AI and Security

By default, our product uses Anthropic’s Claude Haiku 5.5, a fast and capable large language model. It is great for data analysis, query generation and configuring visualizations.

You can change the provider and model for your organization from Settings → AI behavior. See Choose your AI provider and model.

We chose Anthropic for the performance and reliability of the Claude models, and for Anthropic’s strong focus on AI safety and responsible data handling.

By default, our AI features are powered by Anthropic’s API (Claude). All interactions with the AI are processed through their secure infrastructure.

No. Anthropic does not use your data or inputs submitted via their API to train or improve their models. Your data remains private and is only used to provide the service you request.

Yes, all data transmitted to and from Anthropic’s API is encrypted in transit using industry-standard protocols (TLS).

Our AI provider (Anthropic) does not use your data to train its models. API inputs and outputs are retained only for a limited period for trust and safety purposes, after which they are deleted.

We will never sell or share your data with any third parties for marketing or advertising purposes.

To improve the assistant, Toucan AI sends AI assistant conversations to LangSmith (EU region), our AI monitoring service. This lets our team see how the assistant handles your use cases and fix what goes wrong for you.

What is sent: the questions, the answers, and the data excerpts the assistant read to answer them (sample rows, column values, query results). Users appear under a technical ID, not their name. In embedded analytics, this ID is the one your application provides.

How long it is kept: 14 days in LangSmith, including conversations rated with :thumbsup: or :thumbsdown:.

Choosing what is shared: in Settings > General, an admin chooses which conversations are shared: all of them (the default), only the ones a user rates with :thumbsup: or :thumbsdown:, or none. When a rating shares the conversation, the :thumbsup: and :thumbsdown: buttons say so.

On self-hosted installations without LangSmith, no conversation is shared.

Yes, you can request the deletion of your organization data at any time. We will ensure your data is removed from our systems and from third parties it was shared with.

We are evaluating new models regularly as they are released. Any changes to our AI provider will be communicated transparently through updates of the page.

Yes. Each organization can select its AI provider and model from Settings → AI behavior. Leave it on Default to follow Toucan's recommended provider and model, or pick a specific provider (and optionally a model). The choice applies to your whole organization. See Choose your AI provider and model for the step-by-step guide and the list of available providers.

Not yet. You can select from the providers and models we offer (see above), but bringing your own API key or connecting a self-hosted model isn't supported today. We may add this in the future — contact us if it's something you need.