Question 1

Which Claude models are supported?

Accepted Answer

All publicly available Claude models including the Opus, Sonnet, and Haiku tiers. Ana picks the right model based on cost, latency, and task complexity, or admins can restrict the allowed model set. Long-context variants are particularly useful for warehouse document analysis.

Question 2

Does warehouse data go to Anthropic?

Accepted Answer

Only the data Ana explicitly passes to the API. Admins can configure column-level allowlists to prevent sensitive warehouse fields from being sent to Anthropic, and every outbound payload is logged so you can audit exactly what left your environment.

Question 3

How is the Anthropic key secured?

Accepted Answer

The API key is encrypted at rest in TextQL's credential vault and injected by our egress proxy at the network layer. Ana's Python sandbox never sees the raw key value. You can rotate the key any time from the TextQL admin console without changing prompts.

Question 4

Is prompt caching supported?

Accepted Answer

Yes. The connector supports Anthropic prompt caching, which can dramatically reduce cost when Ana reuses the same context across multiple calls in a session. This is especially useful for workflows that apply the same instructions to many warehouse rows.

Question 5

Can I deploy via AWS Bedrock?

Accepted Answer

Yes. If you prefer Claude via Bedrock for data residency or compliance, an admin can configure the connector to call your Bedrock endpoint and whitelist the corresponding AWS domain. The same sandbox security and credential handling apply.