Overview AI Assistant


The AI Assistant answers questions about the uni-PRO Studio in natural language. The answers are based on the local HTML help (including this page) and are generated by a configurable language model.

The assistant is a help facility — it does not control the Studio. It does not modify the open project nor invoke Studio functions. If you want an AI to actually edit the project, use the MCP Server (AI Interface) instead.


How to open it

The dialog is non-modal — it can stay open while you continue working in the Studio.


Usage

  1. On first launch, click the gear icon in the top right corner and fill in the settings (provider and API key). Without a valid key the assistant cannot generate answers.
  2. Type your question into the input area at the bottom.
  3. Click Send or press Ctrl+Enter to submit.
  4. The answer appears in the conversation view above. At the end of the answer, the sources used are listed — clicking one opens the original help page in the browser.
  5. The clear-history icon in the toolbar resets the conversation.


Settings

Settings are opened from the dialog itself via the gear icon. They are not managed via the global Studio settings, but stored in the Windows registry under the same key the Studio uses (sub-group HelpAssistant).

Setting Description
Provider Selects the API protocol. Anthropic (Claude) uses the anthropic.com Messages API. OpenAI-compatible uses the OpenAI Chat Completions schema, which covers OpenAI, Groq, OpenRouter, Mistral, local Ollama, LM Studio and many other servers.
Base URL API endpoint address without a trailing slash. Examples:
  • Anthropic: https://api.anthropic.com
  • OpenAI: https://api.openai.com/v1
  • Groq: https://api.groq.com/openai/v1
  • OpenRouter: https://openrouter.ai/api/v1
  • Ollama local: http://localhost:11434/v1
API Key Secret access token of the provider. Use the Show button to reveal it in plain text. For local servers (e.g. Ollama) the field may be left empty. The key is stored as plain text in the registry, same as other Studio passwords.
Model Name of the language model. The field is freely editable; the entries are only suggestions.
  • Anthropic: claude-sonnet-4-6 for best quality, claude-haiku-4-5-20251001 for speed/cost.
  • OpenAI: gpt-4o-mini is cheap and usually enough, gpt-4o gives better answers.
  • Ollama: name as installed (e.g. llama3.2).
Max Tokens Maximum answer length in tokens (about 4 characters per token). Range 256–8192, default 1024. Higher values allow longer answers but cost more.
Top-K Sources Number of help excerpts sent as context to the model. Range 1–10, default 5. More excerpts give the model more context, but make requests slower and more expensive.


How the assistant works internally

  1. On the first request the HTML help for the current language is indexed (BM25 full-text search, held in memory). This takes only a few seconds.
  2. The query is matched against the index and the Top-K best help pages are selected.
  3. For each match a snippet of about 400 characters is extracted (around the first match position).
  4. Question and snippets are sent together with a system instruction to the configured provider.
  5. The model's answer is shown; the sources it cites at the end are rendered as clickable links.

The assistant does not perform any action in the Studio — it only reads the HTML help.


Provider setup

Anthropic (Claude)

  1. Create an account at console.anthropic.com.
  2. Under API Keys generate a new key (starts with sk-ant-...).
  3. In the assistant: Provider = Anthropic (Claude), Base URL = https://api.anthropic.com, paste the key, pick a model (e.g. claude-sonnet-4-6).

OpenAI

  1. Create an account at platform.openai.com.
  2. Under API Keys generate a key (starts with sk-...).
  3. In the assistant: Provider = OpenAI-compatible, Base URL = https://api.openai.com/v1, paste the key, model gpt-4o-mini or gpt-4o.

Local model with Ollama

Ollama runs entirely offline on your own machine. There are no recurring costs, but answer quality depends on your hardware.

  1. Download and install Ollama.
  2. Pull a model, e.g. in a terminal: ollama pull llama3.2
  3. In the assistant: Provider = OpenAI-compatible, Base URL = http://localhost:11434/v1, leave API Key empty, model llama3.2.


Privacy

For cloud providers, the question and the selected help excerpts are sent over the internet. No project data, no visualisations and no variable values are transmitted — only text from the local HTML help files.

If you do not want any data leaving your machine, use a local model (see the Ollama section above).


Troubleshooting