Not a separate chat on the side, but a block on your bot's canvas — right next to buttons, broadcasts and conditions.
Role, rules, tone of voice. You can insert contact variables into the prompt, so the agent knows who it is talking to from the start.
Any of the connected AI models, plus toggles for tools: knowledge base, web search, tags, variables.
The customer keeps moving through the flow, and once the reply is ready, they return to the block and receive it.
History is stored separately for each customer and each block. You set the depth, the retention period and whether to reset before starting.
The agent decides which branch to take next — from the ones you have set up.
If the model is unavailable or the balance runs out, the customer isn't left hanging in silence — they take a fallback branch, for example to a manager.
Each tool is switched on with a toggle in the block — the agent gets only what you allow.
The agent looks for the answer in your materials — price lists, policies, service descriptions — and answers from them instead of making things up.
When the answer isn't in your internal documents, the agent runs a web search and pulls up-to-date data.
A customer sends a link to a product or an article — the agent opens it and answers based on the content.
Segments your audience as the conversation goes: adds and removes tags that already drive your broadcasts and automated funnels.
Name, city, budget, product model — the agent writes the data to the contact card, where it is visible in the CRM and can be inserted into messages.
You draw the exits from the block — “Place an order”, “Call a manager”, “Clarify the size” — and the agent decides where to take the customer.
A customer sends a photo or a PDF — the agent sees the image and reads the document, as long as the selected model supports it.
Answers correctly about “today”, “tomorrow before 6 pm” and opening hours — not based on what the model remembers from training.
Upload a price list, a policy or a description of your services — the agent will answer from them instead of speaking in generalities.
Each knowledge base gets its own model for computing vectors: from light and cheap to the most accurate. It also determines the price of the base.
Documents added by link can be rebuilt automatically — the base doesn't go stale when your website changes.
You can check by hand what a query finds and review the log of requests to the base.
The knowledge base is exposed as an MCP server — you can connect it to external models and services, not just to the bot.
The knowledge base is a standalone service with a monthly fee, from $4 per month. The price depends on the embedding model and the volume. Auto-renewal can be turned off in the dashboard.
The platform has 40 models. Each block and each chat gets its own model: some tasks need the smartest one, others a fast and cheap one.
Besides the agent in your bot, the dashboard has a full AI chat with the same model catalog and the same balance. Handy for testing a prompt before you put it into a flow.
Three parts are billed separately, so you pay only for what you actually use.
A regular subscription for the bot and its contacts. When you sign up, you get 7 days of the Pro plan for free.
Billed monthly and deducted from your account balance. The price depends on the embedding model you choose and the size of the base. Auto-renewal can be turned off.
A separate prepaid balance: deducted per token actually used, with no subscription fee for each model.
The AI balance is prepaid. When the money runs out, the agent doesn't answer halfway: the customer takes the “AI error” branch.
The logs show which model was used, how many tokens it took and what it cost, broken down by model. It's easy to estimate your spend on a small volume.
A regular block follows a fixed script: press a button, get a prepared text. An AI agent understands the customer's free-form text, answers in its own words based on your materials and decides on its own which branch of the flow to take next. It still lives inside the same canvas, so buttons, broadcasts and conditions keep working as before.
To keep it factual, connect a knowledge base: the agent searches your documents and answers from what it finds. On top of that, you write a prompt with rules — what it may and may not say — and decide which tools it can use.
When you sign up, you get 7 days of the Pro plan for free — enough to build a bot and set up the AI Agent block. But for the agent to actually work, you need two separate things: a knowledge base (if you want answers from your documents) and an AI balance. Until the balance is topped up, the agent doesn't answer and the customer takes the “AI error” branch.
The knowledge base is a separate service with a monthly fee, deducted from your account balance rather than included in the plan. The smallest option starts at $4 per month. The price depends on two things: the embedding model you choose (from light and cheap to the most accurate) and the size of the base in tokens. Auto-renewal can be turned off in the dashboard.
The AI has its own prepaid balance: you top it up with the amount you need, and it is deducted as you go — per token used, with no subscription fee for each model. All spending is visible in the logs, broken down by model, so you can estimate the cost on a small volume and plan ahead.
A document can be added as text, a file or a link to a page — the platform fetches the content, splits it into chunks and prepares it for search. Documents added by link can update on a schedule.
No: the bot doesn't freeze while the reply is generated. The customer keeps moving through the flow, and once the reply is ready, they return to the block and receive it. Only one reply is generated per contact at a time, so no queue builds up.
All the ones the platform supports: Telegram and Viber. The block is placed in the bot's flow and isn't tied to a specific channel.
When you sign up, you get 7 days of the Pro plan for free — enough to build and set up the AI Agent block right away. The knowledge base and AI replies are billed separately: the base monthly, the AI per token used.