Google Gemini Integration

1. Introduction: What is Google Gemini?

Google Gemini is a state-of-the-art "Large Language Model" (AI) that understands text, images, audio, and video. By connecting Gemini to Picky Assist, you provide your WhatsApp number with a "digital brain" that can:

• Understand Natural Language: Customers don't need to use specific keywords.

• Use Static Training Data: You can feed the AI your business FAQs, price lists, and rules so it always provides accurate, branded answers.

• Handle Multi-Modal Inputs: Customers can send photos or voice notes, and the AI can understand them to provide a text response.

2. How Gemini Billing Works

Google offers two primary tiers. Most users start with the Free Tier, which is ideal for testing and low-volume businesses.

The Free Tier (Standard)

• Cost: $0 USD.

• Privacy Note: Google may use your prompts and responses to improve their AI models. Do not use this tier if you handle highly sensitive or private customer data.

• Model Access: Includes models like Gemini 2.5 Flash and Gemini 3 Flash (Preview).

The Paid Tier (Pay-As-You-Go)

• Cost: You only pay for what you use (per "token" or word).

• Privacy: Strictly Private. Your data is not used to train Google's models.

• Typical Costs (Gemini 2.5 Flash):

  • Input: ~$0.30 per 1 million tokens (about 700k words).

  • Output: ~$2.50 per 1 million tokens.

3. Understanding Rate Limits

Rate limits are the "speed limits" for your bot. If you exceed these, the bot will stop responding until the limit resets.

Limit Type	Free Tier (Default)	Paid Tier (Tier 1)
Requests Per Minute (RPM)	15 Requests	300 Requests
Requests Per Day (RPD)	1,500 Requests	Unlimited
Tokens Per Minute (TPM)	1 Million Tokens	2 Million Tokens

Important: Limits apply per Project, not per API key. If you use the same project for multiple bots, they share these limits.

---

4. Training Data & "Context Caching"

A common question is: "Does Google charge me for my training data every time a customer sends a message?"

The Answer: Yes. Because the AI has no memory of previous messages, Picky Assist sends your Static Training Data (instructions) + the Customer's Question with every single request.

How to Save Costs: Context Caching

If your Training Data is very long (over 2,048 tokens), you can use Context Caching in the Paid Tier.

• How it works: Google "remembers" your training data for a set time.

• The Benefit: You get a 90% discount on the cost of your training data for every message sent after the first one.

• Storage Fee: You pay a small fee ($1.00 per 1M tokens/hour) to keep the data cached.

5. Best Model Comparison

Picky Assist supports several models. Here is which one you should choose:

Model Name	Tier	Capability	Best For...
Gemini 3 Flash (Preview)	Free/Paid	Default. Fast & Smart.	Most Users. High reasoning & speed.
Gemini 2.5 Flash	Free/Paid	Reliable & Cheap.	Standard chatbots and high-volume FAQs.
Gemini 2.5 Pro	Free/Paid	Deep Reasoning.	Hard math, coding, or complex logic.
Gemini 2.5 Flash-Lite	Free/Paid	Ultra-Fast.	Highest volume at the lowest possible cost.

Note: While Gemini 3 Flash can understand images and audio, the output sent back to your customers via Picky Assist will always be Text.

6. How the Billing Works in Detail

To understand Gemini billing, you must remember that AI has no memory. Every time a customer sends a message, Picky Assist must "remind" the AI of your training data.

The "Token" Basics

• Tokens are small chunks of text (1,000 tokens ≈ 750 words).

• Google charges for Input (what you send) and Output (what the AI replies).

Real-World WhatsApp Chat Scenario

Imagine you have a Flower Shop chatbot.

• Static Training Data: Your price list and shop rules (approx. 500 tokens).

• Customer Message: "Do you have red roses?" (approx. 10 tokens).

• AI Response: "Yes! Red roses are $20 a dozen." (approx. 15 tokens).

Scenario A: The Free Tier

• Cost: $0.

• Usage: This one interaction consumes 1 Request from your 1,500 daily limit.

• Token Usage: It uses 510 Input Tokens and 15 Output Tokens from your 1-million-per-minute limit.

Scenario B: The Paid Tier (Pay-As-You-Go)

Using Gemini 2.5 Flash (the most cost-effective model):

1. Input Cost: (500 Training + 10 Customer) = 510 tokens. At $0.10 per 1M tokens, this costs $0.000051.

2. Output Cost: 15 tokens. At $0.40 per 1M tokens, this costs $0.000006.

3. Total Cost for this Reply: $0.000057.

The Big Picture: At this rate, you could have roughly 17,500 messages for just $1.00 USD.

Usage Breakdown: The "Hidden" Data

When your WhatsApp bot goes from a few test messages to a real business tool, the way Google counts "usage" becomes very important.

Here is a breakdown of the costs and limits based on your specific setup: 2,000 characters of training data and an average of 20 messages per customer.

In AI, you are charged for the "Input" (instructions + question) and "Output" (answer).

• Your Training Data: 2,000 characters ≈ 500 Tokens.

• Average Input Message: ~50 Tokens.

• Average AI Response: ~100 Tokens.

For every 1 message a customer sends, Gemini processes:

• Input: 550 tokens (Training Data + Question)

• Output: 100 tokens (Response)

• Total per message: 650 tokens.

Price Estimate Comparison (Daily)

This table shows your estimated cost using the recommended Gemini 2.5 Flash model.

Number of Customers	Total Messages (20/cust)	Daily Cost (Free Tier)	Daily Cost (Paid Tier)
5 Customers	100 Messages	$0	~$0.04
50 Customers	1,000 Messages	$0	~$0.40
100 Customers	2,000 Messages	N/A (Limit hit)	~$0.80
500 Customers	10,000 Messages	N/A (Limit hit)	~$4.00

When to Upgrade to Paid: > 1. Volume: As soon as you expect more than 75 customers a day (which totals ~1,500 messages).

2. Privacy: If you want to ensure customer data is not used for AI training.

3. Speed: If you have more than 15 people messaging at the exact same minute.

Rate Limits & Concurrency

Rate limits are the "speed bumps" that prevent the system from crashing.

Free Tier: The "Starter" Limits

• Requests Per Day (RPD): 1,500 messages total. If 75 customers send 20 messages each, your bot stops working for the rest of the day.

• Requests Per Minute (RPM): 15 messages. If 16 people click "send" at the same time, the 16th person gets an error.

• Concurrency: Very low. This tier is for testing or very low-traffic shops.

Paid Tier (Tier 1): The "Business" Limits

• Requests Per Day (RPD): Unlimited. Your bot never sleeps.

• Requests Per Minute (RPM): 300 messages. This allows for high concurrency (many people chatting at the exact same second).

• Tokens Per Minute (TPM): 1 Million. With your 650-token message, you could handle ~1,500 messages per minute without hitting this limit.

How the "Concurrent" Limit Applies

"Concurrency" is how many people can use the bot at the same time.

• On Free Tier: If you have a marketing blast and 20 people message you at 10:00 AM, the first 15 get a reply instantly. The other 5 will fail and Picky Assist will show an "API Limit Reached" error.

• On Paid Tier: You can comfortably handle a "burst" of 300 messages every single minute. This is usually enough for most medium-sized businesses.

Getting Started

Go to the Picky Assist account

1. Login into your Picky Assist Admin account

2. Navigate to App and find the "Gemini" and click on Activate button

3. Click on Generate API Key and follow the below instructions

Generting Gemini API Token in Google AI Studio

To use Gemini, you must generate an API Key.

1. Visit Google AI Studio.

2. Login with your Google Account.

3. Click on "Get API Key" in the sidebar.

4. Copy this key and paste it into the Picky Assist App - Google Gemini

Possibilities with Picky Assist & Gemini

Picky Assist has achieved a deep integration with Gemini across its platform, enabling businesses to build custom AI models tailored to specific processes using tools like Mini App Builder and Macro Buttons. Here’s how this integration enhances various Picky Assist products:

Integration with Smart Replies as an AI Chatbot Picky Assist’s smart replies are now powered by Gemini, allowing for more natural, improved responses. The key advantage is the ability to train Gemini with your own segmented data, categorizing responses to better match customer needs.

Integration with Chatbot Builder When creating WhatsApp chatbots, you can now select the Gemini App to handle conversation details. This includes summarizing conversations, extracting valuable information in JSON format, and automatically updating Picky Assist CRM records. The integration streamlines processes like adding events to feeds or pulling key insights from interactions.

Create AI Agents & Handling Non Interactive Response : With the chatbot builder, you can create AI agents to handle non-interactive responses. For example, when you present a button but the user replies with a voice note, the Gemini integration can read the voice note, understand the context, and respond based on the training data. This ensures users don’t get stuck in a predefined flow.

Integration with Workflow & Funnels Gemini enhances Picky Assist’s workflow and funnels by allowing data-driven decisions. For example, when a CRM status changes, Gemini can suggest the next steps for a lead or deal, such as crafting custom sales scripts, personalized emails, or WhatsApp messages, automating the communication process.

Integration with Connectors Picky Assist’s Connectors allow seamless data exchange with third-party platforms. Gemini now acts as an intermediary step in these processes, such as automatically extracting relevant details from a job application email, matching it with job descriptions in the CRM, updating candidate records, and sending WhatsApp messages, all without manual intervention.

Integration with Mini App & Macro Button Mini App Builder and Macro Buttons streamline tasks by grouping actions. Businesses can create custom Gemini models for specific tasks, such as correcting grammar, training employees with business-specific processes, or summarizing content from uploaded PDFs, enhancing productivity.

Training with Your Own Data One of the standout features of Picky Assist’s integration is the ability to train Gemini with your business data at every execution level. This deep integration allows businesses to customize AI responses and actions, making automation more relevant and effective for specific needs.