Eight hundred people expected a tarot pull, a horoscope, and a fitness nudge from me every morning. I wrote all three by hand, at 6am, for sixteen straight days. On day seventeen I overslept, and by day twenty I just didn't. The community didn't leave - but the ritual frayed. The problem wasn't ideas; it was showing up. I needed the content to generate itself, on a clock, whether I'd had coffee or not. That's the T-2 Auto Content module in x-rush, and it's the reason my mornings are mine again.
What Auto Content actually is
Open /workbench/text/auto and you get two tabs. Content is a feed - every push the agents have ever made for you, scrollable and filterable. Tasks is the control panel - where you decide which agent pushes, when, and how often.
The mental shift that matters: this is not a chat. You don't talk to it. It talks to you, on a timer, and you read. Seven built-in agents - tarot, trivia, zodiac, fitness, study, MBTI, nutrition - take turns generating a daily card from rules you set. When you actually want a back-and-forth, that's a different button (more on that later).
Setting up the first push (the actual clicks)
I started with tarot because it needs zero setup - a card pull is a card pull.
- Click the Tasks tab, then + Add Task. A modal opens with a handful of fields.
- Agent - pick from a grid of seven, each showing its emoji (🔮 tarot, 💡 trivia, ♈ zodiac, 💪 fitness, 📚 study, 📊 MBTI, 🍎 nutrition). I picked tarot.
- Push toggle - flip it on. Off means the task exists but stays dormant.
- Push time - a 24-hour clock. I typed
08:00. - Frequency - daily, weekdays only, or custom (tick any combo of Mon-Sun). I chose daily.
- Description - free text. I wrote "morning pull, focus on action and energy."
Hit Create. The task shows up as a card in the list. The toggle is optimistic - it flips the moment you click, the server confirms in the background. If the confirm fails it snaps back and you see it. No spinners, no "saving..." theater.
The next morning at 8am I opened the Content tab and there it was: a card with a bright bar on its left edge - the unread marker. I clicked it. Two things happened at once. The bar faded (marked as read), and a detail modal opened with the full structured payload. Not a wall of text. The tarot card did a 3D flip, showed upright and reversed interpretations, and a five-dimension star-rated advisory underneath. That's the whole point of seven specialists instead of one generalist - each one lays its content out differently.
The part most people skip (and regret)
Here's the trap. Five of the seven agents are personalized, and they're only as good as what you tell x-rush about yourself. The other two - tarot and trivia - are web-wide and push fine with no setup at all.
The personalized five need real input, and the system prompts spell out exactly which:
- Zodiac wants your sign.
- Fitness wants your goal, training level, available days per week, and equipment.
- MBTI wants your type.
- Study wants the field you're learning and your goal.
- Nutrition wants your goal, dietary preferences, and restrictions.
I set up zodiac for the community default (Leo, since I had to pick something) and nutrition for myself - goal "maintain," preferences "high-protein," restrictions "no shellfish." The difference was immediate. Before I filled those in, the nutrition push was a generic "drink more water" plan. After, it came back with a three-meal breakdown (breakfast 350 kcal, lunch 500, dinner 450), a macro ring, and swap suggestions: "no eggs today? Greek yogurt hits the same protein for fewer calories." Skip the personalization step and you'll blame the agent for being generic. It isn't - it's working blind.
What each push actually looks like
This is the part that sold me, because I'd been hand-formatting these every morning. The detail modal is structured per agent:
- Tarot - 3D card flip, upright/reversed read, five-dimension star rating.
- Zodiac - a six-axis score card (love, career, finance, health, study, social) plus lucky elements: color, number, direction, time window.
- Fitness - a training table with exercises, sets and reps (squat 4×8, bench 4×10, pull-up 3×6), a diet check-in tracker, and a calorie target.
- Nutrition - three-meal breakdown with calories, a macro ring, and swap suggestions.
- Study / MBTI - a study plan block, or a personality deep-dive with E/I, S/N, T/F, J/P dimension rows.
I used to spend twenty minutes making the horoscope look like a horoscope. Now it just does.
Managing tasks without the dread
Editing a task reopens the same form, pre-filled. Deleting pops a DeleteTaskConfirmModal - yes, it asks twice, because deleting a task also drops its push history from the schedule, and that's annoying to rebuild. I appreciate the double-confirm; I've fat-fingered enough delete buttons in my life.
One thing that confused me at first: I couldn't find the floating chat bubble anywhere inside the workbench. That's deliberate - it's hidden across all of /workbench/* so it doesn't interrupt you while you're actually producing. It only shows on marketing pages, the user center, and the blog. Not a bug.
When a push makes you curious
Reading is passive; sometimes a card makes you want to push back. Every detail modal has a Continue chatting link at the bottom. Click it and you land at /workbench/text/chat?agent=<the-agent>&context=<this-card>. Two things happen: the agent is pre-selected (you don't re-pick tarot just because you were reading a tarot card), and the card you came from is quoted at the top of the thread as context. One click, full context, mid-conversation. Each agent keeps its own history too - switch from the Nutritionist to the Tarot Reader and back, and the nutrition chat is exactly where you left it.
From there the loop closes. A daily push drops an idea, I take the interesting ones into a deep chat, and when something's worth sharing I copy it into the HTML Pages editor (/workbench/text/html), hit Publish, and get a standalone URL at /p/{id} that anyone can open without logging in. Push generates, chat deepens, publish ships. The teal #00D4AA accent runs through all three, which sounds cosmetic until you realize how fast it orients you - teal means text land.
Pro tips I learned the hard way
- Fill the personalization fields before your first personalized push. An empty sign means a generic horoscope, and the agent won't nag you about it.
- Use the frequency settings for a five-day week. "Weekdays only" and custom day-picking exist for a reason - weekend pushes into a community that's quiet on Saturdays just feel like noise.
- The optimistic toggle is your friend for testing. Flip a task off, confirm the push stops, flip it back on. The history stays either way.
- Don't delete to pause. Use the toggle. Delete is permanent and double-confirmed for a reason.
What's doing the generating
Every agent runs on x-rush's smart routing across the industry's top-tier text models - Claude Fable 5, GPT-5.6, Gemini 3.5, Qwen3.7 Max, and the rest. You don't pick a model; the router sends each task to the one that fits it best, and the pool refreshes as the world ships new ones. A tarot pull and a nutrition plan don't need the same engine, and they don't get it. Generation stays in the one-to-three-second range, which matters when a card is supposed to land at 8am sharp.
Where to go next
If any of this sounds like the morning you want back:
- Auto content generator - the full operation manual: all seven agents, task scheduling, the chat handoff
- Text module overview - how HTML Pages, Auto Content, and AI Chat fit together
- AI characters & chat - the T-3 side, where pushes turn into conversations
I came in trying to claw back twenty morning minutes. I stayed because the ritual stopped depending on me having a good day. Set it once, let it run, and the cards show up whether you opened the workbench or not.