An AI agent trained on your own knowledge
The FlexInbox AI agent is trained on your content; it answers customers with your own knowledge, in the customer's own language. You're always in control.
Knowledge base
Train the agent on your own content
Everything the agent knows comes from you. Feed your knowledge base from four different sources.
Crawl your website
Automatically crawls your entire website and pulls content — up to thousands of pages — into the knowledge base.
Upload files
Upload your PDF, Word (docx), csv, json, md, and txt files; their contents are added to the agent's knowledge.
Add Q&A
Enter frequently asked questions and their answers (FAQ) one by one so the agent uses them as a priority.
Individual URLs
Add specific pages one at a time by their address; pull only the content you want into the knowledge base.
Answers come from your data
With RAG, every answer comes from your data — the agent doesn't make things up; it relies on the source it finds.
Multiple knowledge bases
Create multiple knowledge bases and assign them to whichever agent you like.
Operating modes
You decide how automatic it runs
Run each agent in one of four modes; switch whenever you need to.
Off
The agent is disabled; agents write every reply themselves.
Suggestion
The AI suggests a reply; the agent sends it or skips it in a single click.
Semi-automatic
The agent prepares a reply; it's sent to the customer once approved.
Automatic
The AI replies to the customer on its own; you step in whenever needed.
You're in control
When the AI stays quiet and when it hands off
The agent never takes control away from you. In the situations below it automatically backs off or hands the conversation to a human.
Goes quiet when an agent replies
As soon as an agent types in a conversation, the AI automatically goes quiet in that chat.
Confidence threshold
When it drops below the confidence threshold you set, it hands the conversation to a human.
Handoff on keywords
On keywords like "agent" or "human," it hands the conversation to a person.
After-hours only
In after-hours-only mode, it steps in based on the working-hours window you define.
Human-like reply delay
With the human-like reply delay setting, you decide how quickly the agent responds.
Agent design
You design the agent from start to finish
The agent's personality, goal, and rules are entirely yours; there are no hidden background rules in the system.
Persona
You set the agent's style and tone.
Goal
You define what the agent aims to accomplish in the conversation.
Rules (system prompt)
You write the rules the agent follows directly; there are no hidden rules in the background.
Model and creativity
You choose the model and use the creativity (temperature) setting to decide how free the replies are.
Agent settings
Persona
A polite, concise, solution-focused sales advisor
Goal
Guide the customer to the right package and answer their question
Model
GPT-5
Mode
Automatic
Automatic follow-up
If the customer doesn't reply, the agent follows up
If the customer doesn't reply, the AI sends a follow-up message it writes itself. It can follow up over multiple steps; the number of steps and the maximum limit are in your control.
Follow-up timeline
First message sent
Follow-up 1 hour later
Final follow-up the next day
The number of steps and the maximum follow-up limit are in your control
Where it works
On every channel your customer is on
On direct messaging channels and every channel connected to Bitrix24 Open Channel; in the customer's language.
WhatsApp and Telegram
Works directly on WhatsApp and Telegram; talks to customers through your inbox.
Bitrix24 Open Channel
Connects to Bitrix24 Open Channel as a bot; also replies to Instagram, Messenger, and web chats linked to the Open Channel.
Multilingual & translation
It replies in whatever language the customer writes in. You can translate an incoming message or a draft in a single click.
Learning loop
Test, rate, and improve over time
Try the agent before going live; keep raising its performance by evaluating the answers it gives.
Test in the Playground
Test in the Playground before going live; see in advance how the agent answers real questions.
Rate and improve
Rate the AI's answers (feedback) and review the records to improve the agent over time.
Train your AI agent on your own knowledge
Train it on your content, choose the operating mode, and stay in control the whole time. Get started free.