a draft in your room for you to open, edit, and download, and tells you what it made. Nothing is sent or spent, it is a draft until you say so.

Everything your crew hands you is checked and tidied to a professional standard first. Before a document, deck, or spreadsheet reaches you, it is quietly cleaned up: stray em dashes become plain punctuation, headings and numbers line up, and each column reads consistently. Anything that genuinely needs your judgment is flagged honestly instead of silently changed, and you get a short note of what was cleaned up and what needs your eye.

Everything your crew produces now carries your brand, not ours. Once you have approved your brand, the Word documents, slide decks, and spreadsheets your crew hands back wear your color, your wordmark, and your fonts, so every deliverable looks like it came from your company. With no approved brand, everything looks exactly as it did before.

You can now teach your crew a skill and approve it right where you wrote it, on one page. Draft a skill on your Playbook and it shows up there for your ok immediately, right beside everything the crew has learned that is waiting on you. No more being sent off to another screen to say yes. Approve it and it joins your crew's skills on the spot.

Your crew can now keep watch on any page you care about and tell you the moment something that matters to you changes. Point it at a page, say what to watch for in plain words (a price drop, a status change, anything important), and the crew checks it on a schedule and notifies you the instant it happens, with a clear note of what changed. It stays quiet when nothing meaningful moves, and tells you honestly if a page stops being reachable.

Your workspace now tidies its own storage as it grows, and nothing you can see or audit is ever lost. Older document versions and finished background records move to cold storage instead of piling up, and your crew brings any of them back instantly the moment you open one.

Your crew now tells you the moment a job is done, with the result one tap away, and tells you honestly if something could not be finished. Notifications stick around after a reload, so you never miss the finish.

Your crew can now make you a clean branded web page or one-pager. Give it the content, a headline, a few sections, a call to action, and the crew hands back a polished, on-brand page in Cruma's look, ready to share or save as a self-contained HTML file. You can also save any document as a branded web page.

Your crew can now build you a live financial model in Excel, with working formulas. Give it your assumptions and the lines you want projected, and the crew returns a real .xlsx where the numbers are wired together, so when you change an assumption the whole projection recomputes right in the spreadsheet.

Your crew can now resize, crop, convert, or watermark an image for you. Hand it a photo or a logo and the crew can shrink it to size, crop it to a shape, switch it to PNG, JPEG, or WebP, rotate or flip it, stamp a watermark on it, and strip the hidden location and camera data before handing it back.

Your Playbook now shows what your crew can do, not just what it has learned. A new Capabilities lane lists the built-in skills the crew can wield out of the box, grouped by area, each showing what it produces and how careful it is, so you can see your crew's reach at a glance.

Your crew can now fill out a PDF form for you. Hand it a fillable disclosure, application, or agreement and the crew fills in the fields and returns the completed PDF, with an option to flatten it so the answers are locked in and can no longer be changed.

Your crew can now redline a contract. Hand it an agreement and the crew marks up a revised version with real tracked changes and margin comments, ready to open in Word and accept or reject line by line.

Your crew can now read the Word documents you share. Drop a .docx contract, letter, or agreement into a chat or a room and the crew reads its text and tables, and you can open it right in the app to see the content, no download needed.

Your crew can now hand you a real PowerPoint deck. Open any write-up and save it straight to .pptx, with a title slide, a slide for each section, and your tables laid out ready to present.

You can now download any document as a Word file. Open a write-up and save it straight to .docx, with its headings, text, tables, and lists intact, ready to hand off or keep editing in Word.

You can now upload Excel files. Drop an .xlsx workbook into a chat or a room and the crew reads it and opens it as a clean grid, the same way it already handled CSVs.

Your documents can now hold a live table or a design right inside them. A crew write-up can drop a forecast table, a task board, or a launch cover into the middle of the writing, and you read and edit it right there in the flow. Click the table to open it full size, change a cell, and it saves back into the document. Click the design to open the canvas editor on it.

Your designs are saved for keeps. Open the design canvas, lay out a poster or a one-pager, and your work is filed alongside everything else, exactly as you left it when you reopen it. You can also read any design as a clean document and export it to PDF in one click.

The Document view now takes you straight to editing. Reading your file and see a line you want to change? Click it and you land in the editor right on that line, ready to type.

Your boards remember how you like to see them. Switch a table to a Kanban board, a gallery, or a calendar, and that choice sticks the next time you open it.

Your crew's memory now keeps what matters sharp and folds old noise into wisdom. Facts you and the crew actually use rise to the top, while stale ones you never touch quietly sink instead of crowding the crew's attention. And when dozens of similar one-off notes pile up ("invoice sent", "invoice sent" ...), Cruma proposes a single durable takeaway ("invoices are net-30, usually paid in about 12 days") to your review card. As always, nothing new steers the crew until you approve it.

Booking and acting just got real: the crew now reaches the tools that matter on each app you connect (create an event, draft an email, update a deal) instead of an alphabet-ordered grab bag of admin functions. "Book a meeting Tuesday" now produces the actual booking card to approve.

The crew now checks its own answer before replying: if what it wrote does not actually accomplish what you asked, it takes one more pass to finish the job or tells you plainly what it could not do, instead of replying as if the work happened.

You can now tell Cruma to stay out of things: say "keep out of the encryption files" in chat and it sticks immediately, workspace-wide. Files matching your rule never reach the crew, on any connected tool, and when something is skipped the crew says so plainly ("1 file withheld") instead of pretending it saw everything. Change your mind later? Forget the rule in Settings, Memory and the crew can read those files again.

Your crew can now use tools from external MCP servers you connect, right alongside your app integrations. Point Cruma at a compatible service (like an agent or data tool that speaks MCP) and its read tools show up for the crew to use on your behalf, under the same review-and-approval rules as everything else, so it stays safe by default.

Your crew now catches its own mistakes. When it grades one of its finished runs as weak, even one that looked successful, it proposes a lesson to your review card so the same slip is less likely next time. As always, nothing changes how the crew works until you approve it.

Your crew now knows its way around your tools. Ask about your email, calendar, CRM, or any of hundreds of connectable business apps and Cruma reaches the ones you've connected, no magic words needed. When an app isn't connected yet, Cruma says so and offers to connect it right there in the chat, instead of telling you it can't help.

Connecting an app now confirms itself: when you come back from authorizing, Integrations tells you plainly that the app is connected and the crew can use it. No more wondering if it worked.

You can stop the crew mid-run: a Stop button sits right on "Cruma is working," and clicking it halts the work cleanly with a plain "Stopped." from Cruma, never a spinner you have to refresh away. You can also just keep typing while the crew works: a follow-up like "actually, keep it short" folds into the run already in flight and shapes the answer, and a message that lands just as a run finishes gets its own reply instead of being lost.

Your crew can now read the PDFs you share: attach a contract, invoice, or listing and Cruma reads its text and works from it, the same way it already reads spreadsheets. A scanned or locked PDF is handled honestly: Cruma tells you it can see the file but can't read the words, instead of guessing.

Click a file in a chat or a room to open it right inside Cruma: PDFs show inline, spreadsheets and CSVs open as a clean read-only grid, and anything else offers a one-tap download. No new tab, no leaving your work.

Every document now has a polished "Document" view: switch from the editor to see your file laid out as a clean, branded letter, and export it to PDF in one click. What you see is exactly what prints.

You can read what Cruma remembers, in the open: Settings, Memory now lays out what it knows about your business and how you like to work, grouped and labeled, each entry marked personal or workspace so you can see who can see it. Get something wrong? Correct it in place or tell Cruma to forget it. Personal entries stay yours; teammates only ever see workspace entries.

The workspace now opens with a real first draft: while you talked, Cruma drafted the most useful thing for your stated goal and filed it in the right room. If the conversation was too thin to draft anything real, it simply does not pretend.

Two-step verification: protect your account with an authenticator app from Settings. Once it's on, signing in asks for your 6-digit code, so a password alone is never enough.

Your crew now works while you sleep: email crew@yourworkspace and by the time you look, a reply is drafted and waiting for your one-tap approval. Nothing ever sends without you.

"While you were out" is now the truth: the morning briefing shows what the crew actually did, each item openable, with anything needing your decision on top. A quiet night says so.

You can buy credits: checkout, top-ups, and the billing portal are live once payments are switched on for your workspace.

Sturdier under the hood: a run that starts repeating itself gets corrected then stopped instead of burning your budget, and interrupted work leaves a resumable trail instead of vanishing.

Ask for a document, plan, table, or projection and the crew files it as a real, living file in your room: typed columns, working formulas, ready for your team to open and edit. Task boards, financial projections, and campaign plans all work this way now.

The crew can send files: ask for a projection "emailed to the team as a spreadsheet" and the approval card shows the exact file that will go out; your approve sends a real Excel attachment with live formulas. If the file changes after you approved, the send safely refuses.

Your crew can now act on any tool you connect: refund a charge in Stripe, book a calendar slot, post to Reddit with the right flair, update a deal in your CRM. The crew proposes the exact action as a card showing precisely what will happen; nothing runs until you approve it, and declining means nothing ever runs.

Approve means approve: the Approve button now actually executes the approved action (with the exact arguments you saw on the card), instead of just telling the crew you agree.

Watch the crew work: drafts and action cards now appear on your canvas as they are being made, from real work, live.

No more pretending: a workspace whose crew isn't fully set up now says so honestly instead of performing fake work.

The crew now remembers your conversation: follow-ups like "make it shorter" work, because each chat message carries the thread's recent history instead of starting from scratch.

The crew follows your ratified rules everywhere: workspace charter rules, approved procedures, and your brand voice now steer every chat run, not just onboarding.

Cruma's research about your business is now quote-backed: a fact only appears on your canvas if it can point to the exact words on your website that support it, and a parked or for-sale domain is never mistaken for your site.

The freshest facts win: when your business changes (new prices, new staff), the crew now answers from the newest confirmed information instead of the oldest.

Honest failures: if the crew can't produce a proper result, it says so and stops instead of presenting something half-formed as done.

Your workspace is now durable: everything you and the crew create, files, rooms, invites, sessions, credits, survives restarts and lives in a real database, with real sign-in. Picking up exactly where you left off now includes the app itself.

And it's fast: the app now reuses its database connections and reads only what each screen needs, so pages that took several seconds load in well under one.

Your crew now runs on a real model: chat with the crew and real work comes back, grounded in your workspace's approved knowledge, with every run hard-capped on cost and time, an emergency stop that halts all agent work instantly, and honest messages when a run stops early instead of silent failure.

See what each run cost: every chat run meters against your workspace credits, and the thread shows a quiet line with what it did and what it cost, plus a clear note when a run stopped at its budget, step, or time limit. Ask the crew a question and you get a straight answer, not a list of steps.

The crew can now work in steps: for a task that needs it, the crew reasons, looks things up, and builds toward a fuller answer instead of a single quick reply (opt-in per workspace).

The crew learns from your reactions: mark a crew reply Helpful or Not quite, and Cruma proposes what it learned to your review card. Nothing becomes standing guidance until you approve it, and every completed task is quietly remembered as case history you can look back on.

Your Playbook is real: the skills your crew has learned for your business now show up in the company Playbook, each with what wakes it up, how it works, what needs your ok, and what it's learned. If your company has more than one workspace, a new one starts with your approved skills already in place, then grows its own.

Bring your AI with you: upload your Claude or ChatGPT export during onboarding and Cruma reads it, shows you what it found, and proposes what it learned about your business to your review card. Nothing becomes standing knowledge until you approve it, and importing the same file twice never creates duplicates.

If your workspace is part of a bigger company, the Company room now shows your family of businesses: every workspace under your umbrella with what it does, one click to switch into the ones you belong to. Company admins can rename the umbrella, add a new business (it starts with your brand and its own private space), manage who helps run the company, see each business's plan and credits in one view, and invite someone to several businesses at once. Nothing changes about privacy: each business keeps its own private space, and a company role never opens a business you haven't been invited into.

Run more than one business? When you tell Cruma about your businesses during setup, it can now give each one its own private workspace under one umbrella, with your brand carried into each as a starting point. Nothing is shared between them behind the scenes; you switch between them in one click from the workspace menu, where they appear grouped under your company's name.

Workspaces in the same company can now send each other requests, like a brokerage asking its studio to schedule a listing shoot. The other workspace sees exactly the details you chose to send and nothing else, and decides for itself whether to accept. You approve what goes out; they approve what comes in.

Your sign-in and password-reset emails now arrive with a cleaner, on-brand Cruma design, and a clear reminder of when the link expires.

When you invite a teammate, they now receive an invitation email carrying your workspace's own color and name, so it feels like joining your company, not a generic tool.

Workspace onboarding: invite teammates by email with role selection; first-run welcome that introduces your crew; the workspace interview that assembles your rooms from how you actually work (with your timezone confirmed from your browser).

Approvals you can trust from anywhere: approval requests reach you with a deep link to a trusted surface; approvals expire safely instead of executing stale actions.

Every crew task is bounded by your budget and verified before anything consequential happens — including a kill switch and ask-before-overage, in every plan.

Workspace memory that compounds: your crew proposes what it learned; nothing becomes standing procedure until a human approves it, and you can review or retract anything.

Plans and credits: workspace credits meter agent work transparently; your plan's included credits arrive monthly; payment hiccups pause new work but never touch your data.

Privacy defaults: your workspace's run history is never used as training data unless you explicitly opt in (off by default, every plan).

What Cruma learns in your private chats stays yours: a lesson from your session steers your work, and an admin can promote it to the whole team in one click when it should apply company-wide. Your personal notes to Cruma (how to address you, private details) never reach your teammates.

Mention an app you haven't connected yet ("pull my stripe payouts") and the connect button is always right there in the chat, every time, so the fix is one tap away instead of a trip to Settings.

Booking and acting just got real: the crew now reaches the tools that matter on each app you connect (create an event, draft an email, update a deal) instead of an alphabet-ordered grab bag of admin functions. "Book a meeting Tuesday" now produces the actual booking card to approve.

The crew now checks its own answer before replying: if what it wrote does not actually accomplish what you asked, it takes one more pass to finish the job or tells you plainly what it could not do, instead of replying as if the work happened.

You can now tell Cruma to stay out of things: say "keep out of the encryption files" in chat and it sticks immediately, workspace-wide. Files matching your rule never reach the crew, on any connected tool, and when something is skipped the crew says so plainly ("1 file withheld") instead of pretending it saw everything. Change your mind later? Forget the rule in Settings, Memory and the crew can read those files again.

Your crew can now use tools from external MCP servers you connect, right alongside your app integrations. Point Cruma at a compatible service (like an agent or data tool that speaks MCP) and its read tools show up for the crew to use on your behalf, under the same review-and-approval rules as everything else, so it stays safe by default.

Your crew now catches its own mistakes. When it grades one of its finished runs as weak, even one that looked successful, it proposes a lesson to your review card so the same slip is less likely next time. As always, nothing changes how the crew works until you approve it.

Your crew now knows its way around your tools. Ask about your email, calendar, CRM, or any of hundreds of connectable business apps and Cruma reaches the ones you've connected, no magic words needed. When an app isn't connected yet, Cruma says so and offers to connect it right there in the chat, instead of telling you it can't help.

Connecting an app now confirms itself: when you come back from authorizing, Integrations tells you plainly that the app is connected and the crew can use it. No more wondering if it worked.

You can stop the crew mid-run: a Stop button sits right on "Cruma is working," and clicking it halts the work cleanly with a plain "Stopped." from Cruma, never a spinner you have to refresh away. You can also just keep typing while the crew works: a follow-up like "actually, keep it short" folds into the run already in flight and shapes the answer, and a message that lands just as a run finishes gets its own reply instead of being lost.

Your crew now looks the part. Each specialist wears its own mark, a magnifier for Scout, a drafting frame for Architect, a pen for Closer, a gear for Ops, a coin for Collector, so at a glance you can tell who is on a piece of work, wherever the crew shows up. Cruma keeps its own mark, and your teammates keep their initials.

Cruma listens harder during setup: it reacts to what you just said with a point of view and digs into the interesting thing before moving on, instead of checking boxes. If you are starting from scratch, it talks to you like a co-founder shaping an idea, never as if the business already exists.

Your crew reads warmer and thinks straighter: replies now sound like a colleague who is on your side rather than a corporate bot, and the crew is better at catching the trap in a request (it will tell you to drive the car to the car wash, not walk) instead of taking the easy wrong answer.

Formatting now reads cleanly everywhere your crew writes to you. Bold, emphasis, lists, quotes, code, and links render properly across chat, your morning brief, room workcards, the overnight walkthrough, your rules, and long-text fields, instead of showing raw asterisks and brackets. Links open safely in a new tab and long ones wrap instead of spilling off the edge.

Files you share in chat now stay attached to the conversation after reload, so you can reopen them from the message instead of having them disappear when the thread refreshes.

The documents your crew makes now read cleanly: lists look like lists instead of one jammed line, wide tables fit the page and scroll inside their own box, and your crew stops trying to learn from small talk.

When your crew finishes making something, it now tells you plainly what it made and where it is, and spreadsheets, images, and filled PDFs it produces are now real files you can download.

Opening a direct link to your Playbook, Company, Replay, or Runs now lands you right there instead of quietly sending you home first. Shared and bookmarked links to those pages work on a fresh load.

Editing the words around a live table no longer loses the table. Opening a crew document that held a table, a board, or a design inside it and changing a paragraph used to quietly destroy that content on save. Now it stays exactly where it was, with every row intact.

Onboarding reliably reads your business now: the canvas that used to sometimes come up blank (your logo, colors, and what you do failing to load) no longer starves itself, retries a slow site instead of giving up, and reads your brand colors and wordmark straight off your logo. A quiet site still says so honestly instead of looking broken.

Onboarding now actually reads its own homework: Cruma no longer asks "what does the business do?" while the answer it found on your website is sitting right there on the canvas, and "that's us" is understood for what it is. Setup questions and canvas research also no longer compete, so the conversation can't stall mid-answer.

If you dismissed something Cruma learned, you can now teach it again later: re-proposing the same fact creates a fresh review card instead of silently staying dismissed forever.

Reacting to a crew reply with Helpful or Not quite no longer risks your chat list failing to load afterward. The reaction is recorded quietly; your threads stay exactly as they were.

If your connection blips, the chat now reconnects without ever dropping a message you just sent: what you wrote and what the crew answered both survive the reconnect.

Your message stays above the crew's reply: a fast answer no longer jumps ahead of the message you just sent, so the conversation reads top to bottom the way it happened.

The crew stops repeating itself: when you tell it something lasting (a price, your preferred name, a policy) and later change it, the new answer replaces the old one instead of both sticking around, and the same fact told twice no longer piles up. [Unreleased]: https://github.com/cruma/cruma-product

The crew now asks before it guesses: a booking with no time, a refund with no amount, a send with no recipient gets you a direct question naming exactly what's missing, instead of a proposal built on an invented detail you'd have to catch at the approval card.

Approval cards now speak your language: "List upcoming events from your primary Google Calendar" instead of a raw tool code, with the app's real name and one consistent risk level. What will run is still shown argument by argument, exactly as it will execute.

The crew's live activity now reads like work, not code: "Searching the web", "Reading Gmail", "Checking Google Calendar tools" — plain words instead of technical labels.

Connecting Gmail actually works now. The Google authorization screen used to reject the request before you could approve it; Gmail, Calendar, and Drive connections now ask Google for the right permissions (reading and drafting only, sending still asks you first).

Checking your calendar or inbox no longer asks for approval. Reading is silent and instant, the way it was always meant to be; approvals are saved for things that leave your business, like sending, posting, or charging.

Ask for a document, plan, table, or projection and the crew files it as a real, living file in your room: typed columns, working formulas, ready for your team to open and edit. Task boards, financial projections, and campaign plans all work this way now.

The crew can send files: ask for a projection "emailed to the team as a spreadsheet" and the approval card shows the exact file that will go out; your approve sends a real Excel attachment with live formulas. If the file changes after you approved, the send safely refuses.

Your crew can now act on any tool you connect: refund a charge in Stripe, book a calendar slot, post to Reddit with the right flair, update a deal in your CRM. The crew proposes the exact action as a card showing precisely what will happen; nothing runs until you approve it, and declining means nothing ever runs.

Approve means approve: the Approve button now actually executes the approved action (with the exact arguments you saw on the card), instead of just telling the crew you agree.

Watch the crew work: drafts and action cards now appear on your canvas as they are being made, from real work, live.

No more pretending: a workspace whose crew isn't fully set up now says so honestly instead of performing fake work.

The crew now remembers your conversation: follow-ups like "make it shorter" work, because each chat message carries the thread's recent history instead of starting from scratch.

The crew follows your ratified rules everywhere: workspace charter rules, approved procedures, and your brand voice now steer every chat run, not just onboarding.

Cruma's research about your business is now quote-backed: a fact only appears on your canvas if it can point to the exact words on your website that support it, and a parked or for-sale domain is never mistaken for your site.

The freshest facts win: when your business changes (new prices, new staff), the crew now answers from the newest confirmed information instead of the oldest.

Honest failures: if the crew can't produce a proper result, it says so and stops instead of presenting something half-formed as done.

Your workspace is now durable: everything you and the crew create, files, rooms, invites, sessions, credits, survives restarts and lives in a real database, with real sign-in. Picking up exactly where you left off now includes the app itself.

And it's fast: the app now reuses its database connections and reads only what each screen needs, so pages that took several seconds load in well under one.

Your crew now runs on a real model: chat with the crew and real work comes back, grounded in your workspace's approved knowledge, with every run hard-capped on cost and time, an emergency stop that halts all agent work instantly, and honest messages when a run stops early instead of silent failure.

See what each run cost: every chat run meters against your workspace credits, and the thread shows a quiet line with what it did and what it cost, plus a clear note when a run stopped at its budget, step, or time limit. Ask the crew a question and you get a straight answer, not a list of steps.

The crew can now work in steps: for a task that needs it, the crew reasons, looks things up, and builds toward a fuller answer instead of a single quick reply (opt-in per workspace).

The crew learns from your reactions: mark a crew reply Helpful or Not quite, and Cruma proposes what it learned to your review card. Nothing becomes standing guidance until you approve it, and every completed task is quietly remembered as case history you can look back on.

Your Playbook is real: the skills your crew has learned for your business now show up in the company Playbook, each with what wakes it up, how it works, what needs your ok, and what it's learned. If your company has more than one workspace, a new one starts with your approved skills already in place, then grows its own.

Bring your AI with you: upload your Claude or ChatGPT export during onboarding and Cruma reads it, shows you what it found, and proposes what it learned about your business to your review card. Nothing becomes standing knowledge until you approve it, and importing the same file twice never creates duplicates.

If your workspace is part of a bigger company, the Company room now shows your family of businesses: every workspace under your umbrella with what it does, one click to switch into the ones you belong to. Company admins can rename the umbrella, add a new business (it starts with your brand and its own private space), manage who helps run the company, see each business's plan and credits in one view, and invite someone to several businesses at once. Nothing changes about privacy: each business keeps its own private space, and a company role never opens a business you haven't been invited into.

Run more than one business? When you tell Cruma about your businesses during setup, it can now give each one its own private workspace under one umbrella, with your brand carried into each as a starting point. Nothing is shared between them behind the scenes; you switch between them in one click from the workspace menu, where they appear grouped under your company's name.

Workspaces in the same company can now send each other requests, like a brokerage asking its studio to schedule a listing shoot. The other workspace sees exactly the details you chose to send and nothing else, and decides for itself whether to accept. You approve what goes out; they approve what comes in.

Your sign-in and password-reset emails now arrive with a cleaner, on-brand Cruma design, and a clear reminder of when the link expires.

When you invite a teammate, they now receive an invitation email carrying your workspace's own color and name, so it feels like joining your company, not a generic tool.

Workspace onboarding: invite teammates by email with role selection; first-run welcome that introduces your crew; the workspace interview that assembles your rooms from how you actually work (with your timezone confirmed from your browser).

Approvals you can trust from anywhere: approval requests reach you with a deep link to a trusted surface; approvals expire safely instead of executing stale actions.

Every crew task is bounded by your budget and verified before anything consequential happens — including a kill switch and ask-before-overage, in every plan.

Workspace memory that compounds: your crew proposes what it learned; nothing becomes standing procedure until a human approves it, and you can review or retract anything.

Plans and credits: workspace credits meter agent work transparently; your plan's included credits arrive monthly; payment hiccups pause new work but never touch your data.

Privacy defaults: your workspace's run history is never used as training data unless you explicitly opt in (off by default, every plan).

System emails (sign-in, password reset, confirm, invites, notifications) redrawn on the product register: lighter warm sheet, the traced Cruma lockup, same jade action. Dark mode follows your mail app.

The sign-in page got the new look: the living green aurora, your workspace morning on the stage, and Continue with Microsoft next to Google. Magic link is now the primary path; passwords still work, one tap away.