Create a workspace, paste your website URL, customize the widget, copy one snippet. Done.
4 AI providers
Pick your model
OpenAI (GPT), Gemini, Claude, or Grok — multiple versions of each. Switch anytime from the dashboard.
One training, every channel
Omnichannel from day one
Train the AI once. It works on your website, WhatsApp, Instagram DMs, and Facebook Messenger.
01
First steps
Create your workspace and first bot.
A workspace is your account-level container. Inside it, you create bots (AI agents), manage team members, handle billing, and control API access. You can create unlimited workspaces — useful for agencies or resellers managing multiple clients.
Sign up at app.zybots.ai and create your first workspace.
Give your workspace a name (e.g., your company or client name).
Create a bot inside the workspace — this is your AI assistant.
Each bot has its own training, settings, channels, and analytics.
1
Create workspace
Sign up, name your workspace, and you're in the dashboard. Takes 30 seconds.
2
Create your first bot
Click 'New bot', give it a name, and you land on the Assistant tab — your control center.
3
Set up AI behavior
Go to Settings > AI. Choose your model, set the temperature, pick an agent preset (Sales, Support, Technical), and edit the system instructions.
Each workspace has its own billing, members, and API key. Agencies can create a workspace per client, pick a plan, set their own pricing, and manage everything independently — fully white-labeled.
02
Knowledge base
Train your agent on real content.
Your AI assistant only knows what you teach it. The Training tab lets you add content from multiple sources. The agent answers strictly from this knowledge base — no hallucinations, no internet searches.
Website training: paste a URL and Zybots crawls your pages. Supports sitemap discovery, path include/exclude rules, and page limits.
File upload: upload PDFs, documents, or text files. Zybots reads the content and adds it to the knowledge base automatically.
Q&A pairs: manually add question-answer pairs for precise control over specific topics.
Free text: type or paste any content the agent should know — policies, procedures, product specs.
Corrections: when the agent gets something wrong, add a correction to override its behavior for that specific question.
1
Add your website
Go to Training > Add source > Website. Paste your URL. Choose discovery mode (full site or single page), set max pages, add path filters if needed.
2
Upload documents
Go to Training > Add source > Files & text. Upload PDFs, docs, or paste raw text. Zybots processes and indexes everything automatically.
3
Fine-tune with Q&A and corrections
Add Q&A pairs for questions that need exact answers. Use corrections when you spot a wrong answer in a real conversation.
What happens when you train from a website
text
1. You paste your URL (and optional sitemap)
2. Zybots discovers and reads all your pages
3. Content is processed, indexed, and ready for chat
4. When a visitor asks a question, the AI searches your indexed content
5. It finds the most relevant sections and builds an answer from them
6. If nothing matches, it says it doesn't know — no guessing
Auto-retrain: on Growth+ plans, you can schedule automatic retraining (daily or custom) so your knowledge base stays current when your site changes.
03
Model selection
Choose the right model for your use case.
Zybots supports multiple AI providers and model versions. You pick the primary model and an optional fallback. The model is set per-bot in Settings > AI — you can switch anytime without retraining.
OpenAI: GPT-4 and GPT-5 family (multiple versions — mini for speed, full for depth).
Anthropic: Claude models for nuanced, safety-conscious conversations.
Google: Gemini models for multilingual and multimodal use cases.
xAI: Grok models for fast reasoning and direct conversation style.
Fallback model: set a secondary model that kicks in if the primary fails or times out.
1
Pick primary model
Settings > AI > Chat model. Choose provider and version. Lighter models are faster and cheaper; larger models handle complex questions better.
2
Set temperature
Lower = more consistent, factual answers. Higher = more creative, varied responses. For support bots, keep it low. For sales, slightly higher.
3
Choose agent preset
Pick from Base, Sales, Teacher, Customer Support, or Technical Support. Each preset loads optimized system instructions — you can edit them further.
The model you choose doesn't affect training — your knowledge base stays the same. You're only changing how the AI reasons about the content it already has.
04
Integrations
Let your AI do things, not just talk.
AI Actions let your assistant execute real operations during a conversation — look up orders, process payments, book appointments, escalate to your team. Actions run mid-conversation, triggered by user intent, with optional confirmation before execution.
Shopify: search products, look up orders, check stock, cancel orders.
WooCommerce: browse catalog, check order status, verify stock levels.
Stripe: create subscriptions, look up invoices, process cancellations, swap plans, update billing.
Calendly & Cal.com: check available slots, book appointments, send confirmations.
Slack: escalate conversations to your team with full context and a 'View in Dashboard' link.
Zendesk: create support tickets from conversations automatically.
Zybots App: real-time dashboard notifications with sound alerts when a conversation needs human attention.
Web Search: let the agent search the web when the knowledge base doesn't have the answer.
Custom REST API: connect any system via custom endpoints — CRM updates, inventory checks, proprietary workflows.
1
Add an action
Go to your bot > API Actions tab. Choose a preset (Shopify, Stripe, etc.) or create a Custom REST action. Configure credentials and endpoints.
2
Configure behavior
Define what information the AI needs to collect from the conversation. Enable confirmation prompts for sensitive actions (payments, bookings, cancellations).
3
Test it
Use the built-in test mode to try it live. Make sure the AI understands the request and the integration responds correctly.
Example: booking an appointment via chat
text
1. Visitor asks: 'Can I book an appointment for Friday?'
2. The AI understands what they need and checks your calendar
3. It finds available slots and replies: 'I have 10am, 2pm, and 4pm open. Which works?'
4. Visitor picks a time → the AI books it instantly
5. Confirmation happens right in the conversation — on whichever channel they're using
05
Deployment
Put your AI where your customers are.
Embed the chat widget on your website with a single script tag. Then extend to WhatsApp, Instagram DMs, and Facebook Messenger — same AI, same knowledge, every channel. All conversations flow into one unified dashboard.
Widget: copy the embed snippet into your site. Works with any platform — HTML, WordPress, Shopify, React, Next.js.
Appearance: customize colors, launcher style, welcome messages, starter questions, and brand assets in Settings > Appearance.
Domain controls: restrict which domains can load your widget in Settings > Security.
WhatsApp: connect your WhatsApp Business account. The AI responds to incoming messages just like on any other channel.
Instagram: connect your Instagram Business account. The AI handles DMs automatically.
Facebook Messenger: connect your Facebook Page. Same AI, same training, Messenger channel.
1
Embed on your site
Go to Assistant tab > copy the script snippet. Paste it before </body> in your HTML. The widget appears immediately.
2
Customize appearance
Settings > Appearance. Set your brand colors, upload a logo, configure welcome messages and up to 5 starter questions.
3
Connect social channels
Go to Omnichannel tab. Connect WhatsApp (via Meta embedded signup), Instagram, or Messenger. Each channel can be toggled on/off independently.
All conversations — widget and omnichannel — appear in the Activity tab. You can search, filter, review, and export from one place.
06
Insights
Track what matters. Export what you need.
The Analytics tab shows real numbers — sessions, messages, leads, resolution rates, top countries, and channel performance. The Activity tab gives you a conversation inbox with lead management. Everything is exportable.
Dashboard overview: total sessions, messages, leads, and active bots at workspace level.
Bot analytics: daily trends, lead capture rate, session size distribution, user vs assistant message split.
Top countries and top sessions: available on Growth+ plans for deeper segmentation.
Lead capture: leads are collected via AI extraction during chat, custom forms, or manual capture. Each lead has name, email, phone, notes, and custom fields.
Conversation inbox: search, filter, and review every conversation. Link leads to their chat history.
Export: CSV, XLSX, or PDF exports for conversations and leads (plan-gated).
Sentiment analysis: available on Enterprise — tracks customer sentiment across conversations.
1
Review conversations
Activity > Conversations. Click any session to read the full transcript. Spot gaps in training, missed escalations, or incorrect answers.
2
Monitor lead quality
Activity > Leads. Filter by status, source, or date. Review which leads came from AI extraction vs custom forms.
3
Close the feedback loop
When you find a wrong answer in a conversation, go to Training > Corrections and add a fix. The agent learns from it on the next retrain.
07
Configuration
Every setting in one place.
Each bot has a full settings panel covering AI behavior, appearance, lead forms, and security. Workspace-level settings handle team members, billing, API keys, and roles.
Forms: custom lead capture forms with configurable fields — presets or fully custom. Form count limited by plan.
Security: allowed widget domains (limited by plan), rate limiting with custom messages per locale.
1
Configure AI behavior
Settings > AI. Pick your model, set temperature, choose a preset, then edit the system instructions to match your brand voice and rules.
2
Brand the widget
Settings > Appearance. Match colors, upload your logo, set the launcher position, and customize the welcome experience.
3
Set up lead forms
Settings > Forms. Create custom forms that appear during conversations. Define fields, labels, and validation. The AI decides when to show them based on conversation context.
08
Collaboration
Manage teams, roles, and multiple workspaces.
Workspaces are the top-level organizational unit. Each workspace has its own bots, billing, members, and settings. You can create unlimited workspaces from one account — ideal for agencies, resellers, and multi-brand companies.
Members: invite team members with email. Assign roles that control access to bots, settings, analytics, and billing.
Custom roles: on higher plans, create custom roles with granular permissions (e.g., view-only analytics, training-only access).
Workspace switching: jump between workspaces from the sidebar. Each workspace is completely isolated.
API keys: each workspace has its own API key for programmatic access to exports, training, and conversation data.
Audit logs: on Enterprise, track who did what — every settings change, training action, and member update is logged.
Built for resale: agencies and consultants can create separate workspaces for each client, choose a plan, set their own pricing, and manage everything from one account — fully white-labeled.
Workspace isolation means one client's data, training, and conversations never mix with another's — even if managed from the same account.
09
Human handoff
When the AI should step aside.
Not every conversation should be handled by AI. Zybots supports structured escalation — the AI detects when it can't help and hands off to your team through the channel you've configured.
Slack escalation: the AI posts the full conversation transcript to a Slack channel with a 'View in Dashboard' button. Your team picks it up from there.
Zendesk: automatically creates a ticket with conversation context.
Zybots App: real-time notifications in the Zybots dashboard when a conversation needs human attention — with sound alerts and full context.
Custom escalation: use any REST endpoint — CRM, helpdesk, internal tool.
Operator mode: when a human takes over, the AI stops responding. The conversation stays in the dashboard and the agent pauses until the operator releases it.
Bidirectional Slack: with the bidirectional install, operators can respond directly from Slack and the reply goes back to the customer.
Multi-participant conversations: you can take over a conversation from the Zybots dashboard and from Slack at the same time. This means three participants — the AI, a dashboard operator, and a Slack operator — can all be part of the same chat with the customer.
More integrations are coming — we're actively adding new escalation channels.
You control when escalation triggers — configure it as an AI Action with specific conditions, or let the AI decide based on conversation complexity and confidence.
Ready to build your first agent?
Create a free account, paste your website, and have a working AI assistant in minutes. No credit card required.
