This is the entry-point doc for the text workbench, one of four lanes in x-rush alongside image, video, and audio. Three features live here, and each is built around one object you manage. Everything you create lands inside a Project: the workbench always has a current Project, switching it from the top-left navigation switches every list below, and the first Project on every account is a protected Default Project that can't be deleted.
The accent color on this lane is #00D4AA, a teal-green you'll see on buttons, focus rings, and active tabs. Image, video, and audio each carry their own accent, so the color tells you which workbench owns the page before you read a word.
Where the Text Workbench Fits
Text is the writing-and-publishing lane: pages you publish, characters you talk to, content that gets written on a schedule while you're away. Generation, editing help, and chat replies all run through the same router, which picks the text model that fits each task and updates its pool as better models ship. You never pick a vendor by hand.
Three Features at a Glance
| Feature | What it produces | The object you manage | Entry path |
|---|---|---|---|
| HTML Studio | Web pages on a public URL | Workspace + publish versions | /workbench/text/html |
| AI Chat | Conversations and written output | Character + persistent history | /workbench/text/chat |
| Automations | Scheduled text, written by a character | Automation + Run History | /workbench/text/auto |
HTML Studio — the page you can ship. A workspace starts from a name, optional tags, and a plain-language task description; the first draft generates for a fixed 0.5 credits, shown on the confirm button before you pay. After that, each confirmed AI edit bills the same fixed 0.5 credits as the first generation while your own code edits stay free, and both update one auto-saved draft. Nothing enters history until you publish. Publishing creates an immutable version and points your stable public URL at it: the /p/{id} address opens for anyone with no login, and republishing moves that same URL to the new version. Old versions stay in your publishing history with no cap, and any of them can be re-published later. Publish also asks who can see it: a plain public link, or community, where the page additionally appears in Explore and other users can Remix it. A Remix is a one-time copy into the other user's own Project; it carries no attribution banner and no link back to you. Exporting the current draft as a ZIP is free. The boundaries: no backend, no database, no custom domain in this version.
AI Chat — the character you own. The character list is the entire entry point. There is no "new conversation" button and no session management to learn; you pick a character and continue its thread. Built-in characters belong to the Default Project, so you see them only while that Project is current. Switch to a normal Project and you see what that Project owns: characters you created, copied, or Remixed. Every character keeps one persistent chat history that survives refreshes, closed tabs, and other devices. Switching characters switches histories; nothing bleeds across. A character is yours to shape: role setup, behavior rules, background material, a web-search ability toggle, avatar, chat background, opening message. Built-in characters protect their core configuration; to change one, copy it as a custom character into whatever Project you pick and edit the copy. When a thread produces something worth keeping, share selected messages as a read-only snapshot, or save a reply as a document in the current Project.
Automations — the schedule you set once. An automation runs one character on a schedule, and the character has to exist first, in the current Project. The binding is one-to-one and permanent: a character bound to an automation can't be bound to another, and an automation can't be re-bound to a different character. Setup is a six-step form: basics, task requirements, web-search permission, output control, schedule, confirmation. The schedule means daily, weekdays, or custom weekdays, one time per day, in a timezone you state explicitly. Enabling runs through a free config preview first, with an optional paid trial run at the same 0.5 credits as any text generation. When a scheduled run succeeds, its output lands as a message in that character's chat history, marked as an automated run. The Automations list and AI Chat show the same message, and clicking through from the list lands you on that exact message, already in context. Every execution attempt lands in the automation's Run History, including user-cancelled, success, failure, skipped, insufficient balance, and content blocked. Two behaviors worth knowing: a run your balance can't cover pauses the automation instead of failing silently every morning, and deleting an automation stops future runs without touching anything it already wrote.
Delivery is in-site only. Characters with an enabled automation appear in the site-wide quick chat bubble, unread counts add up across all your Projects, and clicking an item opens that message. No email, no SMS, no webhook.
How the Three Chain Together
Say you've built a custom character called "Landing Page Critic", fed it your product details as background material, and bound an automation to it: 08:00, weekdays, your timezone. Overnight, a run executes. In the morning the bubble shows one unread, and clicking it lands you on a fresh critique sitting in the character's history.
You push back in the same thread. "Angle two, but lead with the pricing objection" gets a rewrite in context, because the automated message and your replies share one history. When a version survives, you open HTML Studio, create a workspace, and type a one-line brief: "single-screen waitlist landing page, dark theme, mobile first". 0.5 credits, one generation, a few edits, then publish. The page lives at the same /p/{id} URL your previous version used.
That's the loop: Automations writes on a schedule, AI Chat develops the idea, HTML Studio puts it on a public URL. Each handoff is a click or a paste, and nothing gets duplicated on the way.
How It Talks to the Other Modules
A published page is plain HTML, so it embeds whatever the other workbenches produce: an image from Image Studio or a clip from Video Studio renders inline. The reverse direction is looser — a script worked out in AI Chat can become a voiceover or a caption set, and today that handoff is mostly copy-paste. The formal bridge is the document: save a reply as a document and it belongs to the current Project, where other features can pick a saved version from their content selector. There is no "send to another workbench" button, on purpose. What leaves the chat leaves because you chose it.
Where to Start if You're New
Start with HTML Studio. Open /workbench/text/html, create a workspace with a name and a one-sentence brief, and run it. You'll have a real page on screen after one generation, and publishing it to a working public URL teaches the shape of the product faster than this doc does.
Then open AI Chat in the Default Project and talk to a built-in character. If one clicks, copy it into your own Project as a custom character and reshape it: new rules, your background material, your color on the chat background.
Save Automations for last, for a practical reason: the binding is permanent and the schedule runs without you, so it only makes sense once you know which character you want to hear from every morning. Set it up, read the free preview, and let the first scheduled run close the loop. The morning the bubble shows an unread message written while you slept is when the three features start behaving like one workflow.