Amazon CloudWatch Omni

Amazon CloudWatch Omni is AWS's AI-first observability experience, generally available since September 22, 2026 in US East (N. Virginia), US West (Oregon), and Europe (Ireland). It is an evolution of CloudWatch rather than a separate product: the same ingestion, storage, and query substrate, reorganized around teams and the applications and agents they run, with a natural-language layer on top. AWS frames the problem it targets directly, which is the case where an agent completes a task without throwing a conventional error, so latency, availability, and error-rate dashboards all stay green while the answer is still wrong.

Key features

• Spaces as the unit of organization: you create a space in a central account to see telemetry across AWS accounts and regions, plus other clouds including Azure workloads. The first centralized copy is free; each additional copy costs $0.05/GB.
• OpenTelemetry first: Omni keeps CloudWatch's OpenTelemetry interoperability, so logs, metrics, traces, and events arrive from any source rather than only from AWS services.
• Automatic discovery and dependency mapping: Omni discovers services, maps dependencies, and surfaces golden metrics, which is the part teams usually build by hand.
• Three ways to investigate: ask a question in natural language, click through guided console views, or drive it from your own tooling through the Agent Toolkit for AWS. Root-cause help is powered by AWS DevOps Agent.
• Agent evaluation built in: a dedicated agent observability experience runs an evaluation-driven development loop across LangGraph, CrewAI, OpenAI Agents SDK, Vercel AI SDK, and Strands, so you can score every prompt, model call, and tool invocation and validate a fix before shipping it.
• Local-first option: the CloudWatch Omni extension for VS Code, Cursor, and Kiro instruments, debugs, and evaluates agents on your machine with no AWS account required, which is unusual for an AWS product and makes it usable before you commit to the service.

Who should use it

• Platform and SRE teams already on CloudWatch who need agent behaviour in the same investigation surface as the infrastructure underneath it.
• Agent developers who want traces and evals without standing up and operating a third-party observability stack.
• Multi-cloud teams that want one telemetry view across AWS and Azure rather than two consoles.

Problems it solves

1. Silent agent failures: evaluation, not just error rates, becomes the signal that catches a bad tool choice or stale retrieval.
2. Tool sprawl: traces, metrics, logs, and agent evals share one space instead of four dashboards.
3. Instrumentation cost of entry: the local extension means you can start without an AWS account or a data pipeline.

Pricing

Omni bills on three dimensions, at standard CloudWatch rates for the first two. Prices below are for US East (N. Virginia), and AWS includes $1,000 in free credits to start.

Cost note: the free local extension is genuinely free, but every agent evaluation you run in the cloud is query volume. Prototype locally, then watch the analyze line.

Alternatives

If you are not on AWS, Langfuse, LangSmith, Helicone, and AgentOps cover much of the same agent tracing and evaluation ground. Omni's advantage is not features, it is proximity: if your infrastructure already lives in CloudWatch, the agent data lands next to it with no new vendor.