AI Assistants & Knowledge Base

AI Assistants are the intelligent backbone of JackAI Nexus. They automatically respond to customer messages, answer questions, and help qualify leads — all without human intervention. You can create multiple assistants, each with its own personality and capabilities, and assign them to different customer communication channels.

---

What Are AI Assistants?

An AI Assistant is an AI-powered bot that handles customer conversations on your behalf. When a customer sends a message on WhatsApp, Messenger, Instagram, or your web chat, the assistant reads the message, searches your knowledge base for relevant information, and generates a helpful response in seconds.

Key capabilities:

• Answers customer questions automatically
• Qualifies leads based on their responses
• Collects customer information (phone numbers, email, preferences)
• Escalates conversations to human agents when needed
• Learns from your knowledge base and business documents
• Available 24/7 across all channels

Each organization can create multiple assistants, giving you flexibility to customize responses for different products, teams, or channels.

---

AI Provider Types

JackAI Nexus supports three types of AI providers. Choose the one that fits your needs.

Provider Comparison

Not sure which one to choose? Start with Built-in (OpenAI) — it works for most businesses and is the easiest to set up.

---

Built-in (OpenAI)

The built-in provider uses OpenAI's powerful language models (like GPT-4o-mini) to power your assistant. You provide your own OpenAI API key (your secret credential for connecting to OpenAI), giving you full control and transparency of costs.

Configuration options:

• Model: Choose which OpenAI model to use (e.g., GPT-4o-mini for fast, cost-effective responses, or GPT-4 Turbo for more complex reasoning).
• Temperature: Controls how creative or deterministic the AI is. Low values (0.1–0.3) produce consistent, factual answers. High values (0.7–1.0) are more creative and varied.
• Max tokens: Limits the length of each response (higher token count = longer responses but higher cost).
• Custom system prompt: Write instructions that define the assistant's personality, tone, expertise, and behavior.

Who it's for: Most organizations. You control costs and have full transparency. Requires an OpenAI API key (you'll need an OpenAI account and billing setup).

OpenAI Prompts API

This provider uses OpenAI's Prompts API, which stores and manages your system prompts on OpenAI's side. It's ideal if you want centralized prompt management and advanced features like prompt versioning and testing.

Features:

• Store and version-control your prompts in OpenAI's system
• Advanced prompt search and discovery
• Better for teams managing many prompts

Who it's for: Organizations with mature AI workflows or those managing multiple prompts across teams.

n8n Workflow

Instead of using OpenAI directly, you can delegate AI processing to a custom n8n workflow. This is powerful for organizations that need advanced logic, CRM integration, or custom business processes.

Great for:

• Integrating with your existing CRM or business tools
• Complex multi-step workflows (e.g., check inventory, query database, then respond)
• Custom business logic that OpenAI alone can't handle
• Organizations already using n8n

How it works: When a customer sends a message, Nexus calls your n8n webhook with the conversation data. Your workflow processes it (using your own AI provider, databases, or custom logic) and returns the response.

---

Creating an AI Assistant

To set up a new AI assistant:

1. Go to AI Assistants (in the main navigation)
2. Click "Create Assistant"
3. Fill in the details:
   • Name: A descriptive name (e.g., "Sales Support," "Product FAQ Bot")
   • Provider: Choose Built-in (OpenAI), OpenAI Prompts API, or n8n Workflow
   • Model settings: Configure temperature, max tokens, and other provider-specific options
4. Write your system prompt (see below)
5. Click "Create"

The assistant is now ready to be assigned to channels.

---

System Prompt: The Brain of Your Assistant

The system prompt is the instruction set that tells your AI assistant how to think and behave. It's the most important part of your assistant configuration.

A good system prompt includes:

• Your business context: "You are a customer support specialist for an e-commerce fashion retailer"
• Tone of voice: "Be friendly, professional, and concise"
• What to answer: "Provide answers about product availability, shipping, returns, and sizing"
• What NOT to answer: "Never provide technical support; direct those to our IT team"
• When to escalate: "If a customer is angry or mentions a refund dispute, pause AI and notify an agent"
• Key facts: "Standard shipping takes 5-7 business days. Returns are free within 30 days"

Organization-level default prompt: Your organization can set a default system prompt in Settings. All new assistants inherit this prompt, but you can override it per assistant if needed.

Tips for writing better prompts:

• Be specific about your business, products, and values
• Clearly define boundaries (what the AI should NOT do)
• Include tone instructions (friendly, formal, technical, casual, etc.)
• Add key facts that customers frequently ask about
• Tell the AI when and how to escalate to humans

System Prompt Template

Use this template as a starting point for your own system prompts. Fill in the bracketed sections with your business details:

You are [ROLE] for [COMPANY NAME].

Your tone is [TONE - e.g., friendly and professional, cheerful and casual, technical and direct].

You help customers with: [LIST MAIN TOPICS - e.g., product recommendations, order tracking, technical support, pricing questions].

You do NOT: [LIST BOUNDARIES - e.g., share pricing details, make promises about refunds, schedule appointments, provide technical advice outside our products].

If a customer is upset or needs human help: [ESCALATION RULE - e.g., apologize sincerely, acknowledge their concern, and let them know that an agent will follow up within 2 hours].

Key information about our business: [ADD KEY FACTS - e.g., we're open Mon–Fri 9am–6pm EST, standard shipping takes 5-7 days, returns are free within 30 days].

Example (filled in):

You are a customer support specialist for TechFlow, a software company.

Your tone is friendly, professional, and helpful.

You help customers with: product setup, troubleshooting, billing questions, and feature requests.

You do NOT: make promises about new features, override billing policies, or access customer data beyond what's needed to help.

If a customer is upset or needs human help: apologize, acknowledge their frustration, and let them know a senior agent will call them back within 4 hours.

Key information: We have a 30-day money-back guarantee. Our live support team is available Mon–Fri 9am–5pm PST.

---

Assigning Assistants to Channels

Once an assistant is created, you assign it to your customer communication channels (integrations). Each channel can have one AI assistant.

To assign an assistant:

1. Go to Integrations (in the main navigation)
2. Click on the channel you want to configure (e.g., WhatsApp, Messenger)
3. Under "AI Configuration," select the assistant from the dropdown
4. Configure per-channel AI settings (see below)
5. Save

---

Per-Channel AI Settings

Once an assistant is assigned to a channel, you can fine-tune how it behaves on that specific channel.

Enable/Disable AI

Toggle AI responses on or off for the channel. If disabled, only your team can reply to customers.

Voice Message Handling

Control how the AI handles voice messages (audio files):

• Transcribe: Automatically convert voice to text before the AI responds
• Disable: Ignore voice messages (team must reply manually)

Use transcription for WhatsApp and other voice-enabled channels to handle customers who prefer speaking.

Image & Document Processing

• Process: AI can see and analyze images and documents (e.g., invoices, product photos)
• Ignore: AI ignores images/documents, team must handle manually

Phone Collection

Enable AI to ask for the customer's phone number before starting the main conversation. This is useful on Messenger, Instagram, and Webchat, which don't capture phone numbers automatically. The AI will tactfully ask: "To better assist you, may I have your phone number?"

Why enable it:

• Build a richer customer database
• Enable WhatsApp follow-ups for leads from other channels
• Omni-channel deduplication (Nexus automatically merges leads with the same phone across channels)

Auto-Resume Timer

When your team replies to a conversation, the AI automatically pauses (so both AI and human aren't responding). After this timer expires, the AI auto-resumes.

Default: 60 minutes You can adjust: Set to 0 to disable auto-resume (team must manually resume AI)

Batch Processing

Batch processing waits a few seconds (configurable) for the customer to finish typing before the AI responds. This prevents the AI from replying to every individual message when a customer sends several messages in rapid succession, creating a more natural conversation flow.

Example:

• Customer sends: "Hi" → "Are you there?" → "I have a question about pricing"
• Without batching: AI responds to each message separately (3 separate replies)
• With batching: AI waits 2–3 seconds, collects all three messages, then responds once as a single conversation

Default: Enabled (2–3 second delay) When to adjust: If your customers tend to send very short messages, increase the delay. If they send long chunks of text, decrease it.

Rate Limiting (Message Speed Limits)

Prevent the AI from responding too frequently. Set a maximum number of AI responses per minute per conversation. This prevents the AI from sending too many messages too quickly.

Default: No limit Use case: In high-volume customer support, you might limit AI to 1 response per minute to avoid overwhelming your team with follow-ups. Example: If 10 customers message at once, rate limiting ensures the AI sends replies at a steady pace rather than flooding your agents all at once.

---

Knowledge Base: Train Your AI

Your knowledge base is a collection of documents that the AI reads to answer customer questions accurately. Think of it as your assistant's training material.

Adding Files to the Knowledge Base

1. Go to Knowledge Base (in the main navigation)
2. Click "Add File"
3. Upload a document (text, Markdown, or FAQ format)
4. Name the file: Use a clear, descriptive name (e.g., "Product Catalog - Winter 2026")
5. Add a description: Brief summary of the file's contents so the AI knows when to use it
6. Set as active: Toggle "Active" so the AI includes this file in its responses
7. Save

Supported File Types

• Text files: Plain text (.txt), paste content directly
• Markdown: Formatted documents with headers, lists, emphasis (.md)
• FAQ format: Question-and-answer pairs, ideal for structured knowledge

Best Practices for Knowledge Base Files

What to upload:

• Frequently asked questions and their answers
• Product catalog or service descriptions
• Pricing information
• Shipping, returns, and refund policies
• Troubleshooting guides
• Your company's values and mission
• Common objection handling (e.g., "Why are you more expensive than competitor X?")

File tips:

• Keep files focused on one topic (separate "Pricing FAQ" from "Shipping FAQ")
• Use clear headers and subheadings
• Include specific details (prices, timelines, phone numbers)
• Update files when information changes
• Inactive files are ignored by the AI, so deactivate outdated documents
• Don't include sensitive information (passwords, API keys, personal customer data)

---

MCP Servers: Connect External Tools

MCP (Model Context Protocol) lets you connect external tool servers to your AI assistant. Instead of hardcoding business logic, you build a separate MCP server that exposes your tools, knowledge, and behavior rules — and Nexus discovers and uses them automatically.

What MCP Provides

Connecting an MCP Server

1. Go to AI Assistants > select your assistant > MCP Servers tab
2. Click Add MCP Server
3. Enter the server URL, transport type, and auth token
4. Click Test Connection — Nexus discovers available tools, resources, and prompts
5. Click Refresh to update the discovered capabilities

Media Responses

MCP tools can send files to customers — not just text. When a tool returns a specially formatted JSON response with a media key, Nexus automatically downloads and delivers the file via WhatsApp, Messenger, or other channels.

Supported media types:

• Documents — PDFs, brochures, catalogs
• Images — Product photos, infographics, certificates
• Audio — Pre-recorded voice explanations

Learn More

For a complete guide on building your own MCP server with code examples, see Building an MCP Server.

---

How AI Processing Works

Here's the step-by-step flow when a customer sends a message:

1. Customer sends message on your channel (WhatsApp, Messenger, etc.)
2. Nexus receives it and loads the full conversation history
3. AI reads the message along with your knowledge base and system prompt
4. AI generates a response based on customer intent and your training materials
5. Response is sent back to the customer in seconds
6. Lead score updates automatically based on the conversation
7. Lead status may change if certain keywords or intents are detected
8. Conversation is summarized by AI (stored and visible in the conversation detail)

All of this happens automatically and continuously across all your channels.

---

Next Steps

• Managing Conversations — Learn how to monitor, reply to, and escalate conversations to human agents
• Lead Management — Understand lead profiles, scoring, and how to convert leads