Skip to main content
AWS Bedrock Runtime integrations for Ona Agent are deprecated and maintenance-only. Do not create new integrations.Existing Enterprise integrations on AWS EC2 Runners keep working. You can update the model ID or token limit, or enable, disable, or remove the integration.AWS Bedrock Mantle is not deprecated. It provides OpenAI model access for Codex Agent.
This page documents existing IAM-based AWS Bedrock Runtime integrations that use bedrock://<model-id> endpoints to provide Anthropic Claude models to Ona Agent.

​
Requirements for existing integrations

  • An AWS account with Bedrock enabled and access to the configured model.
  • An Ona Enterprise organization.
  • An AWS EC2 Runner in the same AWS account with IAM permissions to invoke Bedrock.

​
Network access (VPC endpoints)

If your runner operates in a private subnet without internet egress, you need a VPC endpoint for the Bedrock Runtime service. Ona only calls the Bedrock Runtime API (bedrock-runtime). No other Bedrock endpoints are required. Create an interface VPC endpoint for com.amazonaws.<region>.bedrock-runtime in the VPC where your runner is deployed. See the AWS documentation on Bedrock VPC endpoints for setup instructions.

​
Maintain AWS permissions and model access

  • Keep access to the configured Claude model enabled in the AWS Bedrock console.
  • Keep the runner’s IAM permissions to invoke the Bedrock Runtime API.
  • Keep the configured model available in the runner’s AWS region.
See the AWS Bedrock Model Access documentation for access requirements.

​
Check the Bedrock endpoint

​
Endpoint format

Check the configured model ID against the AWS Bedrock model names for the runner’s region.

​
Verify an existing integration

  1. Create a new environment with the configured runner.
  2. Open Ona Agent and start a session.
  3. Test the integration with a code generation request.

​
Supported models

Existing integrations support Claude model presets including Claude Opus 4.8 and Claude Sonnet 5 on AWS Bedrock.

​
Identifying available models

There are two ways to identify models: foundation models and inference profiles. Inference profiles are resources that define a model and one or more regions for routing requests, enabling cross-region inference, usage tracking, and cost monitoring. Availability varies by region. Some regions only support foundation models, while others support both foundation models and inference profiles. Check model availability per region in the AWS documentation. You can check availability using AWS CLI commands from your environment, assuming you have proper authentication and region environment variables set:

​
Testing model connectivity

To run a simple smoke test and verify a model works:
If the smoke test succeeds, the model should work with Ona Agent.

​
Get help with an existing integration

If you encounter issues:
  1. Check Ona Agent logs for detailed error messages.
  2. Verify AWS quotas and Bedrock model access.
  3. Contact your account manager for additional support.

​
Troubleshooting

Use a model ID that is available in your runner’s region. For regionalized models, use the correct prefix, such as us.anthropic.*.
Ensure the runner role has bedrock:InvokeModel and bedrock:InvokeModelWithResponseStream. Confirm model access is approved in the Bedrock console.
Use bedrock://<model-id> and include the full version, such as ...-v1:0.
Lower the maximum tokens with gitpod runner config llm-integration set-max-tokens .... Review Bedrock service quotas and request increases if needed.