Documentation · App Builder
App Studio
Where it is and who can use it
App Studio is at /dashboard/app-studio in the Cloud Dashboard. The list of apps is at
/dashboard/app-studio/apps.
| Organization role | What it can do |
|---|---|
| viewer | Nothing in App Studio: it shows a read-only notice instead of the editor and the apps list. |
| developer | Design, save drafts, publish updates. |
| admin | Everything a developer does, plus the first publish of an app (it creates the app's API key), enabling payments, connecting GitHub and exporting, deleting an app. |
The top of the page
The top band holds the controls for the app you are working on.
- App Studio badge, then a selector of your organization's apps. It reads Open an app… (or No apps yet);
each entry shows the app's name followed by
· v<N>when it is published or· draftwhen it is not. - + New app closes the current app and starts again from the empty page.
- All apps opens the apps list.
- On the right, the status:
status: draftorstatus: published v<N>, followed by· unsavedwhen you have changes not saved yet and· <N> problem(s)when the manifest does not validate. On the empty page it readsnew app.
The co-pilot box
On a new app the heading is Describe your app. The co-pilot builds it. and the input placeholder is Describe your app — e.g. “a valuation report for real-estate agents, paid with credits”. The button is Design the app.
On an open app the heading is Describe a change. The co-pilot redesigns your app., the placeholder is Describe a change — e.g. “add a Paphos option and ask for the budget”, and the button is Propose change.
Press the button or Enter to start. The co-pilot always receives your text and the current manifest, so a change request only changes what you ask for. See the co-pilot run.
Examples
On a new app with an empty box, Start from an example shows three cards: Property valuation report, Supplier due-diligence and Trip planner. Use this → copies the card's text into the box. Edit it, then press Design the app.
Action buttons
Once an app is open, a row of buttons appears:
| Button | What it does |
|---|---|
| Save draft | Saves the manifest and the canvas layout. Shows Saving… while it works. |
| Open preview | Opens https://preview-<slug>.webrobot.eu in a new tab: the saved draft, served by the runtime. Shown once the app has been saved. |
| Export code | Opens or closes the export panel. |
| Export diagram (PDF) | Opens a print window with the canvas as a vector drawing, the legend of the arrows and the list of the boxes. Save it as PDF from the browser. |
| Publish | Saves unsaved changes, then publishes. Disabled while there are problems (tooltip Fix the problems first). Shows Publishing…. See Publishing. |
Blank app
When no app is open, a card Or start from a blank app offers Start blank. The browser asks for the App name. The slug is made from the name: lower case, accents removed, other characters turned into hyphens, at most 40 characters. If the result is not a valid slug the message is Choose a name with at least 3 letters.
The blank app is a working skeleton:
- sign-in by email link, with two roles:
admin(from the app's admin emails) anduser(anyone who signs up); - a home page visible without signing in;
- a form with one required text area, What should the report be about?;
- one agent,
main, with the modelclaude-sonnet-5-5, 6 turns and a budget of 0.5 USD per run; - the pages New report, My reports, the report itself (printable) and Credits;
- billing in credits: 50 free credits at sign-up, pay per use with 10 credits held per run, a markup of 50%, 1 credit = 0.01 EUR, and one pack of 200 credits for 5 EUR.
spec:
pattern: report
billing:
modes: [credits]
signupCredits: 50
costs:
"agent:main": { estimate: 10, metered: true }
metering: { creditEur: 0.01, markupPct: 50, drivers: [llm_tokens, compute] }
packs:
- { key: small, credits: 200, price: 5 }
The blank app is not saved until you press Save draft.
The co-pilot run
- The co-pilot is an agent of the platform. It runs in the background. The button shows Designing… and the page says The co-pilot is designing the app (usually 1–3 minutes). You can leave the page: the proposal waits for you. Larger apps take longer; a full app with several agents and approvals took about 2 minutes in our tests.
- App Studio checks the run every 5 seconds. If you reload the page or come back to the same app within 30 minutes, it resumes following the same run.
- When the proposal arrives, App Studio validates it. If it has problems, or only design warnings, App Studio sends it back to the co-pilot once, with the list (at most 30 items), and shows The proposal had N problem(s): the co-pilot is fixing them…. The second answer is shown as it is, even if problems remain.
- If the co-pilot needs something it cannot guess, it answers with a question instead of an app. The page shows The co-pilot asks: and the question. Add the answer to your text and run it again.
- If the run fails, the page shows its error, or the co-pilot failed. If it ends without an app or a question, the page shows the co-pilot gave no app — describe it again.
The co-pilot never changes your app by itself: you always see a proposal first.
The proposal
While a proposal is shown, the canvas draws the proposal, not your app, and a bar above it reads The canvas shows the proposal: <app name>.
- Green border and a new badge: a box that does not exist in your app.
- Amber border and a changed badge: a box whose content differs. A changed goal on the arrow into an agent marks that agent as changed.
- would remove: followed by names: boxes of your app that the proposal no longer has.
- If the proposal still has schema problems the bar says N schema problem(s): you can apply it and fix them.
Then:
- Apply replaces your manifest with the proposal. On a new app, the app takes the slug the co-pilot chose. On an existing app, the slug stays the same: the address never changes under you. The message reads Proposal applied — click the boxes to review, then Save draft.
- Discard drops the proposal and shows your app again.
While a proposal is on screen you cannot select boxes, the add buttons are disabled, and the inspector is hidden. Applying does not save: press Save draft afterwards.
The canvas
The canvas draws the manifest as a flow from left to right. It is a projection of the manifest: every edit goes back into the manifest through the panels. You can drag boxes; their positions are saved with the draft, outside the manifest. Zoom and pan with the controls in the corner.
Lanes
Boxes sit in lanes that say what they are for, left to right: Users & pricing, Navigation, Pages, Input & data, Agents, App data, What users see. Many views wrap into more columns of five.
Boxes
Hover a box to see its type and one line on what it does. Click it to edit it in the inspector.
| Icon | Box | What it is |
|---|---|---|
| 👤 | Users | Who uses the app: how they sign in and their roles. |
| ☰ | Menu | The header for visitors and the app menu for signed-in users, in order. |
| 🏠 | Home page / Page | A page built from blocks (/ for the home, /v/<key> for the others). Double-click to open the page editor. |
| 📝 | Form | The fields the user fills in; they become the agent's goal. |
| 🤖 | Agent | An agent of the app. Its lines show the model (or the team size), whether it acts on the platform, and the start of its instructions. |
| 📄 | View | A page of the app that is not built from blocks: report, history, table, cards, chart, chat, approvals… |
| 💳 | Pricing | How users pay: modes, free credits, packs, plans. Shown when billing is not free. |
| 🗃 | App collection | Records kept by the app. Shown when a view reads it or an agent uses it as a tool. |
| 🐍 | Python slot | Your own Python, run in a sandbox. |
| 🔌 | External service (MCP) | An MCP server of another service whose tools an agent can use. |
| 🗄 | Data job | A pipeline whose dataset the app reads. |
| ⏱ | Schedule | Runs an agent or a data job at fixed times. Its title is the time in words, for example Every day at 06:00 UTC. |
| 👥 | Team | Agents working together, as a sequence or a swarm. |
| ⚡ | Trigger | An event that starts an agent, a team or a data job. |
| ✋ | Approval | A human checkpoint and the role that decides. |
| 🧩 | Agent template | An agent the app's users create for themselves. |
| 🧠 | Memory | What an agent remembers about each user. |
| 📞 | Channel | Where conversations reach an agent. |
A box with a dashed border and not run yet in its hover card is kept in the manifest, but App Studio does not
expect the runtime to run it. This is the case, for example, for data jobs, channels other than web, and supporting agents that the form, a
schedule or a when condition does not start.
Arrows
Every arrow has one meaning and its own colour and line style. Hover an arrow to read it; click it to select it.
| Arrow | Meaning |
|---|---|
| Menu item | An entry of the app menu that opens this page (signed-in users). |
| Link | A button or a header link that takes the user to this page or section. |
| Contains | The page shows this view inside it (the form, the latest report, the chat…). |
| Starts the agent | Submitting the form runs the agent with the goal built from the fields. Its label shows the first line of the goal and the price. |
| Writes | The agent's result lands in this view or collection. |
| Each message | Every chat message is a run of this agent, with the conversation so far. |
| Data | Rows or results flowing in: a data job, an app collection, a schedule. |
| Uses as a tool | The agent can call this while it works: app tools on a collection, a Python slot, an external MCP server. |
| Helps | A supporting agent of the main one. |
| Pays | How users pay for the runs. |
| Uses | What users do: fill in the form, open the menu. |
| Delegates | The agent hands this agent a sub-task (tool ask_<agent>) and waits for its answer. |
| Then | In a team's sequence: this step runs after that one and receives its output. |
| Emits an event | Something happens here (a record written, a purchase, a decision, a call) and the trigger hears it. |
| Starts | When the event happens, the trigger starts this agent, team or data job. |
| Waits for approval | The agent asks a person to decide (tool request_<approval>); the decision resumes the work through a trigger. |
| Creates their own | Each user of the app creates instances of this template. |
| Remembers | The agent reads what it remembers about the user at every task and updates it (tool remember). |
| Answers on | Conversations from this channel reach the agent. |
Adding parts
The bar above the canvas adds a part with sensible defaults, selects it and opens it in the inspector.
| Button | What it adds |
|---|---|
| + Agent | An agent (main role if the app has none, else helper), model claude-sonnet-5-5, 6 turns, 0.5 USD budget, a goal with one optional line per form field. |
| + Form field | A text field named field, field2… |
| + Output | A printable report view, or a My runs history view if a report already exists. |
| + Page | A page with a hero block. The first one is the home page, visible without signing in. |
| + Pricing | Credits billing like the blank app, and a Credits view if there is none. |
| + Data | A data job with an example pipeline. Drawn dashed. |
| + Python | A Python slot with an example main(input), 3000 ms time limit. |
| + Schedule | Runs the main agent every day at 06:00 UTC (0 6 * * *), by the app. |
| + Team | A sequence team with the first two agents as steps. |
| + Trigger | A webhook trigger that starts the main agent. |
| + Approval | An approval decided by the admin role, 30-minute timeout, rejected on timeout; the main agent may request it, and a To approve view is added if there is none. |
| + Template | An agent template on the main agent, with one parameter, a weekly schedule, at most 5 per user and 10 credits per run, plus a My agents view for the user role. |
| + Channel | A phone channel answered by the main agent. Drawn dashed. |
The inspector
The column on the right edits the selected box or arrow. Its header repeats the box type, the box name and one line on what it does. With nothing selected it reads Click a box or an arrow to edit it. Drag the boxes to arrange the flow.
| Selected | What opens |
|---|---|
| Users | The app settings (as in the App tab). |
| Menu, a menu arrow or a header link | The menu editor: which views are menu items and in what order, and the header for visitors. |
| Form | The form fields. |
| The arrow from the form to an agent | The goal editor. |
| Agent or its memory | That agent's settings. |
| An MCP box | The settings of the agent that uses it. |
| Pricing | The billing settings. |
| View or page | That view's settings; for a page, Open page editor. |
| Python slot, schedule, team, trigger, approval, template, channel | Their own panels. |
| Data job | The JSON editor. |
The goal editor
The goal is the text the agent receives on every run. Click a field chip to append it; select fields also offer
{field_label}, and {date} and {lang} are always available. {?field}…{/field} keeps a line only when the field
is filled; {!field}…{/field} only when it is empty. When the app is not free, Credits per run sets the price
of the run. See Agents and automation.
Python slots
The slot panel shows the code (Python 3.12, define main(input), input is {values, lang}), the time limit
(100 to 10000 ms, default 3000), and which agents receive its values. Run test runs the code with the test input
on the real sandbox and shows the output and the time it took. Keep as example stores the test input in the
manifest.
Triggers with a webhook
A webhook trigger has Show address and secret. It shows the address https://<slug>.webrobot.eu/api/hooks/<key>
and the secret the calling service must send. The webhook answers on the published app only, not in the preview.
The page editor
Double-click a page box, or press Open page editor in its panel. The editor fills the screen: the blocks on the left, the page as users see it on the right. The top bar shows the page's address. Done or Esc closes it.
Drag a block from the palette into the page, drag blocks to reorder them, and edit each one in place:
| Block | Use |
|---|---|
| Hero | Headline, text, buttons. |
| Steps | Numbered 1 · 2 · 3. |
| Cards | Features, sources, plans. |
| Text | Free Markdown. |
| Data | Table, cards, numbers from a source. |
| App view | The run form, the history, a table, the chat… inside the page. |
| HTML | A custom block. Scripts, forms and iframes are removed; inline style is kept. |
| Small print | Disclaimer, footer. |
Buttons link to the app's own pages or to a block of the page with an anchor, chosen from a list. See Views and pages.
All settings
Under the canvas, ▸ All settings opens every section as a form, in tabs:
| Tab | What it edits |
|---|---|
| App | Name, address (read only), tagline, languages, pattern, customer domains, theme and colours, fonts, buttons, page width, the email when a report is ready, service hours, extra colours, the header. |
| Form | The form fields: name, type, required, label, help, options. |
| Data | Collections: name, fields, who records belong to, who else reaches them, retention in days; services users connect with their own API keys; quotas per plan. |
| Agents | Every agent: key, role, profile, goal, app tools, external MCP servers, model, turns, budget, instructions. |
| Views | Every view: key, type, roles, menu, title, data source and display. |
| Billing | Payments status, modes, credits, pay per use, packs. |
| JSON | The whole manifest. |
The JSON tab is applied when you leave the text area. If the text is not valid JSON it shows Not valid JSON: and the error, and nothing changes. Use it for what the panels do not show.
The Billing tab starts with the payments status of your organization: Payments are on…, Your Stripe account is not ready yet… with Finish the Stripe onboarding, or Payments are off… with Enable payments. Only an admin of the organization can enable payments. See Billing.
What the user sees
Under the inspector, What the user sees renders the selected box as your users will see it, in the app's colours and fonts: a page, the form, a sample report, the sign-in, the prices. It is drawn from the manifest you are editing, before you save. The real thing is the preview address.
Status, problems and warnings
The card under the preview shows Published v<N> or Not published yet, then valid or <N> problem(s), the app's address and each customer domain with the CNAME it needs.
- Problems (in red, up to 8 shown) are errors against the schema and the manifest rules. A draft with problems can be saved, but Publish stays disabled until they are fixed.
- Warnings (in amber, marked ⚠, up to 6 shown) are design checks, for example a collection nothing writes. They are shown only when there are no problems, and they never block publishing.
See Validation.
The apps list
/dashboard/app-studio/apps lists every app of your organization, most recently updated first. Search by name,
slug or domain.
Each card shows:
- the name, the slug and when it was last updated;
- a badge: Live · v<N> or Draft;
- Address:
<slug>.webrobot.eu, with the date it was published (only when live); - Preview:
preview-<slug>.webrobot.eu; - Domains: the customer domains of the published version.
Actions:
| Action | What it does |
|---|---|
| Edit | Opens the app in App Studio. |
| Open app | Opens the live address (only when published). |
| Export code | Opens App Studio with the export panel open. |
| Delete | Admins only. Asks for confirmation, then stops serving the app at once and revokes its key. The agents and pipelines it created stay in the organization. |
Moving an app to another organization is done by WebRobot platform administrators. See Publishing.