AI agents
Build an AI agent without code
A bot that answers customers on your website, in Telegram and VK from your own documents. You assemble it from ready-made blocks, and after every answer you see what it did and what it cost.
All nodes · 128
- Input / Output3
- AI33
- Logic & Flow9
- Data31
- Integrations52
Answer to the customer
AI agent
Prompt
You answer customers of an online shop. Keep it short and stay inside the company documents. If the answer is not in them, say so and offer to take a contact.
Model
The answer is put together by an AI agent: search across documents to its left, conversation memory below, delivery to its right. Each tool is a node you connected, and it has no others.
300 quarks a month once your address is confirmed, no card needed. That is a few hundred answers, enough to find out whether the agent copes with your questions.
Once the agent is answering customers
«Will it do something it should not?»
The agent only goes where you let it
The agent's tools are the nodes you connected to it. If you did not connect sending mail, it cannot send mail, whatever it made up in its answer.
After that it is the same branching as everywhere else: no matching paragraphs, a service refused, the budget ran out, and each case follows a branch you wrote, not one the model chose. What happened in a particular conversation is laid out by the run breakdown.
- It cannot call what is not there tools are connected nodes, and no others appear
- It cannot loop forever the number of rounds is capped, and on the last one it must answer
- It cannot outspend what you gave the workspace has its own budget in quarks, and that is what stops it
Next
What one answer costs
You are not charged a subscription per seat but for work done: the scenario run, the model call and the tokens on top. A typical answer from documents costs less than one quark, and that arithmetic is laid out on the pricing page rather than hidden in an average.
Your own number shows up straight away: on the runs page every answer is broken down node by node, so you can see which one ate the most.
One answer to a customer
- Running the scenario0.40
- The model call0.10
- 1,940 tokens in0.194
- 88 tokens back0.044
- Total0.74
One particular conversation on one model, not a tariff.
Build an agent for your own case
Take the three questions you are asked most often and give the agent the documents that hold the answers. Half an hour later you will know whether this suits you.
Build your agent