# Your own AI model – assistant and AI on cards

> Connect your organization's own language model (OpenAI-compatible, Azure OpenAI, Mistral, Anthropic, Google Gemini or a self-hosted server) and use the assistant, card summaries, checklists, labels, translation and boards from a prompt – with a data policy you control.

Source: https://contals.com/docs/en/ai/own-model/  
Last updated: 2026-10-16

Contals works with **your organization's own AI model**: you bring the provider, the contract and the API key, Contals sends only what your data policy allows, and only from its servers. Nothing is sent anywhere until an organization owner or admin switches AI on.

## 1. Choose a provider

Pick a provider that fits your data protection needs – ideally an endpoint in the EU or Switzerland, for example Azure OpenAI in an EU region, Mistral, or your own server. You need a data processing agreement with that provider, because board content leaves Contals for it.

Supported: **OpenAI-compatible** (OpenAI, Azure OpenAI, Mistral, Ollama, LM Studio and other servers with the Chat Completions API), **Anthropic (Claude)** and **Google Gemini**.

## 2. Enter it in the organization settings

Open the organization's settings page and go to the **AI** section (owners and admins):

1. **Provider** – and, for OpenAI-compatible servers, the **Endpoint (base URL)**, e.g. `https://<resource>.openai.azure.com/openai/v1` or `https://api.mistral.ai/v1`.
2. **Model** – the exact model name (Azure: the deployment name).
3. **API key** – stored encrypted and never shown again, not even to you.
4. Tick that you have a data processing agreement, click **Test connection**, switch **AI on for this organization** and **Save**.

Endpoints must use https and a public address; private networks are refused.

## 3. Decide what may be sent

Under **Data policy**: titles of boards, lists and cards are always needed. **Card descriptions** and **Checklists** are on by default; **Comments** and **People's names** are off by default, because comments often hold personal details (pastoral care, health, children). Without names, comment authors are sent as "Person 1, 2 …". E-mail addresses are never sent, and **confidential boards are never sent**.

Under **Features** you can switch off each feature – all are on once AI is switched on, except **Summarize my week (home page)**, which stays off until an owner or admin switches it on – and under **Workspaces** you can switch AI off for workspaces whose boards must never reach a language model. **Usage (last 30 days)** shows tokens per day and feature – the usage log stores no prompts and no answers.

## 4. Use it

- **Assistant** – on a board, click **Assistant** in the board header (it opens in the sidebar, like the card details); on the home page use the **Assistant** button. Ask questions or describe changes. The assistant uses the same operations as the API, with your permissions. Every change is shown first and runs only after you click **Run**. Conversations are kept for 30 days; delete them under **Conversations**.
- **On a card** – the **AI** section offers **Summarize**, **Draft description**, **Checklist**, **Suggest labels** and **Translate a card**. Each result is a preview: apply it or discard it.
- **Your week** – on the home page, **Summarize my week** writes a short summary of what is overdue, due and waiting for your answer, from titles and dates – only when you click it ([your home page](/docs/en/views/for-me/)).
- **Since your last visit** – on a board, **Summarize** in the strip writes two or three sentences about what others changed since your last visit, from titles and counts – only when you click it, never for a confidential board ([since your last visit](/docs/en/boards/basics/#since-your-last-visit)). The feature is **Summary since the last visit** under **Features**.
- **Board with AI** – on the home page, describe a project (e.g. "Plan our summer camp for 40 children"), check the suggested lists and cards, then **Create this board**. The new board is private until you share it.

## Costs

Contals sells no AI credits: your organization pays its provider directly for the tokens it uses, and the AI features are included in every plan, the free plan too. The usage overview in the AI section shows how many tokens each feature used.