Skip to main content
Version: Current

AI Assistant

The AI assistant is a chat built into the Designer. It answers questions about Nussknacker from this documentation, and it can read and change scenarios for you through a set of tools, the same way you would do it on the canvas.

The assistant is available only when an administrator has enabled and configured it (see AI assistant configuration). On Nu Cloud, account and instance administration is under the avatar menu → Admin; on-premise installations are administered in their own way.

Opening the assistant​

  • In the main menu, choose Help → AI assistant, or click the assistant button on the right-hand rail. The AI Assistant window opens; it can be moved, resized and maximized, and it stays open while you switch between pages.
  • In the command bar (Ctrl+K / Cmd+K), choose AI Assistant, or type a question and choose Ask assistant....
  • In a node's details, an explain? button appears next to validation errors. It opens the assistant and asks it to explain those errors. You can hide these buttons by turning off the assistant.showHelp user setting.

Type your request in the message box (Ask anything or request a change…). The assistant replies in the language you write in.

What the assistant can do​

Answer questions. The assistant searches the Nussknacker documentation before answering questions about components, SpEL, helpers and Designer features. It is limited to what the documentation describes, so it may answer that something is not documented rather than guess.

Know where you are. With every message the assistant gets the Designer page you write from, so a question about "this page" or "this tab" is answered for the page in front of you - on the MCP page, for example, about MCP servers. It also knows whether the instance runs in Nu Cloud or on-premise, where users and the instance are administered differently.

Work with scenarios through tools. When the assistant needs to look at or change something, it calls a tool. Each call is shown in the chat, and a series of calls is grouped (for example 3 tools called) and can be expanded. The available tools are:

AreaToolsAvailable
Scenarioslist_scenarios, open_scenario, create_scenario, archive_scenarios, unarchive_scenarioseverywhere
Skillslist_skills, get_skilleverywhere
Kafka topicslist_topics, get_topic, preview_topic_messages, create_topic, update_topicwhen the Topics extension is installed
Reading the opened scenarioget_scenario, get_validation_resultsin the scenario editor
Editing the opened scenarioadd_node, add_node_between, replace_node, delete_node, add_edge, delete_edge, change_scenario_values, apply_scenario_graph, import_scenario_graph, run_auto_layoutin the scenario editor
Savingsave_scenarioin the scenario editor
Components and integrationsget_components, get_component_definition, list_configured_servicesin the scenario editor
Running scenarioget_live_data (counts, traffic and sample records of the deployed version, see Testing and Debugging)in the scenario editor

The scenario tools always work on the scenario that is currently opened in the editor. To work on another scenario, ask the assistant to open it, or open it yourself. list_configured_services returns the integrations (Kafka connections, LLM providers, databases, HTTP APIs, ...) configured for your instance, so the assistant can pick a matching component.

Follow skills. Skills are short playbooks for tasks that take many steps, such as building a scenario from a description, choosing components, writing SpEL or testing a scenario. They are published by the Designer itself, and the assistant reads them with list_skills and get_skill before a larger change. The same skills are available to external AI agents.

Confirming the assistant's actions​

Tools that only read data run without asking. Before any tool that changes something (editing the graph, saving, creating or archiving a scenario, creating a topic, ...) the chat shows a card with three buttons:

  • ALLOW – run this call.
  • ALWAYS ALLOW – run it, and stop asking about this tool. The choice is remembered in your user settings.
  • DENY – do not run it. The assistant is told that you refused.

The assistant acts with your permissions: it cannot do anything that you could not do in the Designer yourself.

Changes are drafts until saved​

Edits made by the assistant change the scenario in the editor only, just like edits made by hand. They appear on the canvas immediately, but they do not become a saved scenario version until you use Save or the assistant uses save_scenario, which asks for your confirmation first.

With Settings → Scenario → Enable draft on, unsaved changes are kept as a draft and restored when you reopen the same scenario version, including after a page reload. A draft does not create a saved version or deploy the changes. With drafts off, unsaved changes are lost when you reload the page.

Review the result before saving: check the canvas and the validation errors, and ask the assistant to fix anything that is wrong.

What the assistant cannot do​

  • It cannot deploy or cancel a scenario. Use Deploy and Cancel in the scenario's Actions panel (an external agent can do it through nu-cli, see External AI agents).
  • It cannot delete scenarios or topics, only archive scenarios.
  • It does not click the user interface. It only does what its tools allow; anything else, such as opening node details, changing panels or running scenario tests, you do yourself.
  • Scenario editing tools are not available outside the scenario editor.
  • Its knowledge of Nussknacker comes from the documentation. Answers may still be wrong, so verify important changes before saving and deploying.

Conversations​

The assistant remembers the current conversation, so you can refer to earlier messages. To start over, for example when switching to an unrelated task or when a long conversation makes the assistant lose track, click RESET THREAD at the top of the chat. Conversations are not stored permanently; reloading the page starts a new one.