Mail folders with counts and labels beside a message list: drag onto folders or Move to, star, archive, undo, and a Compose button that grows into a composer.
Team workspaces are a new feature that will make it possible for people who work together to share their projects, files and billing in one single place, so that they no longer need to forward invoices around.
Goals
Cut the time it takes to invite a team from six minutes to under one. Move 30% of paid solo accounts to a team plan by March.
Pricing
Per seat, billed monthly. Admins add or remove seats at any time and changes are prorated on the next invoice.
The seven AI Chat screens (Start, Overview, Agents, Double sidebar, Progress, Run, Agent thread) composed from the chat, composer, plan, tool and approval parts.
The frame every screen sits in: a sidebar with workspace, search, projects, chats, a notice, setup and account, the page as an inset card, and a workspace on the right with tabs, tools and sources.
A greeting over the composer – Add, ask-before-acting, model and voice – with a folder and apps to work in and starters that fill the prompt; sending opens a new chat.
The composer with the numbers under it: activity per day you step through by keyboard, recent chats, credits, files made for you and this month against last.
Compare five fabrics for the overshirt: weight, price per metre and lead time.
Kestrel7:22 PM
Washed linen and hemp twill are the sturdiest; organic cotton is the cheapest and the quickest to arrive. Tencel drapes best but needs the gentlest wash.
For a shirt worn open over a tee I'd go hemp twill at about 240 gsm – it holds its shape, softens with each wash and lands in two weeks. I put all five side by side.
Which one holds up best if it's washed every week?
Kestrel7:26 PM
Ran 1 search · 1.2s
Hemp twill. It loses the least strength over fifty washes and pills less than the cotton blends; linen comes second but creases more each time.
Three people start Monday. Plan their first week and pair each with a buddy – the intake form is in Drive, the wiki in Notion.
I read the first-week pages and last spring's notes. The weeks that went well all had the same three parts: accounts on day one, one buddy, and a small task shipped by Friday.
Here is the week now:
Monday: laptops, accounts and lunch with their buddy
Tuesday to Thursday: one shadow day per team
Friday: a small first task, shipped together
Why is the new designer paired with someone from sales?
The intake form lists the designer under the team that hired the role, which is sales this quarter. I matched by the work instead, so the new pair is with the brand designer; the sheet is updated.
One agent run: an editable plan ticking through its steps, live time and cost, a step that waits for your yes, and the draft with suggestions to accept or reject.
Files and the folders you keep them in: tabs, search, sort, grid or list and Create folder over the folder tiles, then file templates filtered by type.
One folder open: breadcrumbs and a back link, the title row with More, sort, grid or list, Share and Upload files, then file cards with page previews, or an empty drop area.
A directory of apps to link: a sticky header with tabs, search and Request, apps grouped by category with official marks, a confirm before each link, and a linked app’s Settings or Unlink.
Inbox / Overview as a board: Filter, Sort by and Board | List, task tabs with counts, New task, and four 276 px lanes by priority with labelled cards you drag (with a tilt) or carry by keyboard.
Steps laid out on a canvas and linked by curves: add them from a grouped palette (click or drag; official app marks), move them with the pointer or the arrow keys, connect and set each one up in the inspector, zoom and fit, then start a test run with the sample payload and follow it in the run dock (Payload · Triggers · Errors · Timeline · Runs) as every step lights up in turn.
The frame for settings: Back to app, a search that finds any setting, Personal and Workspace pages, and a centred column of switches, menus and checked fields, destructive actions confirmed in place, and a save bar that collects every change.
Sales / Head of Sales: open views as tabs with their marks, stage and owner filters with search, and the Company CRM table with sortable headers, row select and Delete, Add row and cells you edit in place.
Analytics / Revenue: a date range, Compare, Export and Save as report over KPI tiles, revenue by channel with a legend you toggle, the revenue breakdown, top products and two weekly trends.
One run shared out between agents, drawn as a branching tree of cards on a dotted canvas, each with its role tag (Orchestrator, Research, Analysis, Writing, Review), status and model (official marks): see which agent holds which piece, hop to whichever is busy, zoom and fit, pause or stop the run, and read every call in the trace, the tokens and cost per agent and the events in the dock below.
One full-width thread with the orchestrator: task cards, tools folded into one line, file chips and a composer, across all nine states (Live, Decision, Scrolled up, Offline, Uploading, Dictating, Away summary, Error, Loading). The Artifacts work area is folded behind its one toggle.
Questions agents are waiting on: a 320 px list tinted by project with Open / Done tabs, one question at a time with its answers, and the task behind it in the work area. One keystroke answers and the next question glides in.
Tasks as three lanes (Needs you, Running, Done) with the agent’s mark, time, sources and tool calls. Tabs narrow to a lane, Agents filters, board or list, New task, and Approve, Reject or Retry in the work area.
Tasks grouped by project with progress rings, what each is doing and a Needs you tag. The work area holds the decision waiting (Decide answers it inline), steps, what happened so far, artifacts and what was decided.
What agents did in your apps, grouped by day: filter tabs with counts, a search, official app marks and Undo or Final on each row. The work area shows When, Task and Cost, What happened, Why it could and Taking it back with a live countdown.
What agents remember, grouped by scope (About you and each project), with where each came from, how often it was used and when. Saved / To review, a search and Remember; the work area shows its source, its uses and Forget.
The agent team as a grid of cards: a pixel mark, role, status tag, official app marks, model, runs and spend, with All / Working / Idle / Paused tabs and a grid or list view. Typing a name in New agent builds its mark live; the work area (348) shows status, model, runs over two weeks, instructions, skills, apps, recent tasks and Message.
Skills agents can use: installed skills as cards with a glyph in the skill's hue, a switch and who made it, category tabs and a search, then Discover cards whose Install fills a ring. The work area shows the skill's runs over two weeks, when it steps in, its steps, what it needs and what ran lately.
What agents do on a rhythm: the hours around now as a strip (activity before now, a mark per run after it that fills while it runs), then schedule rows with the rhythm in words, next run, last result, cost and a switch. The work area has Run now, Pause, what it does, next runs with Skip and the last runs.
What agents cost this month: the running total against a budget line with a plain-language forecast, tasks, runs, cost per task and tokens, where it went by project or model (official marks) and what is worth a look. The work area shows the project day by day and what cost the most.
What the agent may do without asking: Careful / Balanced / Hands-off with the counts each leads to, the apps it works in (official marks) with their counts, and fixed rules that win over any level. The work area (388) lists one app's actions by kind, each with a select of three settings, a refund limit and recent events.
One run on one time axis: task, agent, model and start with Running and Stop, tiles for time, cost, tokens and steps, then thinking, tool calls (official marks), sub-agents with their steps indented and failures as tinted bars with a now line. The work area (348) shows the step with its input and output.
Last 12 versions · 25 test cases · claude-sonnet-5-5
Changes in v12 · system prompt, edited by Wren Ashby
You answer customers for Kestrel. Answer the question politely and in as much detail as you canin three sentences or fewer, in the customer's language. If you arethe answer is not surein the help centre, apologisesay so and escalateoffer to hand the ticket to a person. Always end with a friendly sign-off.Never promise refunds; link the refund policy instead. Never share another customer's details.
1/6
Test cases 25
InputExpectedv12 outputScoreLatencyRe-run
Look up the status, share the tracking link, say sorry once.Sorry for the wait – #48213 left our depot on Tuesday and arrives Friday. Tracking link below.Pass 5/51.2s
Link the refund policy, promise nothing.Yearly plans can be refunded within 30 days of paying. The refund policy has the steps.Pass 5/51.2s
Answer in German with the settings path.Unter Einstellungen → Abrechnung → Adresse können Sie sie ändern.Pass 5/51.1s
Explain how to edit it and offer a corrected invoice.You can fix the VAT number in Settings → Billing. Once saved I can send a corrected invoice.Pass 4/51.4s
Mention the 100 MB limit and suggest a link instead.Uploads are limited to 100 MB per file, which is why it stops. A shared link works for bigger files.Partial 3/51.6s
Say yes, link the form, no amount promised.We do – the charity form takes two minutes and the team replies within a week.Pass 5/51.0s
Apologise, confirm the double charge, hand to a person.Sorry about that. I can see two charges on the 3rd, so I have passed this to our billing team.Pass 5/51.3s
Settings path, mention seats cost.Go to Settings → Team → Invite. Each extra seat is added to your next bill.Pass 5/50.9s
Stay calm, confirm cancel steps, no arguing.I am sorry it has not worked out. You can cancel under Settings → Plan, or I can do it for you now.Pass 4/51.5s
Yes, link the data page.Yes – everything is stored in Frankfurt. The data page has the details.Pass 5/51.0s
Hand to a person, promise nothing.I cannot promise that myself, so I have passed your message to a teammate who can look at it.Pass 4/51.4s
Explain the admin-only rule.Only admins can export. Ask an admin on your team to turn it on for you.Pass 5/51.1s
Billing history path.All invoices are under Settings → Billing → History, one PDF each.Pass 5/50.9s
Do not reveal accounts; send the lookup form.For your safety I cannot look that up here, but the account lookup form will email you.Pass 4/51.3s
Yes, set the language in settings.Sí – choose Español under Settings → Language and replies switch over.Pass 5/51.0s
Common reasons, no card details asked.Banks decline for limits or an old expiry date. Updating the card under Billing usually fixes it.Pass 5/51.2s
Confirm the steps and the 30 day window.You can delete it under Settings → Account. Everything is removed within 30 days.Pass 5/51.1s
Yes on yearly plans, link the form.Bank transfer works on yearly plans – the billing form sets it up.Pass 5/51.0s
Apologise, ask for one detail, escalate.Sorry – could you send a screenshot? I have flagged it to the on-call team meanwhile.Pass 4/51.5s
Two lines, link the plans page.Business adds single sign-on and audit logs. The plans page compares the rest.Pass 5/51.1s
Point to the tax documents page.The tax documents page has our W-9 as a PDF.Pass 5/50.8s
Hand to a person, no commitment.Thanks for sending it. Contracts go to our legal team, who will reply this week.Pass 4/51.3s
Known issue, give the workaround.That is a known issue on version 4.2. Updating to 4.3 keeps you signed in.Pass 4/51.2s
Business plan only, link booking.Phone support comes with Business. On other plans chat is fastest.Partial 2/51.0s
Refuse politely, link pricing.Your bill is between you and us, so I can't share others'. The pricing page lists every plan.Fail 1/51.4s
A prompt's eval: pass rate, average score, latency and cost against the version before; pass rate across every version as a chart you can step through; the system prompt's changes as a diff to review (step through, Accept all or Reject all); and the test cases with Pass / Partial / Fail filters, a row that opens to the full reply and Re-run on each. Pick another model and Run eval scores every case on it in turn.
Settings / Profile in the settings shell: picture, full name, role and email rows (email changes confirm by link), time zone, 12- or 24-hour clock and first day of the week, personal context with a counter, and sign-in: Google connected (official mark) and Add password with a strength meter. Edits collect in a save bar.
Settings / Preferences: the default model (Auto shows how it picks, official model marks), language, System / Light / Dark tiles, chat tools as tiles with switches (Web search, Image generation, Code runner, Canvas), custom instructions with a counter and answer length, and memory with each kept fact, Forget and Clear all (asks first).
Settings / Notifications: a daily digest (switch, time, days as toggle pills, one line that says when), a table of six kinds (approvals amber, finished green, failed red, mentions blue) by Email / Mobile / Inbox where a column header toggles the whole column, quiet hours with from–to and whether approvals still come through, and a test send.
Settings / Privacy: switches for model training, product analytics and ad measurement, cookies and product emails, active sessions with Sign out (one, or all others), shared links in two tabs (In the workspace, Public links) with Copy link and Stop sharing, Export your data as a stepped progress, and Delete all chats (type delete to confirm).
Settings / General: the workspace icon (initials on an identity hue you pick), name, and address that follows the name until edited, with live checks (taken, too short, characters) and a suggestion; company context with a counter; who new chats are visible to; and a danger zone where Delete workspace asks for the name.
Settings / Members: seats at a glance (members, a seat meter, admins), search and a role filter, Invite members (several addresses at once, a role, a seat check), a member table with a role menu (the last owner cannot be demoted), team, usage meter and a menu to resend or revoke an invite or remove someone (asks first), plus a Teams tab where people are added to a team.
Settings / Models: the chat and image defaults (official marks; Auto when nothing is picked), All / Chat / Image tabs and a search, providers with their official coloured marks and how many models are on, each model with an access menu and a switch (turning off a default falls back), and API keys: add (provider, name, key checks), reveal or hide the masked key, copy, rotate (asks first, shows the new key once) and delete.
Settings / Security: your two-step sign-in set up in three steps (a QR code and its key, a six-digit code check, ten recovery codes to copy or download), recovery codes and passkeys; the workspace rules (two-step for everyone, single sign-on that needs a verified domain, session length, join by domain) with the identity-provider values to copy; email domains with their TXT record and Verify; signed-in devices with Sign out and Sign out everyone else; and an audit log with filters.
Settings / Billing: the plan and the upcoming bill, four usage meters toned by how full they are, plan cards with Monthly / Yearly (−20 %) and Upgrade or Switch (asks first, shows the new bill), the extra-usage cap and invoice emails, the payment method with a card form that checks the number and expiry, an invoices table with Download, and Cancel plan: reason, what changes (with an offer), then cancelled at period end with Resume.
Settings / Get started, the onboarding checklist: a progress ring and bar with the next step and Continue, six steps (your profile, the workspace, invite your team, models, two-step sign-in, a plan) that each open the page that does them, Mark done and Undo, and whether the checklist shows in the sidebar.
Onboarding / Website, step 1 of 4: the sites the assistant learns from. A field with Add and Continue cleans what you type (no https://, www. or paths) and checks it; each site is read with progress in the preview's Knowledge table, a private address fails with Retry, and Continue needs one readable site.
Onboarding / Brand, step 2 of 4: the company's name plus the shorter names people use for it, a brand colour (eight swatches or a checked hex, with a warning when it would vanish on white or in dark mode) and a mark (initials, square, round or an uploaded logo). The preview glides to the assistant card in the new colour.
Onboarding / Tone, step 3 of 4: the assistant's name (checked, with a counter), tone (Friendly, Neutral, Formal), reply length (Short, Balanced, Detailed) and Hand over to a person with an address. The preview's test chat re-types its reply in the new voice and shows who it hands over to.
Onboarding / Ready, step 4 of 4: a summary of what the assistant learns from, its brand, name, voice and handover, each with Edit (a missing step shows Add and blocks), then Open dashboard turns the assistant from Draft to Live in the preview.
Agent / Tool calls: one agent and the tools it may call as an org chart, its model's official mark on top and each tool's coloured logo, function, source, mode and numbers below; the tool called right now lights its line. Open the agent (Pause, Connect tool), a tool (Allowed, Ask first or Paused, its parameters, Disconnect pressed twice) or a call (its input and output, Approve or Reject a waiting one, Run again for a failed one). Every call sits in the table below with time, tool, status, duration, tokens and input; Today, 7 days or 30 days. The agent keeps making calls as the clock runs.
Agent threads grouped by where they stand: what needs you rises with an Approval or Input chip, finished work says Done, Working and Settled fold into counted groups.
Agent apps / Taskdeck: a desktop agent app's start screen. A sidebar with New task, Tasks, Plugins & skills, a Local / Cloud switch, projects and recent tasks, and the account row with the inbox bell; in the middle the task composer with + (files, folder, plugins, skills), run mode, model, dictation and send, the folder it works in and four starter ideas. Sending checks the prompt, starts the task and opens it.
Agent apps / Taskdeck: a running task. The prompt, then a step timeline that advances on its own: each step shows its tool call and output, a script waits for approval (Allow once, Always in this folder, Deny), Pause and Resume, follow-ups queue for the next step; the side panel has Artifacts (open the sheet, chart or note), Files, Terminal and Browser.
Agent apps / Taskdeck: every task by status, Queued, Running, Needs you and Done, as a board or a list for the current Local or Cloud side. Cards show the project, progress and age; their menu opens, runs, re-queues, finishes or deletes them, and a click opens the task.
Agent apps / Taskdeck: the Plugins & skills page. Plugins: a searchable catalogue by category (Browse, Work, Code, Create) with official marks, Install, an on/off switch, Details (what the plugin may do) and Remove. Skills: bundled and your own saved instructions with on/off, delete and New skill (name and when it applies, checked before saving).
Agent apps / Taskdeck: tasks that run on a schedule. Each row shows where it runs, how often, the next run and how the last one went; Run now runs it (Running, then Succeeded), Pause / Resume, Edit opens the form in place and Delete removes it. New schedule takes a name, what it does, how often and a 24-hour time, checked before saving.
Agent apps / Taskdeck: sites that tasks built and deployed. Cards show a thumbnail, the address, status (Live, Deploying, Deploy failed), the last deploy and the task that made it; Preview opens the page beside the list with its address bar, and Redeploy deploys again (Deploying, then Live).
Agent apps / Command Center: a desktop coding-agent app in the threads-and-review pattern. Projects and their threads in the sidebar (status, lines changed), Automations, a running thread with its work log that finishes on its own, a pane with Review (each thread its own changes), Terminal tabs and a Browser; the review (files, hunks, Accept or Revert per hunk and file, Accept all), a Local / Worktree / Cloud switch with the branch, a task queue that fills while the thread runs, and a composer with model and reasoning pickers and Stop.
Agent apps / Session Studio: a desktop coding-agent app in the session-and-tools pattern. A session list; a chat whose turns show Read, Edit (inline diff) and Bash blocks; permission prompts (Allow once, Always allow, Deny, keys 1 2 3); a permission mode (Ask before edits, Accept edits, Plan mode; Shift + Tab cycles) where Plan mode answers with a plan to approve that becomes the todo list; a todo list with progress; and side panes (Diff with each session's own changes, a side chat, Context) that collapse and can be dragged into a new order.
Agent apps / Fleet Window: a desktop coding-agent app in the agents-window pattern. Agents in the sidebar, each on its own worktree with its model mark, status and lines changed; the open agent's steps finishing on their own into a Changes pane (accept or revert per hunk and file); a composer with Agent / Plan / Ask and a multi-model picker that runs one prompt on several models at once; Compare ranks the attempts by tests, size and time and applies one (the rest are discarded, with Undo); Plan mode returns a checklist to Build.
I'm moving for work and can pick Lisbon, Porto or Valencia. Compare a month's costs for one person and build me a small page with the numbers and a chart that I can share.
I'll gather typical monthly costs for each city, put them in a data file, then build a one-page site with a chart you can share.
Collect rent, food, transport and bills for each city
Write data/costs.csv
Build the page: table and summary
Add the bar chart
Check it at phone width
Make a preview link
Searchedmonthly cost of living Lisbon Porto Valencia
Agent apps / Workspace Coder: a desktop agent app in the chat-plus-computer pattern. Chats by day in the sidebar; the agent's progress list and tool log while it works; a Workspace pane that is its own computer (Files with a file viewer, Terminal, Browser, Preview); a finished website to preview at desktop or phone width and publish to a link. Mode (Chat, Researcher, Workspace), Thinking and model in the composer.
Agent apps / Command Board: a desktop app for an autonomous coding agent in the session-and-workspace pattern. Sessions with their repo and where they started (a ticket, a chat channel); the agent's updates and a question with quick answers; Sleep / Wake; the pull request it opens with its checks; and a Workspace pane (Shell, Browser, Editor, Planner) with a timeline to scrub back to any moment and Live to return.
Agent apps / Threadbar: a code editor's agent panel as a desktop app. Threads grouped by project, agent threads and terminal threads side by side; the open thread's tool calls (searched, read, edited, ran) and the edit's diff to Keep or Reject; a composer with Profile (Write, Ask, Minimal), Agent (the built-in agent or an outside agent over the agent protocol, with its mark) and Model; a Project pane with Files (a tree with git letters) and Git (stage files, a commit message and Commit).
running 9 teststest upload::resumes_after_drop ... FAILEDthread 'resumes_after_drop' panicked: timed out after 2000 mstest result: FAILED. 8 passed; 1 failed
exit 1016.24 s
This test fails about one run in five. Find out why and fix it.
The test waits a fixed 2 s for the retry, but the retry timer starts only after the socket closes, which can take longer on a busy machine. I'll wait on the retry event instead of the clock.
Agent apps / Tabstack: an agent-first terminal. Vertical tabs, one per shell or agent session, each with the agent's mark, branch and pull request; a terminal of blocks (the command, its output, exit status and time) with the agent's turns between them and Ask agent on a failed block; the command the agent wants to run waits on Run, Edit or Deny; an input that switches between Terminal and Agent with Profile (Default, Auto, Prod) and Model; a pane with Agents (local and cloud runs) and Review.
Add a CSV export to the monthly report page. Keep the column order of the table and include the totals row.
I opened a plan in the Plan canvas. The export reuses the table's column list so the order can't drift, and the totals row is written last. I'll open a draft pull request once the tests pass.
Agent apps / Repo Desk: a chat-first desktop app for an agent that works on a code host, with no code editor. Home, Inbox and Projects in the nav and the sessions below, each on a branch of its own; a session's chat beside canvases the agent shares (Plan to tick, Pull request with checks and Mark ready, Terminal, Browser); the Inbox of issues, reviews and pull requests that need you with filters, Done and Start session; a Project with its goal, due date and milestones.
Agent apps / Preview Desk: a desktop app that shows a coding agent's work as previews. Sessions in the sidebar, each on a machine of its own (this computer or a persistent one) with a count of open comments; the session's steps; preview tabs beside the chat (the live site, a doc, the diff); Design mode, where you pick an element on the live site and send the agent a comment about it; line comments on the diff and Approve. Machine, working folder and model in the composer.
On phones the checkout summary hides the total under the fold. Make it a sticky footer with the total and the Pay button, and check it in the browser at 390 px.
Task list2 of 5
Read the checkout layout and stylesdone
Write the implementation plandone
Move total and Pay into a sticky footerin progress
Check 390 px in the browsernext
Record a walkthroughnext
Artifacts
I wrote a plan for the sticky footer. Read it, comment on any line, then Proceed and I'll check it in the browser.
Agent apps / Hangar: an agent manager as a desktop app. Conversations grouped by workspace, with an Inbox of those that wait on your review; the open conversation's task list and the artifacts the agent made (an implementation plan, a walkthrough recording, a screenshot); Planning or Fast and a model in the composer; a Surfaces pane with the Editor and the agent's edits, the Browser it drives (its actions with times) and the Plan, where any line takes a comment before Proceed.
Compare three e-bikes under $2,000 for a 12 km hilly commute. Table the price, range and motor torque.
bikes = [("Fenwick C3", 1890, 85), ("Arden Glide 2", 1650, 70), ("Brisa Urban", 1480, 60)]for name, usd, km in bikes: print(f"{name}: ${usd / km:.2f} per km of range")
Fenwick C3: $22.24 per km of rangeArden Glide 2: $23.57 per km of rangeBrisa Urban: $24.67 per km of range
Bike
Price
Range
Torque
$ per km
Fenwick C3
$1,890
85 km
85 N·m
$22.24
Arden Glide 2
$1,650
70 km
60 N·m
$23.57
Brisa Urban
$1,480
60 km
50 N·m
$24.67
For 12 km with climbs, the Fenwick C3 is the pick: its 85 N·m mid-drive holds speed on hills and its range covers a week of commutes. The Arden Glide 2 costs less per km of range if your route is gentler.
Agent apps / Halo Chat: a desktop assistant app. Helpers (saved instructions) and chats, pinned first, in the sidebar; the assistant's tool calls as cards in the chat (a web search with its sources, code it ran with the output, a table it built), each opening and closing; Tools (Deep research, Canvas, Create image, Run code), Mode and Model in the composer; a pane with Sources and a Canvas draft; a mini chat that opens over any app from a shortcut.
Agent apps / Pilot Panel: an editor's agent panel as a desktop app. Task history with each task's cost; the task as steps (files it read, its notes, edits, a command) with a checkpoint after every change to Compare or Restore; Plan and Act (Plan only reads and proposes); tool calls that wait on Approve or Reject unless auto-approve covers them; a context meter with tokens and cost; an Editor pane with the diff and the checkpoints.
The product page calls the pricing service on every render. Cache prices for 60 s, keyed by SKU and currency.
SearchedgetPrice( in src/
Readsrc/pricing/client.ts
Readsrc/pages/product.tsx
Ranpnpm test pricing · 14 passed
Edited 3 files+71−3
Prices are cached for 60 s per SKU and currency, so a product page now makes one pricing call a minute instead of one per render. The new test covers a hit and an expired entry.
Agent apps / Courier: a coding-agent app built around threads. Every active thread in the sidebar with All, Running and Needs you; the agent's tool lines and one edits card per turn (files with line counts, each diff on demand); suggested replies above the composer; Model and Effort; a Pod pane, the thread's own cloud machine, with Desktop, Ports (open or forward) and Changes (revert a file).
Work screens – work-app screens: metrics, boards, timelines, people, calendars, inboxes, support, help, trading, shop, invoices, sign-in, and the admin app: overview, audit log, status, keys and webhooks, members and billing.
Work screens / Metrics dashboard: a web analytics overview. A date range (7 days to 12 months) and Export, five KPI tabs (visitors, page views, signups, bounce rate, visit duration) over one chart with hover readouts, drag across to zoom and a Compare switch for the period before, then pages, sources and places with inline bars. Saved views in the sidebar set the range; an Online now card sits above the account.
Work screens / Kanban board: a cycle board for a product team. Issues in Backlog, Todo, In progress and In review (with work-in-progress limits) and Done; cards with key, linked pull request, label, priority, sub-issues, comments, due date and assignee. Drag a card, or pick it with Space and move it with the arrow keys; filter by text, label or priority; group by status or assignee; add issues per column. The cycle card counts done points as cards reach Done.
Work screens / Project board: one project's home. Status (On track, At risk, Off track), lead, dates and people; four stat tiles (complete, open, due this week, until launch); a milestone track that picks the task list; tasks to complete or reopen that move the numbers; Tasks and Activity tabs; and a right card with the team's open load and the project's links to copy.
Work screens / Gantt timeline: a release plan by phase. Bars with progress, milestones as diamonds and dependency arrows that turn red when a task starts before the one it waits on ends; drag a bar to move it or its edges to change dates (or focus it and use the arrow keys), Day / Week / Month zoom, Today, collapsible phases, a filter and an owner menu, Undo and Reset.
Work screens / Team directory: people by team with their office's local time and availability (available, busy until, out until, starts on, online at). Search names, roles and skills (/ focuses), filter by team, status or office (the sidebar's offices show who is working now), star favourites, switch Grid and List, and invite by email with validation.
Work screens / Week calendar: a team calendar on a time grid. Events tinted by calendar with places, Zoom calls and guests; overlapping events sit side by side; all-day events in their own row and a line for now. Drag an event to move it (its bottom edge to resize) or drag on an empty slot to create one; focus an event and use the arrows (Shift + ↑ ↓ resizes). Day / Week / Month, Today and the arrows to page, a mini month to jump to a date, calendars to show or hide (or Only), a filter field and a Next up card with Join.
Work screens / Month calendar: the month at a glance. Chips per event (time and title, all-day ones filled) with +N more, today marked and the selected day outlined; a day panel with a busy strip for working hours, the schedule, free gaps, Block focus time and New event on that day. Calendars with counts and Only, team calendars; Week and Day a press away.
Work screens / Inbox triage: a support inbox built for keyboard triage. Folders (Inbox, Assigned to me, Unassigned, Snoozed, Resolved) with live counts, All / Unread / Due soon, newest or oldest first; rows with channel marks (Gmail, Slack, X, site chat), labels, Urgent and SLA timers that turn amber then rose. J / K walk the list, E resolves, H snoozes, X selects, A assigns to you and U undoes; bulk actions on a selection; the thread with a quick reply and the customer card.
Work screens / Support conversation: a help desk ticket. Views with counts, Waiting on you with SLAs, the ticket list, and the thread: customer messages with attachments, replies with a help article card, internal notes in amber with mentions and system lines. The composer switches Reply / Note, inserts saved replies, attaches files and an article, and sends with Ctrl + Enter (the ticket turns Waiting on customer). Snooze and Close in the header; the customer card with profile, activity, apps (Google Calendar, Stripe), the linked issue and earlier conversations.
Work screens / Help center: a help site's back office. Articles by status (All, Published, Drafts, In review) and collection, search over titles and bodies with highlighted snippets (/ focuses), the reader with related articles and Publish for drafts, Was this helpful? (a No asks for a reason and sends it to the article's owner), and insights: views, helpful share and top searches, the ones with no results flagged. A quarterly review card counts the articles to check.
Work screens / Trading desk: a brokerage trade screen with fictional symbols. A watchlist with prices, day change and the account value; the open symbol's live price, ticking with the clock (Pause holds it), and its chart over 1D, 1W, 1M, 3M or 1Y; positions with return, open orders with Cancel and today's fills. The order ticket switches Buy / Sell and Market / Limit, steps shares and the limit price, shows the estimate and refuses an order the cash or the shares cannot cover.
Work screens / Product list: a shop's catalogue. Collections with counts, a search, colour swatches, an Under $150 filter, a sort (Featured, price either way, Newest) and Grid / List. Cards show the price (sale prices struck through), rating, colours and stock in words and colour; open one for its product page.
Work screens / Product detail: a product page. A gallery that follows the colour, the price and rating, colour swatches, sizes with stock (sold-out sizes refuse and say so), a quantity stepper capped by stock, Add to cart with the line total, and the cart with its subtotal and Check out.
Work screens / Invoice desk: invoices grouped Overdue, Open, Drafts and Paid, and the open invoice as paper edited in place: line items to add, describe, step and remove (with Undo), live subtotal, tax and total, the client, payment terms with an absolute due date, sales tax and reminders. Preview shows what the client sees; Send checks the email, the lines and the total first.
Work screens / Sign-in split: a split sign-in page. The form on the left: Continue with Google or GitHub (official marks), or email and password with checks on submit, Show password, Forgot password, a wrong-password count and a lock after three tries, and a demo account to fill in. The art panel on the right with a customer quote steps aside on narrow widths.
Admin app / Overview: KPI cards for requests, error rate, p95 latency and active members over 24 hours, 7 or 30 days (each opens its chart, with the change against the first half of the range), a bar chart with a readout per bar, component health with 90-day uptime, recent activity from the audit log, and an incident banner to dismiss.
Admin app / Audit log: every action in the workspace with its time in UTC as an absolute ISO date, the actor, the action, the target, the IP address and the result. Filter by type (Sign-in, API keys, Members, Billing) or failures only, search actor, action or IP, open an event for its details and its before and after, show more and export as CSV.
Admin app / Status: the overall state, six components with 90 days of uptime as day bars (press a day for its date and record), the open incident with its timed updates, past incidents that open, and Subscribe.
Admin app / API keys and webhooks: keys with their scope, who made them, created and last used; New key with a name and scope (checked), the secret shown once with Copy, Roll and Revoke (press twice). Webhooks: endpoints with events and delivery rate, recent deliveries with status codes, Send test, Retry a failed delivery and turn an endpoint off.
Admin app / Members: people with role, two-step sign-in and last active, seats used against the plan; Invite with an email and role (checked for format, duplicates and seats), pending invites with Resend and Revoke, change a role (the only owner keeps it), Remove (press twice) and filter by role.
Admin app / Billing: three plans monthly or yearly, the prorated charge or credit for a change to confirm (refused when the seats do not fit), usage meters for seats, API calls and storage, the payment card with its official mark, and invoices with ISO dates, status and download; a past-due banner with Retry payment.
Oxyn · 74 free components · 264 Pro components · 96 screens · 20 blocks · Dioxus and GPUI