> For the complete documentation index, see [llms.txt](https://docs.toucanai.cloud/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.toucanai.cloud/govern/security-model/pii-and-personal-data.md).

# PII & personal data

{% hint style="info" %}
**Target Audience**: Developers and privacy stakeholders integrating Toucan AI, especially in embed and AI chat scenarios.
{% endhint %}

### TL;DR

* Toucan AI stores **account and session data** for users of the Toucan platform (for example email, name, session metadata).
* **Your end-user data** in embed mode is mostly under **your control**: what you put in tokens and chat.
* **AI conversations can persist** what users type — treat chat as potentially storing personal data.
* **Minimize** what you send: pseudonymous IDs, only the attributes needed for security, no unnecessary context in AI clues.

***

### When to use this

Use this page before going to production with embed or AI features, or when answering privacy questionnaires about where personal data can appear.

***

### Where personal data can appear

#### 1) Toucan platform accounts

When users sign in to Toucan AI directly, the platform stores typical account information such as:

* Email address and display name
* Profile fields you provide at signup
* Session metadata (for example IP address or browser information, when collected)

This data supports authentication, billing, and product operation.

#### 2) Embed integrations (your application)

In embed mode, **you** define what identity and attributes are sent to Toucan AI inside the embed token, for example:

* A stable user identifier (`distinctId`)
* Optional attributes used for row-level security (region, department, role, etc.)
* Optional free-text context for the AI (`aiContextClues`)

Depending on what you send, these fields **may be personal data**. Toucan AI does not require email or legal name in embed tokens.

#### 3) AI assistant chat

Users may type personal data directly into chat. That content can be **stored** as part of conversation history and is necessary to run a conversation, as it will be the context used as input for the LLM answer.

Query results shown in AI workflows may also appear in conversation state (for example summarized or sampled rows). See [Data storage & retention](/govern/security-model/data-storage-and-retention.md).

***

### Data you control vs Toucan controls

| Area                    | Who controls content | Guidance                                                      |
| ----------------------- | -------------------- | ------------------------------------------------------------- |
| Embed token attributes  | Your backend         | Send only what RLS or product features require                |
| `distinctId`            | Your backend         | Use a technical ID, not email or name                         |
| `aiContextClues`        | Your backend         | Avoid personal data; prefer non-identifying context           |
| Chat messages           | End user             | Train users; clear history when needed                        |
| Connected database rows | Your database        | Use RLS; Toucan queries on demand, does not bulk-copy your DB |

***

### Best practices (minimization)

* Use a **pseudonymous** `distinctId` (never email, phone, or legal name).
* Limit embed **attributes** to the minimum required for access control.
* Do not pass personal data in **AI context clues** unless strictly necessary.
* Apply **row-level security** on all sensitive tables.
* Clear AI conversation history when your product workflow requires it.
* Align with your privacy policy and DPA for subprocessors listed in [Third-party subprocessors](/govern/security-model/third-party-subprocessors.md).

***

### Related pages

* [Data storage & retention](/govern/security-model/data-storage-and-retention.md)
* [AI assistant data handling](/govern/security-model/ai-assistant-data-handling.md)
* [Security boundaries (embed)](/embed/embedding-overview/security-boundaries.md)