> 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/data-storage-and-retention.md).

# Data storage & retention

{% hint style="info" %}
**Target Audience**: Developers, security reviewers, and compliance stakeholders evaluating how Toucan AI handles data.
{% endhint %}

### TL;DR

* **Your business data stays in your databases.** Toucan AI does not maintain a separate copy of your warehouse or tables as a downloadable dataset.
* **Query results are fetched on demand** for each chart view or AI action, then held only for the duration of that operation (and related UI/AI workflows).
* **Toucan AI stores platform data** (accounts, Dashboard JSON configuration file refering the different components WITHOUT the related data, chart JSON Configuration including the query but not the underlying data, AI conversation history) in its PostgreSQL database.
* **Database connection credentials** are stored encrypted at rest on the platform.
* **There is no dedicated long-term cache of query result rows** on the platform; see [Query data flow & caching](#query-data-flow-and-caching) below.

***

### When to use this

Use this page to understand what Toucan AI persists versus what it processes transiently, and how that relates to retention and deletion.

***

### What Toucan AI stores (platform data)

| Category                         | Examples                                               | Stored by Toucan AI?                                      |
| -------------------------------- | ------------------------------------------------------ | --------------------------------------------------------- |
| Account & organization           | Users, memberships, invitations                        | Yes                                                       |
| Product configuration            | Dashboards, charts, themes, variables, access policies | Yes                                                       |
| Chart definitions                | Query pipeline, chart type, visualization config       | Yes (definition only — not a snapshot of all result rows) |
| Database credentials             | Connection secrets for your connected databases        | Yes, encrypted                                            |
| AI conversation history          | Chat messages and related assistant state              | Yes                                                       |
| Your table rows / warehouse data | Sales records, HR data, etc.                           | **No** — remains in your database                         |

***

### Query data flow & caching

This section answers: *Does Toucan download my data? Is there a cache? Is that cache encrypted?*

#### Does Toucan copy or download my database?

**No.** Toucan AI does not ingest or replicate your connected database into a Toucan-owned data lake.

When a user opens a chart, runs a filter, or asks the AI assistant a data question, Toucan AI:

1. Builds a query from the chart definition or AI workflow.
2. Applies your security rules (row-level and column-level policies).
3. Sends the query to your database **through a dedicated query execution service**.
4. Receives the result for that request only.
5. Uses the result to render the visualization or answer the question.

Your data remains the system of record in **your** database.

#### Is there a query result cache?

**There is no separate, platform-wide cache of query result datasets** (for example, no shared Redis or disk store of “all rows returned from customer DB X”).

| Layer                        | What happens                                                            | Persisted?                                                                                   |
| ---------------------------- | ----------------------------------------------------------------------- | -------------------------------------------------------------------------------------------- |
| **Connected database**       | Source of truth for your rows                                           | Yes — in your infrastructure                                                                 |
| **Query execution**          | Runs the pipeline against your DB for each request                      | No copy kept after the operation completes                                                   |
| **Application memory**       | Holds the result while building a response (chart data, AI tool output) | No — scoped to the request/worker lifecycle                                                  |
| **Chart / dashboard config** | Stores the *query definition* and visualization settings                | Yes — not the resulting dataset                                                              |
| **AI conversation**          | May retain messages and tool outputs from a session                     | Yes — see [AI assistant data handling](/govern/security-model/ai-assistant-data-handling.md) |

Charts are **re-executed** when data is needed again (for example when a user refreshes a dashboard or changes a filter), rather than served from a long-lived Toucan cache of your rows.

#### Encryption of “cached” data

Because query results are **not stored in a dedicated Toucan query cache**, there is no separate “cache encryption” layer for result rows.

Protection instead relies on:

* **Encryption in transit** between Toucan services and your database (TLS as configured for your connection).
* **Encryption at rest** for **connection credentials** stored on the platform (see [Secrets & encryption](/govern/security-model/secrets-and-encryption.md)).
* **Encryption at rest** for the Toucan platform database.
* **Access control** on every query (tokens, permissions, RLS/CLS) and roles (ADMIN, MAKER or EXPLORER).

#### What about Redis?

Redis is used for **background job queues** (asynchronous tasks such as AI message processing), not as a long-term store of your query results. Job metadata is kept for a short period, then removed automatically.

***

### AI and conversation data

AI assistant conversations (including text you type and outputs from data tools) can be **persisted** so users can continue a thread across sessions.

* Clearing a conversation removes that thread’s stored history through the product API.
* Automatic time-based purge of all conversations is a **product/policy** topic — confirm current behavior with Toucan AI support if you need a specific retention period.

See [AI assistant data handling](/govern/security-model/ai-assistant-data-handling.md) and [PII & personal data](/govern/security-model/pii-and-personal-data.md).

***

### Retention & deletion (overview)

| Data type                    | Typical retention                     | Deletion                                                  |
| ---------------------------- | ------------------------------------- | --------------------------------------------------------- |
| Platform account data        | While the account/organization exists | Contact Toucan AI support for organization-level requests |
| Chart & dashboard config     | Until deleted in the product          | User/admin deletion in product                            |
| AI conversation (per thread) | Until cleared or org removed          | Clear conversation in product                             |
| Query result rows            | Not retained as a platform dataset    | N/A — not stored as a bulk copy                           |
| Connection credentials       | Until database connection removed     | Delete connection in product                              |

For third parties that may also hold data (LLM providers, analytics), see [Third-party subprocessors](/govern/security-model/third-party-subprocessors.md).

***

### Best practices

* Treat Toucan as a **query-and-visualize layer**, not a second database for your raw data.
* Use **row-level security** so every query is scoped to the end user.
* Avoid typing sensitive personal data into AI chat unless you accept it may be stored in conversation history.
* For embed integrations, use **pseudonymous identifiers** in tokens rather than email or legal name (see [PII & personal data](/govern/security-model/pii-and-personal-data.md)).