Velaro Bots and AI Guide

Bot Personas — Overview

Every AI bot in Velaro is driven by a Bot Profile — a reusable character layer that controls how the bot looks, speaks, sounds, and thinks. Profiles live under Settings → Bots → Personas. You must create at least one profile before using the AI/Bot node in a workflow.

Each profile has four layers: Identity, Character, Voice, and AI Model. Any workflow AI node that references a profile automatically inherits all four layers. This means you can switch a bot's entire personality and model by swapping the profile — no need to touch the workflow.

Bot Personas — Identity

Go to Settings → Bots → Personas → New Persona.

• Name: what visitors see as the bot's display name in chat
• Avatar: choose from 24+ presets (AI Bots, Human-style, Illustrated) or upload your own image
• Default: mark one persona as Default — it's used automatically by any AI/Bot node that doesn't specify a persona

You can have multiple personas for different workflows (e.g. "SupportBot" for support, "SalesBot" for sales).

Bot Personas — Character (Tone and Greeting)

The Character layer controls the bot's personality without touching the system prompt.

Tone sets the communication style injected automatically into every response:

• Friendly — warm, casual, encouraging
• Professional — formal, concise, to the point
• Empathetic — supportive, understanding, patient
• Concise — minimal, direct, efficient

Greeting is the bot's opening message when a conversation starts. Leave it blank to use the workflow's default first message. Set it to personalize the welcome per persona (e.g. "Hi! I'm Alex, how can I help you today?").

The tone and greeting override anything set at the workflow node level. Set the persona once; every workflow using it gets the same voice.

Bot Personas — Voice (ElevenLabs)

Enable voice to have the bot read its responses aloud using ElevenLabs text-to-speech.

• Voice: choose from all available ElevenLabs voices
• Voice Model: Turbo (fastest), Flash (balanced), Multilingual (non-English support)
• Auto-play: when on, audio plays automatically without the visitor clicking play

Voice is scoped to the persona. When a conversation hands off to a workflow using a different persona, the voice switches automatically to that persona's voice. Requires ElevenLabs configured under Settings → Integrations.

Bot Personas — AI Model

Each persona can specify which LLM it uses. The model follows this priority chain:

1. The persona's AI Model (if set) — takes priority

2. The site default set by your Velaro admin

3. The workflow node's fallback model

Available models include: GPT-4o, GPT-4o mini, GPT-5, GPT-5 mini, Claude Haiku, Claude Sonnet, Gemini Flash, Gemini Pro, and Velaro Default (platform-managed, always current).

Use "Velaro Default" if you want Velaro to manage model upgrades automatically. Use a specific model if you need deterministic, reproducible behavior or cost control.

What is a bot in Velaro?

A bot is an AI-powered conversation handler that responds to visitors automatically. Bots use a large language model (LLM) with a custom system prompt and optional knowledge sources. They can handle entire conversations or work alongside human agents.

Creating a bot

1. Go to AI Bots in the sidebar.

2. Click New Bot.

3. Give the bot a name (used internally and optionally shown to visitors).

4. Write a System Prompt — this tells the bot its role, tone, and what to do.

5. Optionally connect a Knowledge Base (if your plan includes it).

6. Click Save.

System prompt best practices

The system prompt is the most important setting. Write it as instructions to the AI.

Good example:

You are a customer support agent for Acme Corp. You help customers with order status,
returns, and product questions. Always be friendly and concise. If you cannot answer,
offer to connect the visitor to a human agent.

Customer name: <<firstName>>
Account tier: <<accountTier>>

Bad example:

Answer questions.

Using variables in system prompts

Variables resolve in the system prompt at runtime. This lets the bot know who it's talking to.

• <<firstName>> — visitor's first name
• <<plan>> — account plan (custom field)
• <<accountId>> — any custom field you've defined

Bot channels

Bots work on all channels: web chat, SMS, WhatsApp, Facebook, Instagram. Assign a bot to a workflow or deployment to activate it on a channel.

Connecting a bot to a channel

The recommended way is via a Workflow:

1. Create or open a workflow.

2. Add an AI / Bot node.

3. Select the bot from the dropdown.

4. The bot handles conversation from that node onward.

AI Orchestration (multi-agent)

If your plan includes AI Orchestration, you can have multiple AI bots hand off to each other — all within the same conversation, no human involved.

How it works:

• A primary/triage bot greets the visitor and figures out what they need
• It delegates to a specialist bot that has the right tools for that topic (e.g. an Orders bot with Shopify access, a Billing bot with NetSuite access)
• The specialist handles the request, then calls return_to_previous to hand control back to the primary bot

To set it up — no coding, no tool names to type:

1. Create your specialist bots under Bots (each with its own system prompt)

2. Open your primary bot's workflow and click the AI node

3. In the Specialist Agents panel, use the dropdown to select a specialist

4. Type a short scope hint: "orders, shipping, returns, tracking"

5. Check which tool servers (BigCommerce, NetSuite, etc.) this specialist gets

6. Repeat for each specialist

That's it. You never write delegate_to_anything — the system injects those tools automatically based on what you selected. The primary AI reads the scope hint to decide when to hand off.

What happens automatically (behind the scenes):

• Each specialist you added gets a delegate_to_[their name] tool injected into the primary bot
• The specialist automatically gets a return_to_previous tool so it can hand back when done
• The system tracks a call stack so specialists can always return to whoever called them
• Maximum 5 levels deep (rarely need more than 2)

The primary bot should be lightweight: No tools on the triage/primary bot — just a short prompt that says what it should classify and delegate. All tools (Shopify, NetSuite, BigCommerce) live on the specialist bots.

For full design patterns and demo-ready examples, see the Bot Design Patterns guide.

Knowledge Base integration

If you have a Knowledge Base set up, the bot can search it to answer questions grounded in your documentation. This reduces hallucinations and keeps answers accurate.

To connect KB to a bot: go to the bot settings and enable Knowledge Base under the Sources section.

Escalation to human agents

The bot can escalate to a human agent in two ways:

Automatic (built-in tool): The bot has a transfer_to_agent tool. When the visitor asks for a human, says "I need to speak to someone", or the bot cannot resolve the issue, it calls this tool. The bot always includes a reason and a summary note so the agent knows the context immediately when they open the conversation.

Workflow fallback: You can also add a Route to Team node after the AI node in your workflow as a guaranteed fallback path.

What happens when no agents are available:

• The system checks team availability before routing
• If the team is offline, out of schedule, or all agents are at capacity, the conversation is treated as a missed chat
• If you have an "on unavailable" routing rule configured for that team, that workflow runs automatically (e.g. an offline survey, a "leave a message" flow, or a different bot)
• If no rule is configured, the conversation is marked missed and an SMS alert fires (if enabled)

To control the "no agents" experience: Go to Rules > Routing Rules and add a "Use workflow on unavailable" rule for the team. Set it to a workflow that handles offline visitors gracefully.

Bot testing

1. Open the bot in the editor.

2. Click Test Bot to open a live preview.

3. Type messages and see how the bot responds.

4. Adjust the system prompt and re-test until the behavior is right.

MCP Tools (integrations)

If your plan includes MCP (Model Context Protocol), bots can call external tools during a conversation:

• Look up order status from Shopify
• Check account details in HubSpot or Salesforce
• Create a support ticket in Zendesk
• Read a customer record from NetSuite

Each tool is configured as an MCP server under Settings > MCP Servers. The bot automatically decides when to call the right tool.

Scripted (deterministic) workflows — no AI needed

If you need 100% predictable, word-for-word responses — for example in regulated industries like insurance, healthcare, or finance — you do NOT need to use an AI bot node at all.

Build a workflow using only:

• Send Message nodes (exact text you control)
• Ask Question nodes (capture visitor input)
• Condition nodes (branch on their answer)
• Route to Team nodes (hand off to a human)

This gives you a fully deterministic script where every word is authored by you. No LLM is invoked. The workflow IS the script.

When to use AI instead: Use an AI bot node when you want the bot to handle open-ended questions, understand natural language variations, or search your knowledge base. The AI node is optional — add it only where you want flexible, conversational responses.

Hybrid pattern: Run a scripted section first (collect name, email, issue type) then hand off to an AI node for the open-ended conversation. This gives you structured data capture with conversational flexibility.

Knowledge Overrides — Guarantee Exact Answers

Knowledge Overrides let you define authoritative facts that are always injected into the AI prompt before any vector search results. Use them to guarantee the AI uses your exact policy text for brand-critical topics — returns, warranties, pricing, holiday hours, limited-time promotions.

Go to Settings → Bots → AI Chatbot → Knowledge Overrides to manage them.

Each override has:

• Topic — short label (e.g. "Return Policy", "Holiday Hours")
• Fact — the exact text injected verbatim into the AI system prompt
• Trigger Keywords (optional) — comma-separated words; the fact is only injected when the customer's message contains one of these words. Leave blank to inject for every conversation.
• Priority — lower number = higher priority (injected first). Default: 0.
• Expires On (optional) — auto-deactivates after this date. Useful for seasonal promotions.

When the customer sends a message, active overrides matching the query keywords are prepended to the AI context as:

> "CURRENT FACTS — These override all other knowledge. Always use these exact facts: • Return Policy: We accept returns within 30 days with original receipt."

Best practices:

• Keep each fact under 300 characters — every override adds to the AI prompt budget.
• Use trigger keywords for topic-specific facts (e.g. "warranty, defective, broken").
• Leave keywords blank only for genuinely universal facts (e.g. your company name, phone number).
• Set an expiry date for limited-time promotions so they self-cleanup.
• Use Priority 0 for the most important facts (injected first); use higher numbers for supplementary context.

Subscription requirements

• Basic bot (LLM responses): requires AI/Bot Automation on your plan
• Knowledge Base integration: requires Knowledge Base on your plan
• Knowledge Overrides: requires AI/Bot Automation on your plan
• MCP tool calls: requires MCP on your plan
• Multi-agent orchestration: requires AI Orchestration on your plan

Contact support@velaro.com to add these to your subscription.