Prompt Tips: How to Make KroWork Understand You Better

Say it → Kro it → Keep it — that's the whole flow.

This guide covers how to express yourself clearly, add requirements effectively, pick the right model, and avoid common mistakes.

Ⅰ. Language Optimization: How to Talk So KroWork Gets It

Start with One Sentence, But Don't Stop There

KroWork supports vague starts, but every back-and-forth costs time and credits. If you already have a direction, say it upfront.

One sentence 👇

“Make me a cache cleaning tool”

A few more sentences 👇

“Build a local cache cleaner.

• Features: scan cache junk on my computer, group by type, show how much space each takes, let me check and clean in one click
• UI: click "Start Scan" first, then list results for me to choose, confirm and clean
• Style: green theme, clean and simple”

Just cover the three things KroWork needs: what it does + how it works + what it looks like.

The Universal Formula (4 Sentences for Any Scenario)

1. I want to build a ___ — what you're making
2. To ___ — what problem it solves
3. With ___ — what features it needs
4. Styled as ___ — what it looks like

Apply it to our example:

1. want to build a cache cleaner
2. To free up disk space on my computer
3. With scan cache, display size, select & clean, show results
4. Styled as green theme, clean and simple

Principles of Language Optimization

Principle	❌	✅
Specific > Vague	Make it look nice	Green theme, rounded buttons
Features > Feelings	Feels great to use	Auto-execute, one-click output
Structured > Prose	It scans cache and shows size and lets you check what to clean…	1. Scan 2. Display 3. Select & Clean

Ⅱ. Adding Requirements: How to Make KroWork Get It Right

Language optimization is about being clear; adding requirements is about being precise — giving KroWork specific execution standards and constraints.

Type 1: Describe the Workflow

Write out "what first, what next, what each step produces" — KroWork follows the flow, nothing gets missed.

For example, doing an AI×SaaS research report 👇

Step-by-step:

1. Search for industry signals from the past 12-24 months
2. Organize by category: disrupted / augmented / AI-native newcomers
3. Assign 2-3 company cases per category, with sources
4. Output a startup opportunity priority matrix

Why it works: You've defined every step and its expected output. KroWork doesn't have to guess — it executes exactly as you describe.

Type 2: Write Specific Rules (If There Are Numbers, Write Them)

Where there are clear thresholds, ranges, or lists — write them out. Don't let KroWork decide for you.

For example: past 12-24 months Typical companies: Salesforce, ServiceNow, HubSpot, Zendesk, Intercom, Notion, Canva, Glean, Harvey, Cursor Output: include cited sources, end with a priority matrix

"12-24 months," "these companies," "cited sources" — you define the boundary and depth.

Type 3: Define the Visual Style

One sentence to set the direction 👇

“Style: professional research report, clean structure, present key findings with tables and matrices”

Core principle: tell KroWork what you expect it to look like, rather than letting it guess.

• Key takeaway: Where there are specific numbers and rules, write them out. Don't let KroWork decide your thresholds — it may not choose what you want.

💡 Use Instructions

KroWork has built-in Instructions — click "More" → "Instructions" in the input bar, or manage your library in Settings.

• Save frequently-used prompts as Instructions, bring them in with one click next time
• Supports variables — auto-replaced with your current input at runtime
• Supports Markdown import/export for cross-device sync

Tip: Instructions bind to a single conversation. Remember to attach them when starting a new chat: Instructions: Preset Reusable Prompts for Kro

Ⅲ. Model Selection: Which Model for Which Scenario

Simple Rules

Simple Rules

Scenario	Pick	Why
Building apps with complex features and interactive logic	GLM 5.1 (Global) — Recommended	Strong reasoning, handles complex feature decomposition and multi-step execution
Tweaking copy, translation, simple Q&A	Deepseek v4 Flash (Global)	Fast response, sufficient for simple tasks, low credit cost
Processing long documents or large text	Kro Granite	Good long-context performance, less likely to lose earlier information

When to Switch

• Building a complex app → stick with the recommended model. Don't switch to Flash just to save credits — complex logic is more error-prone on lighter models, and fixing errors costs more than you saved
• Just tweaking a button label or a color value → Flash is enough — fast and cheap
• Quick question mid-app-build that isn't about the app itself → open a new chat with Flash; don't derail the current conversation

One-Liner to Remember

Use the recommended model for apps. Use Flash for copy edits. Not sure? Don't switch.

Note: Switching models mid-conversation requires the new model to re-read the entire context, costing extra credits. For lightweight edits like copy tweaks or color changes, open a new conversation and pick a cheaper model instead.

Ⅳ. Common Mistakes

Mistake 1: Trying to Build a Complex App in One Sentence ❌ "Make me an app"

✅ At least cover: what it does + what features + what style

One sentence can get you started — KroWork will follow up with questions. But every round of follow-up costs time and credits. If you already have a direction, just say it.

Mistake 2: Saying "Redo" When You're Not Satisfied

This is the most wasteful move. Everything KroWork built — gone with one "redo."

Wasteful	Precise
Doesn't look good, redo	The buttons on the results page are too small, make them one size bigger; add icons to the category list
Wrong features	Missing total disk capacity display, add a disk usage progress bar at the top
Not right, try again	No confirmation popup before cleaning — add an "Are you sure?" dialog

KroWork can make targeted edits on existing work. The more specific your request, the faster and more accurate the fix.

Mistake 3: Doing Multiple Different Things in One Conversation

Building a cache cleaner and a bookkeeping app in the same conversation? KroWork's context gets confused, and both tasks suffer. Overloaded context also burns more credits.

Right way: One conversation, one app. New task? Click "New Task" in the sidebar.

Mistake 4: Not Saving Your App When It's Done

The easiest step to skip — but saving isn't just "keeping," it's turning a one-time conversation into a permanent tool.

Not saved	Saved
Dig through conversations to find it	One-click launch from My Kro Apps
Costs credits to rebuild	Launch costs zero credits
Results may vary each time	Same tool, same result, every time
Only you can use it	Share with anyone — no setup needed on their end

After saving, your app works like regular software on your computer. You can see it in My Kro Apps, set it to launch at startup, install to Launchpad / Start Menu, share via link, or publish to Kro Workshop for everyone.

Can you still edit after saving? Yes. Come back to the app's conversation anytime, make changes, and update the app.

Mistake 5: Jumping Into a Complex App Without a Plan

Jumping into a complex app without a plan often leads to things going off-track halfway — and there's no easy way back.

Right way: Turn on Plan Mode, review the plan first, then execute. Fixing misalignment at the plan stage costs far less than rebuilding.

Mistake 6: Not Watching the Execution Process

The right panel shows task progress and file list.Don't walk away.Catch issues early and correct them — it's much faster than fixing after completion.

Scenario Reference

Go to the KroWork homepage (or click "Start" in the top-left corner) to see four template apps. Click any one to view its prompt. Learn from these templates and adapt them — it's a fast way to improve your own results.

Template	What you can learn
Quick Image Generator	Describe a complete interaction flow — input → refine → generate → save → send — and visual style details
Dual-Key Piano	Specify key mappings, sound parameters, chord rules; write numbers when you have them
Deep Analysis	List analysis dimensions, case companies, and output requirements to define scope and boundaries
Cache Cleaner	Break down the UI by states — idle → scanning → done → cleaning → done — and define safety rules

More apps in Kro Workshop: AI Stock Review, Pawker — Texas Hold'em, 2048 Game, Job Hunt, Smart Desktop Cleaner, Images Mosaic Tool, SillyTavern, Pixel To-do List… See how others write their prompts and use similar structures for your own.

Remember: KroWork's core value isn't just conversation — it's turning your ideas into saved desktop apps. Every time you build something, save it. It becomes your personal software — always ready, permanently yours.