AI Explain (Claude on Amazon Bedrock): set up and use
Explain an alarm with Claude on Amazon Bedrock in your own AWS account, and set up AI for the Ask CloudConsole chat. Pro; about a cent per Explain, billed by AWS to your account.
On an alarm that isn't OK, tap Explain this alarm. CloudConsole gathers the evidence, sends it to a Claude model on Amazon Bedrock in your own AWS account, and shows what happened, the likely cause with a confidence level, the evidence it is based on and suggested next steps. The same setup powers Ask CloudConsole, the AI chat behind the sparkle button, for questions such as “Why is my bill high?”.


Where it runs
The request is signed on your phone with the account's own credentials and goes straight to Bedrock in the region you choose. There is no CloudConcur server and no third-party AI service in between, and Bedrock doesn't store prompts or train on them by default.
- Sent: resource names, alarm and metric values, the list of recent changes, and your question.
- Never sent: credentials, secrets, environment variable values, log lines or object contents.
What Explain looks at
The alarm and its history, the metric over the last 90 minutes, What changed? (CloudTrail writes and deploys), and, depending on the alarm's dimensions, the Lambda function, Auto Scaling group, target group health or ElastiCache events behind it. Each source has a few seconds; one that fails is shown as “not checked” and the rest still count. Every evidence chip opens the screen it came from.
Explain never changes anything. Its next steps open the right screen (for example a rollback, which then goes through your usual guardrails) or mute the alarm on this phone. Continue in chat takes the alarm into Ask CloudConsole for follow-up questions.
Set it up
- In the AWS console, open Amazon Bedrock → Model access in the region you'll use and request access to the Claude model (Claude Sonnet 5.5 is the recommended one; Haiku 5.5 is fastest, Opus 5.5 deepest). It usually takes a few minutes.
- Give the role or user you sign in with these IAM actions:
{ "Effect": "Allow", "Action": [ "bedrock:InvokeModel", "bedrock:InvokeModelWithResponseStream", "bedrock:GetFoundationModelAvailability" ], "Resource": "*" }bedrock:InvokeModelis the request behind each Explain and each step of a chat,InvokeModelWithResponseStreamstreams answers, andGetFoundationModelAvailabilityis only for the access check in setup. You can narrowResourceto the model or inference profile ARNs. - In CloudConsole, open Account → Integrations → Explain & Ask (AI) (or tap Explain for the first time), pick the model and the Bedrock region, tap Test with a tiny request, then Turn on Explain.
By default requests stay in the region you pick. You can let Bedrock route them within the same geography or globally with a cross-Region inference profile, which some newer models need in some regions.
What it costs
Explain and the chat are part of Pro; in the demo they work without Pro, on sample data. The model use itself is billed by AWS to your account at Bedrock's prices: a typical Explain is about 4,000 tokens in and 500 out, about a cent with Claude Sonnet 5.5, billed by AWS to your account. Setup shows the estimate for the model you pick and a running count for the month. Nothing runs in the background: each answer is one tap.
Questions
Does my alarm data go to Anthropic or CloudConsole?
No. The phone calls Amazon Bedrock in your own AWS account with your credentials. Neither CloudConcur nor any other service sees the request.
Why does Explain say I don't have access to the model?
Bedrock needs model access per region. Open the Amazon Bedrock console in that region → Model access and request the Claude model, then try again.
Is “What changed?” still free?
Yes. The What changed? timeline is free; Explain, which reads it as evidence, needs Pro.
Email [email protected] with what you tried and any message the app showed. Don’t send access keys or secrets. We reply within two working days.