# AI assistant – TANGENTA

> The TANGENTA AI assistant draws real vectors, operates the program, finds images and checks its own result. Claude via your subscription, OpenAI, Gemini, DeepSeek or Ollama.

URL: https://tangenta.ch/en/ai.html

AI assistant

# An assistant  that really draws.

It does not just answer. It works in your document, with the same tools as you, and looks at the result before it reports back.

![The AI assistant window next to the poster, with the request, the steps and the answer](https://tangenta.ch/img/02-ki-assistent.png)

**01** / How it works

## Ask. Watch. Done.

1

### You describe

In the AI window (Ctrl+Alt+K) you write what you need, in full sentences and in your own language.

2

### It works

It looks at the page, draws, searches for images or runs menu commands. You see every step in the log.

3

### It checks

Finally it views the page as an image, corrects what does not fit and sums up what it did.

**02** / What it can do

## 18 tools at its disposal.

### Draw

Shapes, curves and text as real vectors, by script or as SVG. Everything stays editable.

### Operate

Run any menu command: group, align, weld shapes, convert to curves, trace, export.

### See

View the page or a part of it as an image, and read the properties of individual objects.

### Find images

Search eight image libraries, preview results and insert them with author and licence.

### Search the web

Research via SearXNG or Brave Search and read web pages, for example for colours or facts.

### Tidy up

Select, rename and delete objects – but only when you explicitly ask for it.

**03** / Providers

## You choose the model.

Set up several accounts and switch between them in the AI window. Keys stay locally with your user account.

| Account | How | Cost |
|---|---|---|
| Claude via your subscription | Install Claude Code and sign in; TANGENTA uses it | included in Pro or Max |
| Claude API | API key from Anthropic | pay as you go |
| OpenAI | API key | pay as you go |
| Google Gemini | API key | pay as you go, with a free tier |
| DeepSeek | API key | pay as you go |
| Ollama | your own server, even on your own computer | free, data stays in-house |
| LiteLLM | your own proxy for any model | depends on the model |

**04** / Safety

## You stay in control.

- Without the “darf ändern” (may change) tick it only reads
- One Ctrl+Z undoes everything it changed in one answer
- Deletes only when you explicitly ask
- Does not read web pages on your own network (localhost, 10.x, 192.168.x)
- Limits on rounds, tokens and time per answer
- Keys stored locally with your user account, never in documents

### Try it

> “Draw a simple logo for a bakery: a loaf in a circle, with the name below in a rounded typeface.”

> “Align all the images on this page to the top and distribute them evenly.”

> “Find a free photo of the Rhine Falls and use it as a full-page background.”

> “Make a black-only version of the logo on this page.”

**05** / Setup

## Ready in five minutes.

1. Open TANGENTA and go to **Extras > Optionen > Global > KI** (Ctrl+J).
2. For Claude via your subscription: install [Claude Code](https://claude.com/claude-code), then click **Anmelden** (sign in) in the dialog and confirm with your Claude account.
3. For another provider: **Neu …** (new), choose the provider and enter the API key.
4. Click **Test** to check that the account responds.
5. Open **Extras > KI-Assistent** (Ctrl+Alt+K) and start writing.

The program interface is currently in German; you can write to the assistant in any language. For web search, enter your own SearXNG server or a Brave Search key under **Global > Websuche**.

## Start drawing.

Free of charge, for Windows and Linux, with AI.

[Download](https://tangenta.ch/en/download.html) [View on GitHub](https://github.com/roli8200/tangenta)
