Clear answers for teams evaluating BedrockOps.

BedrockOps is the operational intelligence layer for Amazon Bedrock production. It gives teams one workspace for root-cause analysis across runtime reliability, Knowledge Base operations, and cross-signal incident triage — correlating runtime, retrieval, prompt, and deployment signals into ranked probable causes for quality drops and incidents.

No. BedrockOps is built for explanation and investigation. It focuses on ranked root cause, runtime analysis, Knowledge Base debugging, and cross-pillar incident triage rather than raw dashboarding alone.

It covers quota burn analysis, throttling and failure classification, retry amplification, connection-pool saturation, runtime readiness checks, and agent tool-use correctness.

It covers ingestion-failure intelligence, retrieval-configuration validation, retrieval-quality scorecards, tenant-isolation checks, and candidate-vs-live comparison with audit-logged promotion and rollback visibility.

BedrockOps is not a Bedrock proxy — no requests are routed through it. It is not a vector database — no embeddings are built or stored. It is not a reranker-training tool — no models are trained. It does not replace Bedrock features — Knowledge Bases, AgentCore, Guardrails, and Bedrock Evaluations remain the customer’s primitives, and BedrockOps operates on top of them. It is Amazon Bedrock only in V1, not a multi-LLM-provider product. And it is not a benchmark-aggregation product — customer data is never aggregated across tenants.

BedrockOps is delivered entirely on AWS. All SaaS-side compute and storage run in eu-west-3 (Paris).

Customer operational data stays in Europe today. Broader regional coverage is planned for later versions, but Europe is the active region now.

No. BedrockOps does not require an in-path proxy. The default connection model is a read-only cross-account IAM role, scoped with a per-tenant External ID and verified with a confused-deputy check after every role assumption.

Collection is sharded by activity: a 1-min cadence for active accounts, 5-min for quiet accounts, and 30-min for dormant accounts. The read-only role reads Bedrock invocation logs, Bedrock quota metrics, Bedrock Knowledge Base state and ingestion logs, CloudTrail events, and inference-profile and model-availability listings. Data flows in one direction only: from the customer account to BedrockOps.

It lowers security review friction and helps teams adopt the product without changing live request routing in production. BedrockOps reads, interprets, and recommends — it never proxies a Bedrock call.

BedrockOps starts and bills through AWS Marketplace. There is no direct web checkout flow.

The 7-day trial includes 1 AWS account, 3 monitored Knowledge Bases, 7-day retention, 3 alert rules, and 1 email destination. See the pricing page for the full plan entitlements.

Pricing expands with additional AWS accounts ($49 / account / month) and additional Knowledge Bases ($10 / KB / month), billed through AWS Marketplace.

Yes. BedrockOps bills through AWS Marketplace only — there is no direct web checkout — and every Production-plan customer can set a monthly spend cap. At 50% of cap an in-app projection of end-of-period spend is shown, at 80% admins receive an email warning, and at 100% protected mode activates: high-volume time-series ingestion pauses for the remainder of the period while dashboards, incident history, alerts, and Knowledge Base state writes stay fully operational. Caps can be raised at any time and take effect immediately.