Sending messages
Type your message in the chat panel and press Enter or click the send button. The AI reads your message, plans its approach, and starts working on your app immediately. You can send messages at any time — when creating a new feature, fixing a bug, asking a question, or requesting a design change. The AI maintains context from your entire conversation history within the current thread.What the AI can do
The AI agent has over 40 specialized tools at its disposal. Here is what it can do in your app:| Capability | Description |
|---|---|
| Create files | Generate new pages, components, API routes, and utility files |
| Edit files | Modify existing code with surgical precision — full rewrites or targeted edits |
| Delete files | Remove files that are no longer needed |
| Search code | Find specific patterns, functions, or references across your codebase |
| Run SQL queries | Execute queries against your app’s built-in PostgreSQL database |
| Install packages | Add pre-approved npm packages to your project |
| Connect integrations | Wire up services like Slack, Jira, Salesforce, Stripe, and more |
| Generate API clients | Create typed client code from OpenAPI specifications |
| Run type checks | Validate TypeScript types across your project |
| Run tests | Execute test suites to verify your app works correctly |
| Read logs | Check development server logs for debugging |
| Copy components | Use pre-built Shadcn UI component patterns |
Agent mode vs. Ask mode
Vybe offers two interaction modes, selectable from the mode toggle in the chat panel.Agent mode (default)
In Agent mode, the AI can read and modify your app. It creates files, edits code, runs database queries, installs packages, and connects integrations. This is the mode you use for building and iterating on your app.Ask mode
In Ask mode, the AI operates in read-only. It can read your files and answer questions about your code, but it cannot make any changes. Use Ask mode when you want to:- Understand how a specific part of your app works
- Get explanations of code the AI previously generated
- Ask architecture or design questions without risking accidental modifications
- Review logic before deciding what to change
Attaching images
You can attach images and screenshots to your messages. This is useful for:- Showing a desired design — Attach a mockup or screenshot of a UI you want the AI to replicate
- Reporting errors — Screenshot an error in your browser and send it to the AI for debugging
- Referencing layouts — Share screenshots of existing tools or dashboards you want to recreate
Clarifying questions
When the AI needs more information to proceed, it asks clarifying questions through interactive forms. Instead of guessing, the AI presents you with options to choose from or fields to fill in. For example, if you ask the AI to “add a chart”, it might ask:- What data should the chart display?
- What chart type do you prefer (bar, line, pie)?
- Should it pull data from the database or an integration?
Streaming responses
The AI streams its work in real-time. As it generates code, you can see:- The AI’s reasoning and plan of action
- Each file being created or modified
- Database queries being executed
- The live preview updating as changes are applied
Thread management
Each app can have multiple conversation threads. Threads let you organize different workstreams or features independently.- Create a new thread to start a fresh conversation without losing previous context
- Switch between threads to pick up where you left off on a different task
- Each thread maintains its own conversation history
- The AI reads the full history of the active thread when responding
Changes made by the AI apply to the app regardless of which thread initiated them. Threads organize your conversations, not your code.
Auto-suggestions
After the AI completes a task, it suggests next features or improvements you might want to build. These suggestions appear as clickable prompts below the AI’s response. Suggestions are contextual — they are based on what you just built and what would logically come next. You can click a suggestion to send it as your next message, or ignore them and type your own prompt.Example prompts
Here are effective prompts for common tasks: Building a dashboardBuild a dashboard that shows our sales pipeline from Salesforce with filtering by stage and date range.Adding database features
Add a new page that lets users search and filter our PostgreSQL customer database.Connecting integrations
Connect to our Jira project and create a view that shows all open bugs assigned to the current team.Improving UI
Redesign the customers table to use alternating row colors, add pagination with 25 rows per page, and make the email column clickable.Debugging
The chart on the analytics page is not loading. Check the API route and fix any issues.Adding new pages
Create a settings page where users can update their notification preferences and timezone.
Tips for effective prompting
- Be specific — “Add a sortable table with columns for name, email, status, and created date” is better than “add a table”
- Mention data sources — Tell the AI where data should come from (database, integration, or hardcoded)
- Describe the layout — Mention how you want elements arranged on the page
- Iterate in small steps — Build one feature at a time rather than requesting everything in a single prompt
- Reference existing pages — “Make this page look like the customers page” helps the AI match your existing style
What’s next
- Choose the right AI model for your task: Choosing a Model
- Understand the editor layout: The Editor
- Learn about the live preview: Preview & Sandbox