INSIDE 3.3
AIP-006

AWS Bedrock model connection

AWS Bedrock connects supported Bedrock models to your AI features using the configured AWS region.

admindeveloper
DraftThis article hasn't been reviewed yet.
Where to find it
Settings › AI › ModelsSettings › AI › Accounts

What it is

AWS Bedrock lets AI features use supported models through a Bedrock connection. The account chooses the model and AWS region. INSIDE supports both complete replies and streaming model output.

Who uses it and why

Administrators agree which AI features should use Bedrock. Developers work with Powerfront to check the model, region and authentication. Powerfront manages the connection for you.

Where to find it

Powerfront selects AWS Bedrock as the Provider in Settings › AI › Models. The Model, API Key and Region are configured in Settings › AI › Accounts. Both screens are Powerfront-managed.

How it works

  1. The feature uses an AiFlow account with a Bedrock model.
  2. INSIDE selects Converse or Invoke. Unless Powerfront has configured a request-mode override, model identifiers containing Claude or Nova use Converse first; other models use Invoke first.
  3. INSIDE sends the conversation and supported function information. Supported modes can also carry images or documents.
  4. The reply can include text, function requests and usage information. Streaming output is assembled into the completed result.

If the selected request mode is unavailable and returns a recognised method-compatibility error, INSIDE can try the other mode. This stays with the same model and is not provider failover. A fixed custom endpoint bypasses this mode fallback.

Configure it

Ask Powerfront to prepare the Bedrock model or inference-profile identifier and the account's API Key and Region. If no region is set, INSIDE uses us-east-1. Powerfront should confirm the identifier is available in the intended region.

The model connection uses the credential in API Key as bearer authentication. Amazon Bedrock knowledge retrieval has separate credential fields and setup.

Reasoning Effort defaults to Default, which sends no explicit thinking budget. On the Claude Invoke path, None, Low, Medium, High and XHigh request budgets of 0, 1,024, 8,192, 16,384 and 24,576 tokens respectively. The Converse path does not apply this account control. Ask Powerfront to verify model compatibility before relying on it.

Settings

Setting Where Applies to Default Who can change it What it does
Region Settings › AI › Accounts website us-east-1 PowerfrontChooses the AWS region used to build the default model-service connection.
Reasoning Effort Settings › AI › Accounts website Not set: no explicit thinking budget PowerfrontRequests a thinking budget on the Claude Invoke path; the Converse path does not apply this account option.

Fields

Field Where Required What it is for
Provider Settings › AI › Models Yes Select AWS Bedrock when preparing a Bedrock model record.
Model Settings › AI › Accounts Yes Selects the configured Bedrock model or inference profile for the account.
API Key Settings › AI › Accounts Yes Supplies the credential sent as bearer authentication for model requests.

Limits and good to know

  • A non-empty API Key is required for the Bedrock model connection.
  • Models differ in accepted request formats, content and functions. A catalogue listing alone does not establish compatibility.
  • Certain provider output-limit stops are recorded as incomplete replies.
  • A region setting selects the connection region. It is not, by itself, a guarantee about the routing of a cross-region inference profile.
  • This release does not establish automatic model/profile region validation. Powerfront needs to check the selected profile and region.

Depends on and works with

RequiresConfigured byEnhances
View all connections as a list

This feature uses

Used by

FAQ and troubleshooting

Why does the connection use the wrong region?

An unset Region falls back to us-east-1. Ask Powerfront to check the region and model or profile together.

Can I use my knowledge-base AWS key pair in the model account?

The setups use different authentication fields. The model account sends API Key as bearer authentication; knowledge retrieval has separate AWS credentials.

Why did INSIDE try a second request mode?

A recognised method-compatibility error can cause a Converse-to-Invoke or Invoke-to-Converse retry. It does not change the selected model.

Why did Reasoning Effort not change the reply?

The Converse path does not apply this setting. The Claude Invoke path sends a budget, which the selected model must support.

INSIDE 3.3 · Source commit 992abb980bContent is a pre-release draft© 2026 Powerfront