Observe by Snowflake's AI-powered features use a model from Snowflake Cortex by default. If your organization needs LLM inference to run entirely inside your own AWS environment, you can connect your Amazon Bedrock account so Observe by Snowflake routes that inference to Bedrock instead of Snowflake Cortex.

How it works

Once you connect your Bedrock account, every AI feature that calls a large language model routes that call to your Bedrock account instead of to Snowflake Cortex.

There is no fallback to the default provider if a model is unavailable in your account. Make sure every model listed under Required model access is enabled before you turn the integration on.

You can delete your Bedrock integration at any time; see Delete an Amazon Bedrock integration. Your AWS IAM role stays valid if you reconnect later.

Set up an Amazon Bedrock integration

Each account supports one Bedrock integration at a time. Only account admins can set it up.

1. Request a setup payload: call POST /v1/llm-integrations/setup. Observe by Snowflake returns an externalId, a trustPolicy, and a permissionPolicy.
2. In your own AWS account, create an IAM role for Observe by Snowflake to assume:
   1. Attach the trustPolicy from step 1 as the role's trust policy. It authorizes Observe by Snowflake's AWS principal to assume the role, and requires the externalId you were issued so only Observe by Snowflake can assume this role on your behalf.
   2. Attach the permissionPolicy from step 1 as the role's permissions policy. It grants the Bedrock permissions needed to run inference; you can attach it as-is.
3. In Bedrock's Model access settings for your account, grant access to every model listed in Required model access.
4. Finish setup: call POST /v1/llm-integrations with your IAM role's ARN, its AWS region, and the externalId from step 1.

AI feature inference starts routing to your Bedrock account as soon as setup completes.

Required model access

Grant your IAM role access to every model below in Bedrock's Model access settings. Requests route across all of them — if one is missing, requests that need it fail rather than fall back to another model.

Retention and guardrails

Your Bedrock integration runs inside your own AWS account, so Bedrock's retention settings and guardrails are yours to configure — Observe by Snowflake does not set or manage them for you. Before enabling the integration, review and configure Bedrock's model invocation logging, data retention, and Guardrails settings in your AWS account according to your own data-handling and compliance requirements.

Delete an Amazon Bedrock integration

Only account admins can delete a Bedrock integration.

1. Find the integration's ID: call GET /v1/llm-integrations. The response lists your account's integration and its ID.
2. Delete it: call DELETE /v1/llm-integrations/{id} with that ID.

Once the integration is deleted, every new LLM call goes to Snowflake Cortex. Any LLM call already in progress falls back to Snowflake Cortex within 15 minutes.

Deleting the integration does not change anything in your AWS account. The IAM role you created stays in place, and you can delete it yourself after you delete the integration. Your externalId is also kept, so if you set up the integration again later, the trust policy on your existing role still works.

Related resources

For details on authenticating and making calls to the Observe by Snowflake API, including the setup and integration calls described in this document, see the API documentation.