AI assistant

The constructor's assistant answers questions about the module and can edit the scenario by itself: add and configure actions, create functions and form fields, run tests. It works with the open project and knows it is building a Multogram module.

Open it with the robot button in the bottom-right corner of the window — Open AI Chat.

Modes

You don't have to switch modes by hand. Ask the assistant in Ask mode to build, add or fix something, and it switches to Edit by itself and does it in the same answer: the chat shows a "Switch to Edit mode" step and the mode switch turns to Edit. Questions, explanations and ideas it answers without switching — and offers to build when the answer adds up to a module.

In edit mode each of the assistant's steps is visible in the chat: what it read, what it added, how it checked it. Any change can be undone like an ordinary edit — Ctrl+Z. The assistant doesn't change the protected accounts field.

What the assistant knows about Multogram

• Multogram's modes and your plan. The assistant knows every mode, what each one does and which of them your account has. When you describe something a ready-made mode already does, it says so first — and recommends the mode (or the plan that includes it) instead of rebuilding it as a module.
• The actions open to you. A Telegram action whose mode you don't have is greyed out in the constructor — the assistant doesn't add or run it either, and tells you which mode opens it.
• The module template. What the Main, ProcessAccount and AuthSession functions and the accounts resource of a new module do. In its answers the assistant says what each thing is: "the AuthSession function", "the accounts resource", "the Start Session action".

Where to start

In an empty chat the assistant offers ready requests:

• "Explain step by step what this module does"
• "Add a step: send a message to a chat from each selected account"
• "Add a field for the message text to the module form and use it in the scenario"
• "Add FloodWait handling: wait as long as Telegram asks and repeat the step"
• "Check the module for errors before building"

If the plan doesn't allow the assistant to edit the module, it suggests questions — for example, how the module gets accounts from the form and how to handle Telegram's limits.

AI credits

The assistant runs on the Multogram model and spends your account's AI credits — the same as the assistant in the application. The balance is visible in the model menu: "… of … credits left" and the renewal date. When the credits run out, the assistant tells you when they renew.

There is one model — Multogram: the assistant itself uses the best model your plan includes. The assistant needs no keys or settings of your own. If the module itself needs AI — for example, to write texts — there is the AI Request action with your own provider key for that, see Runtime and interface.

AI agents (MCP)

Modules can also be built by your AI client — Claude Code, Cursor, Codex and others connected to Multogram over MCP (the Pro and Max plans). The assistant's model menu has an MCP block: it shows whether agents are connected and opens the connection guide. Only one party works with a module at a time — the constructor's assistant or an agent; while an agent is working, the space above the input field shows who it is, and the Take over button pauses the agents.

How to give tasks

• Describe the result, not the steps. "Let each account join the channels from the list with a pause from the settings" is better than a list of actions.
• Name the places. A function, a resource, an action — the assistant finds them in the project.
• Ask it to check. "…and check it with a test" — the assistant runs the function and shows the result.
• Big things in parts. First the skeleton, then one function at a time.

If the assistant doesn't answer

Updated October 18, 2026