Use an AWS key that already has the intended Bedrock access when you want to manage that credential yourself. This guide follows Option B in the module’s Connect AWS panel.

Prepare the AWS connection

Have your AWS administrator prepare the access key, secret and intended source region. Keep the secret private. The form takes an access-key pair and region; it does not provide a session-token field.

Ask the administrator to allow the selected model’s invocation and inference-profile access. A profile can route across regions and needs the relevant permissions. See AWS’s inference prerequisites and profile permissions. Do not grant unrelated account-wide access as a troubleshooting shortcut.

Enter the existing key

Open the PD Modules plugin in your staff area, choose PD AI Assistant, then open Settings → Provider & connection. Note the active provider and model before changing them. Enabled staff, customer and automated assistant features share the chosen engine.

1. Select Your own Amazon Bedrock (your AWS account) in Choose a provider. Viewing the form does not activate it.
2. Review the displayed Model. The module offers Nova Micro, Nova Lite and Nova Pro using US inference profiles; check that your intended source region supports the selected profile.
3. In Connect AWS, find Option B — paste a key that already has Bedrock access.
4. Enter AWS Access Key ID, AWS Secret Access Key and AWS region.
5. Click Use this key and check the connection summary. If a connection is already present, the panel shows its summary and Disconnect instead of the entry form; do not disconnect it merely to explore the fields.
6. In the provider section, use Save & use Amazon Bedrock or Save Bedrock settings, as displayed, to save the model and select Bedrock as the active engine.

Confirm a reply

Check the active provider summary, then open Assistant and request a short, non-sensitive answer without a customer lookup or action. Saving credentials does not itself run an inference request. A Connected label therefore needs this separate functional check.

If the request is denied, check the selected region/profile and ask your AWS administrator to review the rejected operation. Do not keep replacing valid credentials when the error concerns model access or a region restriction.

Check Knowledge separately

Custom Knowledge needs embedding access in addition to chat. The module uses Amazon Titan Text Embeddings V2 for this, so Nova access alone is insufficient. Use the Bedrock Knowledge checks if replies work but indexing or reference retrieval fails.

Retain the model, source region, original attempt time and redacted error while your hosting team checks the failed operation. Do not send the access key, secret or private prompt.

See the PD AI Assistant Blesta guide for related tasks, or explore PD AI Assistant.