Settings¶

The Settings window is where you configure Kanzi AI Assistant — your LLM provider and API key, the knowledge base, web search, the MCP server, and general behavior. Open it from AI Assistant > Settings in the Kanzi Studio menu bar, or with the gear icon in the chat window toolbar.

A category tree on the left groups the settings. Select a category to see its settings on the right. The search box above the tree filters the categories and settings as you type, so you can jump to a setting by name.

Changes take effect when you click Save. Some settings — noted below — take effect only after you reopen the Kanzi AI Assistant window or restart Kanzi Studio.

AI Provider¶

Configure which LLM answers your requests and the credentials it uses.

Setting	Description
Provider	The LLM provider: OpenAI, Anthropic, Google, or DeepSeek. Changing the provider takes effect after you reopen the Kanzi AI Assistant window.
API Key	Your provider API key. Kanzi AI Assistant stores it locally in `%APPDATA%`, encrypted with Windows DPAPI under your user account. Each provider keeps its own key, so you can switch providers without re-entering it.
Model	The model to use. Select one from the list, or type any model ID directly if your provider offers a model that is not listed.
Custom API URL (OpenAI-compatible)	Only shown for the OpenAI-compatible provider. The full endpoint URL including the chat-completions path, for example `https://api.example.com/v1/chat/completions`. Enter the full URL, not only the base.

For where to obtain a key for each provider, see Setting up.

Knowledge Base¶

Kanzi AI Assistant ships with a curated knowledge base of Kanzi task recipes, domain concepts, and best practices that it queries while it works.

Setting	Description
Retriever mode	How Kanzi AI Assistant searches the knowledge base: Onnx (semantic embedding search, the default) or BM25 (keyword matching). Changing the retriever takes effect after you reopen the Kanzi AI Assistant window.

Web Search¶

Web search lets Kanzi AI Assistant look up current information on the live web for questions the curated knowledge base does not cover. It is off by default and requires a search-provider API key. See Web search.

Setting	Description
Enable web search	Turns the `web_search` tool on or off. Off by default.
Search API key	Your search-provider API key, shown once you enable web search. Kanzi AI Assistant stores it locally in `%APPDATA%`, encrypted with Windows DPAPI.
Search API URL	The search endpoint, shown once you enable web search. Leave it blank to use the default Tavily search API.

MCP Server¶

The MCP (Model Context Protocol) server exposes Kanzi Studio tools and project state to external AI tools such as Claude Code and Cursor. See MCP server.

Setting	Description
Enable MCP server	Turns the MCP server on or off. Enabled by default. Changing this takes effect after you restart Kanzi Studio.
MCP status	Shows whether the server is running and its address. Use it to copy the server URL for your external client.

General¶

Setting	Description
Show detailed communication	Shows tool-action descriptions inside assistant messages, so you can see what Kanzi AI Assistant did at each step. On by default.
Normalize terminology	Rewrites colloquial terms in your messages to canonical Kanzi terms before the assistant runs. The chat keeps your original text; only the assistant sees the rewrite. Off by default.
Send anonymous usage data	Sends anonymous telemetry that helps Rightware improve Kanzi AI Assistant. It collects no personal data or conversation content. On by default. See Privacy and telemetry.
Request timeout (seconds)	How long to wait before canceling a single LLM request. Range 10–600, default 300.
Max agent iterations	How many LLM turns Kanzi AI Assistant can take on one request. Range 10–500, default 200. Raise it for complex builds, lower it to cap cost.