Skip to main content

Log in to comment and vote

Comments1

  • Jade Petal

    •

    Nov 9

    Organizations in compliance-regulated industries (defense contractors under CMMC/DFARS, government, healthcare, finance) often cannot send internal documentation to commercial AI APIs. Many of them already run AI models through Amazon Bedrock inside their own AWS environment - including AWS GovCloud (US) - where access is private, authenticated with AWS IAM, logged, and covered by their existing compliance boundary. The current Hudini providers (OpenAI, Azure OpenAI, Grok, and Claude via an Anthropic API key) connect only to commercial endpoints, so these organizations cannot use Hudini today. Native Bedrock support would open Hudini to them without requiring any new AI vendor approval.

    Requested capabilities

    1. Provider type "Amazon Bedrock".

    2. Region selection, including AWS GovCloud: us-gov-west-1 and us-gov-east-1 (partition aws-us-gov), not only commercial regions.

    3. Custom endpoint URL (override), so a private VPC endpoint or a FIPS endpoint can be used.

    4. Model ID as free text, accepting Bedrock model IDs and inference profile IDs (cross-region profiles such as "us." or "us-gov." prefixed IDs), so customers can use whichever models are enabled in their account without waiting for Hudu to update a list.

    5. AWS-native authentication with short-lived credentials. In order of preference: a. IAM Roles Anywhere (X.509 certificate -> temporary credentials) for on-prem servers; b. AWS credential profile / default credential chain on the Hudu host; c. Access key + secret with optional session token (least preferred). Requests signed with AWS SigV4 for service "bedrock".

    6. API: Bedrock Converse / ConverseStream (model-agnostic, so one implementation serves all model families). Support for Bedrock prompt caching where the model offers it would reduce cost for repeated context. If Hudini needs embeddings (e.g. for semantic search), support for Bedrock embedding models as well.

    7. Outbound proxy setting and custom CA bundle for the Bedrock connection.

    8. Admin controls: choose which Hudini features may use the provider, and an option to log which records/content were sent to the model (audit trail).