Build
Chat Modes
Rytora BuildLabs has 6 modes for different situations — from generating code to debugging to quality testing. Click the mode dropdown in the bottom-left of the chat to switch.
Quick reference
Deep mode is a separate toggle (not in the dropdown). Activates extended reasoning for complex architecture decisions. Works with any mode. Uses more credits.
⚡ Build
The default mode. Describe what you want and the AI generates production-ready code — React pages with Tailwind and shadcn/ui, Supabase integration for database and auth, and the wiring between them.
How it works
Your prompt goes through 3 AI layers: Architect (plans the structure), Code Generator (produces frontend and Supabase integration), and Validation (ensures quality). Each layer is a specialized AI prompt optimized for its task.
When to use: You want the AI to generate or modify code in your project.
Try it:
Create a landing page with hero section, pricing, and testimonialsAdd a dashboard with charts showing sales dataAdd a user settings page with profile edit, password change, and notification preferencesTips:
Be specific: "dark theme, shadcn/ui cards, responsive grid" beats "make it look nice"
Mention tech when it matters: "use Supabase for the database" or "add Stripe checkout"
For complex features, use Plan mode first — it saves credits by avoiding bad generations
Each generation costs credits based on the amount of code produced
💬 Ask
A conversational mode that doesn't trigger code generation. Ask anything: debug errors, compare technologies, get architecture advice, or understand code the AI just wrote. Works even while a Build is running.
How it works
Your message goes to a chat-optimized AI call that doesn't run the generation pipeline. No files are created or modified. The AI has context of your project and conversation history.
When to use: You have a question — about an error, a concept, or which approach to take.
Try it:
What does this error mean: Cannot read properties of undefined?Should I use Supabase or Firebase for real-time features?Why did you create a separate AuthContext instead of using NextAuth directly?How much would it cost to add Stripe payments?Tips:
Use this while Build mode is running — it won't interrupt the generation
Ask "why" after the AI makes a decision — it explains its reasoning
Great for learning: "Explain how TanStack Router works in this project"
Cheapest mode — uses minimal credits
🔍 Explore
An autonomous AI agent that reads through your project files to find answers. It searches file names, scans code content, reads implementations, and traces how components connect — all read-only. Like having a senior developer who can instantly navigate your entire codebase.
How it works
Spawns a read-only agent that makes multiple tool calls in a loop: searching files with glob/grep patterns, reading source code, listing directories, and running read-only shell commands (git log, ls, etc). Typically makes 3-10 tool calls in 10-30 seconds.
When to use: You want to understand existing code without changing anything.
Try it:
How does the authentication flow work end-to-end?Where is the payment processing logic? What functions handle it?What API endpoints does the app expose?Show me all the database models and their relationshipsTips:
Always start here when returning to a project you haven't touched in a while
Use before Build: "How does auth work?" then "Add role-based permissions"
Costs 1-3 credits — cheaper than Build because it only reads, never writes
The agent shows its search steps in real-time so you can follow along
📋 Plan
The AI explores your codebase and creates a detailed implementation plan before any code is written. It identifies which files need to change, proposes the architecture, lists dependencies, and provides step-by-step instructions. Like having a tech lead do a design review before you start coding.
How it works
Spawns a planning agent that reads your codebase structure, understands the existing architecture, then produces a structured plan with: implementation steps, critical files to modify, new files to create, potential risks, and estimated complexity. The plan appears in chat for you to review and refine.
When to use: You're about to build something complex and want a clear roadmap first.
Try it:
Add a notification system with email (Brevo) and in-app alertsRebuild the dashboard with real-time WebSocket updatesAdd multi-language support (i18n) to the entire appMigrate from REST to GraphQL for the user-facing APITips:
Use for any feature that touches 3+ files — the plan prevents wasted generations
Review the plan and tell the AI to adjust before switching to Build
You can say "Now build step 1" to implement the plan incrementally
Costs 2-5 credits — pays for itself by avoiding bad Build attempts
🤖 Agent
The most autonomous mode. The AI works independently: reading files, searching for the root cause, making targeted edits, running commands, and even searching the web for solutions. It keeps iterating until the task is done or it reaches the time limit (2 minutes). If something goes wrong, it automatically restores your files.
How it works
Spawns a full-access agent with 8 tools: read files, search code (glob/grep), list directories, run shell commands, edit files, create files, and web search (powered by Anthropic). The agent calls tools in a loop, with each response deciding what to do next based on results. Includes snapshot/rollback: original file contents are saved before edits and restored on failure.
When to use: You have a task that needs investigation + action — like fixing a bug, refactoring, or debugging.
Try it:
Fix the broken submit button — it sends the form but shows no success messageAdd input validation to all forms (required fields, email format, min length)The /api/users endpoint returns 500 — find the bug and fix itRefactor the auth middleware to support both JWT and API key authenticationTips:
Describe the problem with context: what should happen vs what actually happens
Agent mode costs 3-8 credits because it makes multiple AI calls in a loop
Watch the real-time progress — you can see each file it reads and each change it makes
Files are auto-restored if the agent fails — your project stays safe
For simple changes, Build mode is faster and cheaper
🛡️ Verify
A QA agent that tries to break your app on purpose. It runs the build process, checks for compile errors, tests edge cases (empty inputs, null values, missing data, concurrent requests), and reports a clear PASS/FAIL verdict with evidence. Catches bugs before your users find them.
How it works
Spawns a read-only adversarial agent. Strategy: (1) run build/lint/typecheck, (2) test core functionality, (3) probe edge cases with intentional bad input, (4) check error handling. Each check includes the exact command run and actual output observed. Final verdict: PASS (all good), FAIL (critical issues), or PARTIAL (works but has problems). Also runs automatically in the background after every generation of 3+ files.
When to use: You've made changes and want to check everything works before deploying.
Try it:
Check if the app builds and all pages render without errorsTest the checkout flow — what happens with invalid card data?Verify the API handles missing auth tokens correctlyDoes the form handle empty submissions, special characters, and very long input?Tips:
Run after every major Build — it's the cheapest insurance against bugs
Pay attention to FAIL results — they include the exact error and where it happened
Verify runs automatically after Rytora BuildLabs generates 3+ files (you don't need to trigger it)
Costs 2-5 credits — worth it to avoid shipping broken code
You can verify specific features: "test only the payment flow"
Recommended workflow
Use the modes together for the best results. Here's the flow most experienced users follow:
Ask or Explore first
Understand your project. Use Ask for quick questions or Explore to navigate the existing code. Know what you have before changing it.
Plan before building big features
For anything touching 3+ files, switch to Plan. Get a roadmap with file changes, architecture decisions, and implementation order. Review and adjust.
Build to generate code
Switch to Build and describe what you want. Be specific. If you have a plan, reference it: "Build step 1 from the plan above."
Use Agent for fixes and debugging
Something broken? Switch to Agent. Describe the symptom: "The form submits but shows no confirmation." The agent investigates and fixes autonomously.
Verify before deploying
Run Verify to catch bugs. It tries to break your app, tests edge cases, and reports PASS/FAIL. Fix any failures, then deploy with confidence.
How to switch modes
Option 1: Mode dropdown
Click the mode button (shows "Build" by default) in the bottom-left of the chat input. Select from the dropdown and start typing.
Option 2: Slash commands
Type the command directly in the chat: /explore/plan/agent/verify/coordinate followed by your message.
Credit usage by mode
Credits vary based on the complexity of the task. Heavier context (large projects) may cost slightly more.