Amazon CloudWatch Omni explained: AI-assisted observability for applications and AI agents

In AWS's words, Omni is "an evolution of Amazon CloudWatch" and "an AI-powered observability experience organized around your teams and the applications they run". It combines OpenTelemetry interoperability with the scale of CloudWatch.

Omni is delivered outside the AWS Management Console. Teams sign in to a standalone web experience with the identities they already manage, through single sign-on. Agent developers can also work in a free IDE extension. AWS's product page describes the aim as unifying monitoring of AI agents, applications and infrastructure, with the advice to "instrument once via OTLP", the OpenTelemetry Protocol.

The CloudWatch User Guide is direct: "Omni is an interface and set of workflows built on Amazon CloudWatch. It is not a replacement, and enabling it does not change anything you already run."

Three terms shape any Omni design.

• Domain. The sign-in point for your organisation, with its own URL. Your identity provider connects at the domain, and one domain can hold several spaces.
• Space. A space shows telemetry from the AWS account and Region where it is hosted.
• CloudWatch Dataset. This makes logs, metrics and traces queryable and correlated inside Omni.

The practical point: "Omni does not aggregate telemetry across accounts and Regions on its own." To see several accounts or Regions in one space, use CloudWatch centralization rules to bring telemetry into a central account first. A space's Region must also match your IAM Identity Center Region, or use multi-Region replication.

For a mid-sized estate, that means deciding the central account and Region before you create spaces.

• Application map. Omni discovers services and their dependencies and reports RED metrics: request rate, errors and duration.
• The Omni agent. The built-in AI assistant takes plain-language questions and writes the SQL or PromQL to answer them.
• Investigations. Omni works with the AWS DevOps Agent to investigate incidents and find the root cause.
• Dashboards, threads and alerts. Build views, keep investigation threads, and send alerts to Slack or Amazon SNS.
• Skills. Call Omni from your own code or from a coding agent.

Traditional signals miss a key failure. As the User Guide puts it, "an agent can return a well-formed response that is wrong." Error rates and latency will not show that.

Omni puts agent tracing, evaluations, datasets and experiments in the same space as your application telemetry:

• Traces, spans and sessions for each prompt, model call and tool invocation.
• Evaluators that run continuously in production and on demand against datasets, with scores shown alongside operational signals.
• Your own evaluators, including criteria scored by a judge model.
• Experiments to validate a fix before you ship it.

AWS names LangGraph, CrewAI, OpenAI Agents SDK, Vercel AI SDK and Strands among the supported frameworks. The free IDE extension for VS Code, Cursor and Kiro lets developers instrument, debug and evaluate agents locally, with no AWS account required. Setup steps are in the User Guide under Monitor AI agents.

Observability is half the picture for agents in production. The other half is what they can access; our guide to securing AI agents' secrets and IAM on AWS covers that.

AWS's 19 October announcement says spaces in central accounts can see telemetry "across your AWS accounts, Regions, and other clouds, including Azure and GCP".

The User Guide page "Send telemetry to CloudWatch Omni" lists these application telemetry environments: Amazon EC2, Amazon ECS, Amazon EKS and AWS Lambda, plus Azure VM, Azure Kubernetes Service (AKS), Google Compute Engine and Google Kubernetes Engine (GKE). Telemetry arrives through OTLP endpoints, and workloads outside AWS federate for temporary credentials.

One limit to note: the AI agent telemetry environments in the same guide are on AWS only, namely Amazon Bedrock AgentCore, Lambda, EC2, ECS and EKS. If your agents run on Azure or Google Cloud, check the current documentation before you plan around them.

Omni launched in three Regions on 23 September 2026 and reached eight more Regions on 19 October 2026, for 11 in total:

If your organisation has a policy on where operational data is processed, an in-Region space keeps Omni next to your telemetry. For the wider question, see our guide to data residency on AWS.

The Amazon CloudWatch Omni pricing page sets out three dimensions:

• Ingest. Telemetry sent to CloudWatch is charged per GB at standard CloudWatch rates.
• Store. Retained data is charged per GB-month, as in CloudWatch. Omni works on the same stored telemetry, not a second copy.
• Analyze. Log and trace queries are charged per GB scanned above an included allowance of five times your monthly log and span ingestion. For PromQL query pricing, see the pricing page.

Dashboards and alerts are included within service quotas. Agent evaluations are charged at Amazon Bedrock AgentCore Evaluations rates. If you centralize telemetry from several accounts or Regions, the first centralized copy is free.

AWS offers $1,000 in free credits for 30 days for OpenTelemetry ingestion, for up to 10 accounts per AWS Organization. Not all accounts are eligible; the console shows eligibility when you enable Omni.

AWS lists prices for US East (N. Virginia) and notes that ingestion pricing varies by Region. Check the pricing page for your Region before you estimate.

• Try the playground. A shared environment with sample data needs no AWS account and no setup.
• Set up a domain and space. You need an AWS account and permission to enable Omni and create a domain and space in the CloudWatch console. Then configure single sign-on.
• Send telemetry. Existing CloudWatch telemetry is already there; add OTLP sources for new workloads.
• Install the IDE extension if you build AI agents.

• Start where you already have CloudWatch telemetry. Omni reads what you send today, so a team with good CloudWatch coverage gets value fastest.
• Decide the central account and Region first, then set up centralization rules.
• Instrument one AI agent with OpenTelemetry and put evaluation scores next to its error rates.
• Keep the fundamentals in place. Omni builds on metrics, logs, alarms and retention you already run. Our guide to an AWS-native observability baseline covers those.

As an AWS Select Tier Services Partner, Upvantix offers consulting on observability design: baseline reviews, alarm and runbook design, and deliberate decisions about managed or third-party tooling. Through managed AWS operations, we also work on AWS monitoring, telemetry, event routing, dashboards and alert-quality improvement.

Talk to us about your observability baseline and where Omni fits.