Quick guide to using the ChatPlace AI Agent
1. General principle of how the AI Agent works
The ChatPlace AI Agent is an intelligent system that functions like a full-fledged human manager: responds to clients, analyzes messages, reads multimedia, remembers information, passes requests to operators, works 24/7.
The agent is not a single model — it's a complex system combining multiple neural networks:
Main model: Claude Sonnet 4.5
Backup / auxiliary models: ChatGPT-5, ChatGPT-4o
Specialized models: voice recognition, image analysis, video and text interpretation
Internal fine-tuning algorithms
This architecture ensures:
Stable performance under any load
Deep understanding of complex content
Long-term context retention
Fast and natural human-like responses
Ability to operate like a real employee
If the main model is overloaded (approx. 1% of cases), the agent automatically switches to ChatGPT-5 — users won’t notice the difference.
2. Why the AI package costs $25 on the Pro Plan and why AI requests are billed separately
The cost includes not just access to a language model, but support for the entire multimodal AI system.
The AI Agent:
Works 24/7
Processes text, photos, videos, voice messages, and video notes
Reads client profiles (posts, bio, content)
Analyzes entire message history
Auto-learns from context
Sends requests to operators
Can save variables into your CRM
Uses multiple models simultaneously
We collaborated with engineers from Anthropic (creators of Claude) to create one of the smartest communication agents on the market.
Every AI response is a computational operation (tokens). Token costs are high — we cover the majority of this cost ourselves. That’s why:
2000 AI requests are included in the package
each additional 1000 requests costs $15. On Premium — $10
At the moment, 1 AI response costs around $0.0165, one of the lowest prices worldwide for multimodal systems of this level. The agent continues to improve — more capabilities and intelligence are added regularly.
We built the system so it requires no technical skills — setup in a few clicks, but inside is a massive computational infrastructure.
3. Supported languages & voice recognition
The agent understands all major world languages, including:
Text
Voice messages
Video notes with speech
Vixed messages
Translations
It automatically replies in the same language the client uses. If needed, you can explicitly specify in the rules:
Which language to use
Communication style
Whether to ignore the client’s language and always respond in a given one
4. How to write an effective Communication rule (Prompt)
The Communication rule determines exactly how the AI Agent behaves. Its purpose is to ensure the agent:
Follows your business processes
Avoids mistakes
Operates strictly within your scenarios
A good prompt must include:
Communication style (friendly/formal, “you” vs “thou”)
How the agent introduces itself
Which language to use
How to handle pricing questions
How to process order requests
Which data to save into variables
Which information to send to operators
Forbidden topics
Sales behavior
Clarifying question logic
Handling spam
When a dialogue is considered “complete”
The more detailed — the better the results.
If you cannot write a proper prompt, write your rules in free form and ask any AI model to convert it into a well-structured prompt.
5. Sending data to an operator
The agent can automatically forward client data to an operator based on your rules. It can send:
Name
Phone
Email
Profile link
Conversation history
Inquiry text
Any extracted information
Any variables saved during the dialogue
The agent sends data only in the cases that you specified. For example:
If the client wants a consultation → send a request.
If they ask for a callback → send their number.
If they want to buy → notify the sales manager.
If it’s a VIP client → always notify the operator.
All logic is customizable.
6. Why the Agent may stop responding
There are several logical reasons:
The agent detected spam or meaningless messages — If a client sends random text or repeated messages, the agent goes into “protection mode” to avoid wasting tokens.
The dialogue is logically complete — If a client writes: “ok”, “got it”, “thanks” — the agent won’t reply.
The agent completed its task and forwarded data to an operator.
The agent avoids unnecessary token usage.
Rare technical delays (<1%). Due to temporary third-party API load.
👉 Important: The agent always resumes the conversation if the client sends meaningful input again.
7. Forbidden topics & censorship
There is no external censorship. But you can configure your own forbidden topics:
Politics
Medicine
Legal advice
Cryptocurrency
Adult topics
In such cases, the agent replies neutrally:
“Unfortunately, I can't discuss this topic, but I can help with something else.”
8. Where to store prices, policies, and critical information
Always store important data in the Communication Rule, NOT in the Knowledge Base. The reason:
The KB contains many topics; critical data may get lost
Updating them requires manual searching
The agent may rephrase or distort numeric values
The KB is not designed for sensitive information
Do NOT store prices or policies in:
the knowledge base
analyzed Instagram posts
large PDF files
The Communication rule overrides all other sources and remains precise.
9. Can the Agent trigger automations?
The agent does not directly start automations (to avoid conflicts). However, it can give the user a trigger phrase, for example:
“If you want to proceed with the order, type: PAYMENT.”
When the client sends the trigger — the automation starts.
10. Client follow-up (re-engagement)
The agent can politely follow up with clients:
Ask if they managed to resolve their issue
Remind them about an earlier conversation
Return to a topic
Guide the dialogue to completion
It only does this when it detects real interest — it never spams uninterested users.
11. Working with multimedia
The agent processes:
Voice notes
Video notes
Photos
Screenshots
Uploaded PDFs
Stories
Comments
It reads text on images, understands context, analyzes stories, and responds meaningfully.
12. How the agent learns
It uses every available source:
Knowledge base (account scan, files, text)
Communication rule (primary logic)
Conversation fine-tuning (if enabled)
Client messages
CRM variables
Dialogue history
Instagram profile data
Reactions, comments, stories
Behavioral patterns
It adapts to your communication style and remembers important topics automatically.
13. Updating the knowledge base
There are three ways:
1. Instagram/TikTok account scan — Reads:
posts
profile bio
writing style
existing conversations
2. File upload — Supports:
PDF
DOCX
XLS
Up to 30 MB each.
3. Manual text input — The KB updates instantly, and the agent begins using new information immediately.
14. Difference between Communication rule and Knowledge base
Knowledge base = factual reference
Products
Delivery terms
Service descriptions
Specifications
Communication rule = the brain
Style
Behavior
Decision logic
Variable saving
Sales scripts
Operator handoff rules
If the two conflict — the Communication Rule always wins.
15. Frequently asked questions
1. Does the agent reply to comments?
Yes.
2. Does the agent reply to stories?
Yes.
3. Can the agent see story content?
Yes. (All three options are configurable in settings.)
4. Can it write in Direct after a comment?
Yes — one message (Instagram limitation).
If the user replies to that message, the agent continues normally.
5. Does it conflict with automations?
No — automation triggers have priority.
You can also configure ignore-windows for automations.
6. Can it work alongside a human operator?
Yes — ideally set rules for when the agent freezes if an operator responds.
16. Saving client data & creating variables
This is one of the strongest features. The agent sees:
Client name
Username
Profile link
Tags
Variables
Message history
Instagram profile data
Subscription status
Previous conversations
The agent can:
Save values into variables
Create new variables automatically
Extract data based on your rules
Detect key phrases (“my name is…”, “my number is…”)
Store values for further processing
Reuse variables in the dialogue
Examples:
“If the client gives a name — save to client_name.”
“If the client writes a phone number — save to phone.”
“If the client mentions a city — create client_city.”
“If the client is subscribed — set is_follower = yes.”
“If the client wants a consultation — send operator name, number, profile, and chat history.”
17. Best practices
Check the Unanswered section during the first 3 days
Regularly update the Knowledge Base
Never store critical data in the KB — only in the Communication Rule
Test every change
Use follow-ups when needed
Enable automatic data transfer to operators