Docs AI Configuration

AI Configuration

Set up Claude or OpenAI as your AI engine, manage API keys, configure auto-reply behavior, and customize AI features.

Overview

Helpon supports dual AI engines — Claude by Anthropic and OpenAI — giving you the flexibility to choose the provider that best fits your use case, budget, and performance needs. You can configure one as your primary engine and the other as a fallback.

All AI settings are located in Helpon → Settings → AI in your WordPress admin dashboard. From this panel you can manage API keys, select models, configure auto-reply behavior, set up the AI copilot for agents, enable image analysis, and customize the system prompt that shapes your AI assistant's personality.

ℹ️
Dual Engine Architecture

You can configure both Claude and OpenAI simultaneously. Helpon uses your primary provider for all AI replies and automatically switches to the fallback provider if the primary is unavailable or returns an error.

AI Provider Setup

Helpon requires an API key from at least one AI provider. Follow the steps below for your preferred provider.

Claude (Anthropic)

1

Create an Anthropic Account

Visit console.anthropic.com and sign up for an account. You will need to add a payment method and purchase API credits before generating a key.

2

Generate an API Key

In the Anthropic Console, navigate to API Keys and click Create Key. Give it a descriptive name like "Helpon WordPress" and copy the key immediately — it won't be shown again.

3

Enter the Key in Helpon

Go to Helpon → Settings → AI, select Claude (Anthropic) as your provider, and paste your API key. Click Validate Key to verify the connection.

Primary AI Provider
Claude (Anthropic) ▾
Claude API Key *
sk-ant-api03-••••••••••••••••
✓ Key validated successfully
Default Model
Claude 3.5 Sonnet ▾

OpenAI

1

Create an OpenAI Account

Visit platform.openai.com and sign up. Add a payment method under Billing to enable API access.

2

Generate an API Key

Go to API Keys in the OpenAI dashboard and click Create new secret key. Name it "Helpon" and copy the key. OpenAI keys start with sk-.

3

Enter the Key in Helpon

In Helpon → Settings → AI, select OpenAI as your provider and paste the API key. Click Validate Key to confirm the connection.

Primary AI Provider
OpenAI ▾
OpenAI API Key *
sk-••••••••••••••••••••••••
✓ Key validated successfully
Default Model
GPT-4o ▾

API Key Management

Helpon takes security seriously when handling your API credentials. Here is how keys are managed within the plugin.

Encryption & Storage

All API keys are encrypted using AES-256-CBC encryption before being stored in the WordPress database. Keys are never stored in plain text, and they are decrypted only at the moment an API call is made. The encryption key is derived from your WordPress AUTH_KEY salt.

Key Rotation

To rotate an API key, generate a new key from your provider's console, paste it into Helpon, and click Validate Key. Once validated, the old key is replaced immediately. We recommend rotating keys every 90 days as a security best practice.

Testing the Connection

Use the Test Connection button in Helpon → Settings → AI to send a test prompt to the configured provider. A successful test returns the model name, response time, and token usage. A failed test shows the specific error code and message from the provider.

Fallback Provider

You can configure a secondary provider that activates automatically when the primary provider fails. For example, set Claude as your primary and OpenAI as your fallback. If Claude returns an error or times out, Helpon seamlessly routes the request to OpenAI so your visitors never see a failed response.

⚠️
API Key Security

Never share your API keys publicly or commit them to version control. If you suspect a key has been compromised, revoke it immediately from the provider's console and generate a new one. Helpon masks keys in the admin UI, showing only the first and last 4 characters.

Model Selection

Helpon supports multiple AI models from both providers. Choose a model based on your balance of cost, speed, and response quality.

Model Provider Speed Quality Cost per 1K Tokens
Claude 3.5 Sonnet Anthropic Fast Excellent $0.003 / $0.015
Claude 3 Haiku Anthropic Very Fast Good $0.00025 / $0.00125
GPT-4o OpenAI Fast Excellent $0.005 / $0.015
GPT-4o-mini OpenAI Very Fast Good $0.00015 / $0.0006
GPT-3.5-turbo OpenAI Very Fast Moderate $0.0005 / $0.0015
Choosing the Right Model

For most healthcare use cases, Claude 3.5 Sonnet or GPT-4o offer the best balance of accuracy and speed. If you handle high chat volumes and need to minimize cost, consider Claude 3 Haiku or GPT-4o-mini for general inquiries and reserve the larger models for complex conversations.

To change your model, navigate to Helpon → Settings → AI → Model and select from the dropdown. The change takes effect immediately for all new conversations. Existing conversations continue using the model they started with.

Credit System

Every AI-generated reply in Helpon consumes credits from your account balance. Credits allow you to control and monitor your AI usage costs independently of the underlying API billing.

How Credits Work

  • 1 credit = 1 AI reply sent to a visitor, regardless of message length
  • Credits are deducted at the time the AI response is generated
  • Agent-written replies (non-AI) do not consume credits
  • AI Copilot suggestions consume 0.5 credits per suggestion
  • Image analysis consumes 2 credits per image processed

Credit Packs

Pack Credits Price Per Credit
Starter 500 $9 $0.018
Growth 2,000 $29 $0.0145
Business 10,000 $99 $0.0099
Enterprise 50,000 $399 $0.008

Monitoring & Alerts

Your current credit balance is always visible in the Helpon → Dashboard widget and in the AI settings panel. Helpon provides configurable alerts when credits run low.

Credit Balance
4,328 credits remaining
Monthly Usage
1,672 credits used this month
Low Credit Alert Threshold
500 credits
You will receive an email notification when your balance drops below this number
Auto Top-Up
Enabled
Auto Top-Up Pack
Growth (2,000 credits / $29) ▾
💡
Auto Top-Up

Enable auto top-up to ensure your AI assistant never goes offline due to exhausted credits. When enabled, Helpon automatically purchases the selected credit pack when your balance falls below the alert threshold.

Auto-Reply Settings

Auto-reply controls how and when the AI responds to visitor messages without human agent intervention. Fine-tune these settings in Helpon → Settings → AI → Auto-Reply.

AI Auto-Reply
Enabled
Response Delay (ms)
1500
Simulates typing time (500 - 5000 ms)
Confidence Threshold
70
Minimum AI confidence (0 - 100) to auto-send
Fallback Message
I'm not sure about that. Let me connect you with a team member who can help.
Max Turns Before Escalation
5
Automatically escalate to a human agent after this many AI turns
Sentiment-Based Escalation
Enabled

Setting Details

Setting Description Default
AI Auto-Reply Master toggle for AI-generated automatic responses Enabled
Response Delay Adds a realistic typing delay before the AI reply appears (in milliseconds) 1500 ms
Confidence Threshold AI replies scoring below this value are held for agent review instead of being auto-sent 70
Fallback Message Displayed when the AI cannot generate a confident response Pre-configured
Max Turns Before Escalation Number of consecutive AI replies before the conversation is routed to a live agent 5
Sentiment-Based Escalation Automatically detects negative visitor sentiment (frustration, anger) and escalates to a human agent Enabled
ℹ️
Confidence Threshold

A threshold of 70 works well for most setups. Lower it to 50 if you want the AI to handle more queries autonomously, or raise it to 85 if you prefer tighter human oversight. Replies below the threshold appear in the agent console as "Pending Review" instead of being sent automatically.

Image Analysis

Helpon can analyze images uploaded by visitors during a chat conversation. When enabled, the AI processes the image and provides contextual responses based on the visual content.

Supported Formats

  • JPEG / JPG — Photos, scanned documents
  • PNG — Screenshots, diagrams, ID cards
  • GIF — Animated images (first frame analyzed)
  • WebP — Modern compressed images

Use Cases

  • Medical imaging review — Patients can share dental X-rays, skin photos, or wound images for preliminary triage
  • ID verification — Analyze uploaded identification documents for patient onboarding
  • Insurance card reading — Extract information from insurance cards shared via chat
  • Document review — Parse prescriptions, referral letters, or lab results

Configuration

Enable image analysis in Helpon → Settings → AI → Image Analysis. The toggle activates image processing for all chat conversations. Images are sent to the configured AI provider's vision API and are not stored on your server after analysis unless you enable the image archive option.

⚠️
HIPAA & Privacy Considerations

If your practice handles protected health information (PHI), ensure that image analysis complies with your HIPAA policies. Images sent to AI providers are processed by third-party APIs. Review your Business Associate Agreement (BAA) with Anthropic or OpenAI before enabling image analysis for medical images. Consider disabling this feature or limiting it to non-PHI use cases if a BAA is not in place.

AI Copilot

The AI Copilot assists your human agents by generating suggested replies in real time. Instead of the AI responding directly to visitors, it provides agents with draft responses that they can review, edit, and send with a single click.

How It Works

1

Visitor Sends a Message

When a visitor sends a message in a conversation assigned to a human agent, the Copilot analyzes the message, conversation history, and your knowledge base.

2

AI Generates a Suggestion

The Copilot panel (docked to the right of the chat console) displays a suggested reply along with a confidence indicator. A green indicator means high confidence, yellow means moderate, and red means low confidence.

3

Agent Reviews & Sends

The agent can click Apply to insert the suggestion into the reply box, edit it as needed, and send. Alternatively, the agent can dismiss the suggestion and write their own reply.

Chat Console
Visitor: What insurance plans do you accept? I have Blue Cross.
Agent is typing...
AI Copilot 92% confident
We accept Blue Cross Blue Shield along with most major insurance plans including Aetna, Cigna, United Healthcare, and Delta Dental. Would you like me to verify your specific plan's coverage?
Apply Dismiss

Copilot Settings

Setting Description Default
Enable Copilot Toggle AI suggestions for agent-handled conversations Enabled
Panel Position Where the Copilot panel appears: Right sidebar or Inline (below chat) Right Sidebar
Auto-Suggest Automatically generate suggestions for every incoming message Enabled
Knowledge Base Priority Copilot prioritizes answers from your knowledge base over general AI knowledge Enabled
Tone Matching Copilot adapts its suggestion tone to match the agent's writing style Enabled

System Prompt

The system prompt defines your AI assistant's personality, knowledge boundaries, and behavioral rules. It is sent as the first message in every AI conversation and shapes every response the AI generates.

Editing the System Prompt

Navigate to Helpon → Settings → AI → System Prompt. The editor supports plain text with variable placeholders. The maximum length is 4,000 characters.

Available Variables

Variable Description Example Output
{business_name} Your business name from the General Settings Bright Smile Dental
{agent_name} The name of the assigned agent (or "AI Assistant" if no agent) Dr. Sarah
{visitor_name} The visitor's name if known, otherwise "there" John

Example System Prompt

System Prompt
You are a friendly and professional virtual assistant for {business_name}.

Your role:
- Answer questions about our services, hours, pricing, and policies
- Help visitors schedule appointments
- Provide general health information (but never diagnose conditions)
- Escalate complex medical questions to a human team member

Tone: Warm, professional, and reassuring. Use simple language.

Rules:
- Never provide specific medical diagnoses or treatment plans
- Always recommend consulting with a healthcare professional for medical concerns
- If you are unsure about something, say so honestly and offer to connect the visitor with {agent_name}
- Keep responses concise (under 150 words unless the visitor asks for detail)
- Greet returning visitors by name: "Welcome back, {visitor_name}!"

Business hours: Monday-Friday 8am-6pm, Saturday 9am-2pm
Emergency: Direct visitors to call 911 or visit the nearest emergency room

Prompt Templates

Helpon ships with pre-built system prompt templates tailored to common industries. Select a template from the dropdown in the system prompt editor to use as a starting point:

  • Healthcare (General) — Balanced template for clinics, hospitals, and health practices
  • Dental Practice — Focused on dental services, insurance, and appointment scheduling
  • Mental Health — Sensitive language, crisis resource awareness, appointment facilitation
  • E-Commerce — Product inquiries, order tracking, returns, and general support
  • Professional Services — Consultations, pricing inquiries, and scheduling
Prompt Best Practices

Be specific about what the AI should and should not do. Include your business hours, escalation rules, and tone guidelines. Test your prompt by having a conversation with the AI from a visitor's perspective and refining based on the responses you receive.

Troubleshooting

Common AI-related issues and how to resolve them:

API Key Errors

  • 401 Unauthorized — Your API key is invalid or has been revoked. Generate a new key from the provider's console and update it in Helpon.
  • 403 Forbidden — Your API account may not have billing enabled or the key lacks required permissions. Check your provider's billing and API access settings.
  • Key validation fails — Ensure there are no extra spaces or line breaks when pasting the key. Try generating a fresh key.

Rate Limit Errors

  • 429 Too Many Requests — You've exceeded the provider's rate limit. Helpon automatically retries with exponential backoff. If this persists, consider upgrading your API tier with the provider or enabling a fallback provider.
  • Throttling during peak hours — Enable the fallback provider so Helpon can route requests to the secondary engine when the primary is rate-limited.

Model Unavailable

  • Model not found — The selected model may have been deprecated or renamed by the provider. Check the provider's model documentation and update your selection in Helpon → Settings → AI → Model.
  • Temporary outage — AI providers occasionally experience outages. Helpon switches to the fallback provider automatically if one is configured.

Credits Exhausted

  • AI replies stop working — Check your credit balance in Helpon → Dashboard. Purchase a credit pack or enable auto top-up to prevent future interruptions.
  • Credits depleting too fast — Review your usage in Helpon → Analytics → AI Usage. Consider using a less expensive model for routine queries or increasing the confidence threshold to reduce unnecessary AI replies.

Slow Responses

  • High response latency — Larger models (GPT-4o, Claude 3.5 Sonnet) may take 2-5 seconds. Switch to a faster model (Claude 3 Haiku, GPT-4o-mini) if speed is critical.
  • Long system prompt — Reduce your system prompt length. Every token in the system prompt adds latency and cost to each request.
  • Server-side delays — Ensure your WordPress hosting has adequate PHP memory (256MB minimum) and the max_execution_time is set to at least 30 seconds.
ℹ️
AI Logs

For detailed debugging, enable AI logging in Helpon → Settings → AI → Advanced → Enable Logs. This records every API request and response, including token counts, latency, and error details. Logs are retained for 30 days and can be exported as CSV.