Answers before you connect an AWS account.

BedrockOps is the operational intelligence layer for Amazon Bedrock production. It correlates runtime failures, Knowledge Base events, prompt changes, model changes and deployments into one ranked cause list, and it gives Knowledge Base changes a release process with candidate versus live comparison, audit-logged promotion and rollback.

No. BedrockOps is built for explanation, not for raw dashboarding. CloudWatch generative AI observability is the right place for Bedrock metrics, traces and token usage. BedrockOps consumes that layer and adds attribution, cross-signal correlation, ranked cause and Knowledge Base release control.

Quota burn attributed to a caller and workload, throttling classification for 429 ThrottlingException and 503 ServiceUnavailable, TPM versus RPM pressure, retry amplification, connection pool saturation, throttle root-cause classification with per-event runbook guidance, runtime readiness checks, and agent tool-use correctness scoring.

Ingestion failure intelligence, retrieval configuration validation, retrieval quality tracking over time, tenant isolation checks, candidate versus live comparison, and audit-logged promotion and rollback.

Yes. BedrockOps runs the candidate and the live Knowledge Base against the same probe set and reports retrieval-quality scores, zero-result rate and tenant isolation results side by side. Promotion is recorded with an owner, a timestamp and the comparison evidence, and the previous Knowledge Base stays available for rollback.

BedrockOps attributes tokens-per-minute consumption to the caller identity and workload behind each InvokeModel call, then ranks callers by burn over the selected window. CloudWatch reports EstimatedTPMQuotaUsage for the account and model; BedrockOps reports which application, role or workload produced it.

A 429 ThrottlingException means your request exceeded a quota you own, usually tokens or requests per minute for that model and region. A 503 ServiceUnavailable means the service side could not serve the request. The remediation differs: 429 means fix client concurrency, spread load with cross-region inference, or raise the quota; 503 means back off, retry with jitter, and fail over.

A retry storm happens when a client SDK retries throttled requests aggressively, so each throttled call generates several more, which consume more quota and produce more throttling. The signature is a retry-to-request ratio well above 1 concentrated in one caller while the underlying request rate is flat. BedrockOps measures that ratio per caller and flags amplification.

Ingestion jobs usually fail on a subset of documents rather than the whole data source: unsupported formats, oversized files, encoding problems, permission errors on the S3 prefix, or embedding calls that were themselves throttled. BedrockOps groups failed documents by failure reason and links the ingestion job to the retrieval quality change it caused.

A cross-account IAM role with read-only permissions, assumed by BedrockOps with an external ID. It reads Bedrock runtime and Knowledge Base signals, quota and throttling data, and deployment events. There are no write permissions and no changes to your Bedrock configuration.

No. There is no in-path proxy and no traffic reroute. BedrockOps reads signals from your account and never sits in the request path of a production Bedrock call.

Customer operational data stays in Europe today, and the product is delivered entirely on AWS. Broader regional coverage will come in later versions.

It changes what the security review is about. Instead of assessing a new dependency in the production request path, the team reviews a familiar cross-account role with an external ID and no write permissions.

A 7-day trial is $0. Production is $199 per month, or $159 per month billed annually, and includes 3 AWS accounts and 10 Knowledge Bases. Overage is $49 per extra account and $10 per extra Knowledge Base. Enterprise pricing is custom — contact us. Everything starts and bills through AWS Marketplace, with a customer-set monthly spend cap.

1 AWS account, 3 monitored Knowledge Bases, 7-day retention, 3 alert rules and 1 email destination, at $0 for 7 days. When the trial ends without a paid subscription, the workspace moves to read-only rather than billing automatically.

As one Knowledge Base, regardless of how many data sources it has. Billing grows with the number of AWS accounts and Knowledge Bases connected, not with telemetry volume.

Yes. A customer-set monthly spend cap, an in-app spend projection at 50%, an email warning to admins at 80%, protected mode at 100%, and opt-in auto-expand. BedrockOps never stores payment details; billing runs through AWS Marketplace.