---
title: "How Spark processes your organisation's data"
description: "What happens to the content members send to Spark: which AI provider processes it, what Talkspirit stores, and who decides."
category: administration-security
section: data-and-compliance
tags: [ai]
type: Reference
lastUpdated: 2026-07-29
locale: en
canonical: https://support.talkspirit.com/en/administration-security/how-ai-processes-your-data
---

# How Spark processes your organisation's data


## How Spark processes your organisation's data
If you are deciding whether to switch Spark on, this is what you are agreeing to. The content members send to Spark is processed by an external AI provider on Talkspirit's behalf, and Talkspirit stores a record of it. Talkspirit chooses the provider. Your organisation does not. And the choice can differ from one organisation to another.

## Which AI provider processes our content?

**Anthropic**, using its Claude models, by default. **Mistral** is the sovereign provider we are moving towards, and Talkspirit's platform team can point an individual organisation at a different provider. The provider in use is not a setting inside your organisation, so if you need it confirmed in writing for a security review, ask your Talkspirit contact for the current data processing agreement.

## What is stored?

Spark does not simply forward a prompt and forget it. Three things are kept on Talkspirit's side:

- The prompt text of jobs run through the AI gateway, stored with the job that produced it.
- Spark conversation history, so a conversation can be picked up again.
- An index of the organisation content Spark uses as context. This is what the **Used org context** note on an answer refers to.

Talkspirit is hosted in France and Switzerland, is GDPR by design, and is ISO 27001 certified. What each AI provider may do with the content we send it is governed by the data processing agreement Talkspirit holds with that provider.

**Important:** Spark's inline generation does not write prompt or response text into our application logs, but the records above mean a prompt does not disappear once the answer arrives. Treat what members type into Spark as data your organisation holds.

## What can Spark do with roles and circles?

On a role or circle that already exists, Spark offers **Suggest purpose**, **Suggest domains** and **Suggest accountabilities**, plus **Refine** to rework text you already have and **Review accountabilities** to check them for clarity. Creating a whole role or circle from a name and a description is not available: the **Create with AI** tile in the create dialog is greyed out.

For the step-by-step, see [Creating and editing a circle or role with Spark](/structure-governance/how-to-create-and-edit-a-circle-or-role-with-ai-assistant).

**Note:** your organisation can rename *purpose*, *domains* and *accountabilities*, so these buttons may use your own words.

## Who switches Spark on?

An administrator turns Spark on for the whole organisation, under **Administration > Spark > Activation**. Whether it is available at all depends on your plan: the toggle can be locked with a *contact sales* state, or be running as a trial. Each Spark feature is a separate switch and stays off until someone turns it on. Switching the main toggle off stops every Spark feature at once.

For the steps, see [Enabling Spark for my organisation](/administration-security/how-to-enable-spark-for-your-organisation).

![The Activation tab of Assistant settings, showing the AI assistant card switched on with its description.](/images/administration-security/how-ai-processes-your-data/01-ai-assistant-activation.png)

## What's Next?

- [Enabling Spark for my organisation](/administration-security/how-to-enable-spark-for-your-organisation)
- [Creating and editing a circle or role with Spark](/structure-governance/how-to-create-and-edit-a-circle-or-role-with-ai-assistant)
- [Privacy & Security Settings](/administration-security/talkspirits-administration)