Streaming operations teams already know the kinds of questions they need to answer:
“What’s our cache hit rate for the West Coast edge locations over the last 72 hours, and how does that compare to last month?”
“Which origin servers are showing elevated latency right now, and which content categories are most affected?”
“Where are we seeing the worst rebuffering rates today, and is there a geographic pattern?”
With tools like MCP, operators can easily make these kinds of natural language queries. But the accuracy of the answers depends on the quality of the data.
CDNs generate data at a volume that traditional platforms can’t handle, leaving teams to work with sampled data and short retention periods. Even when teams have full-fidelity data, answering the questions has traditionally required writing SQL against complex schemas.
By bringing an agentic approach to full-fidelity CDN data, you can answer questions about your data much faster. Uriel Ramirez, Senior Solutions Architect at AWS, and Osmar Bento da Silva Junior, Principal M&E Solutions Architect at AWS, have built a reference solution that combines Hydrolix CDN Insights with Amazon Bedrock AgentCore and the Strands Agents SDK. This approach adds a natural language interface over full-fidelity data in Hydrolix, so the people who need answers can get them quickly even when they aren’t technical operators.
This post will cover:
- Combining CDN data with agentic AI
- Using Amazon Bedrock Agentcore with CDN Insights
- Translating CDN data to operational intelligence
Combining CDN Data With Agentic AI
Agentic AI fundamentally changes how teams interact with their data. Rather than writing SQL or waiting on analysts, anyone can ask questions in plain language and get answers drawn directly from the underlying data. However, an agentic approach will only provide useful, accurate answers if the underlying data is reliable.
CDNs generate massive volumes of log data which can easily run to terabytes of data per day and hundreds of terabytes of data for a major event like the Super Bowl. Traditional observability and analytics platforms have challenges with the scale and cost of this data, forcing operators to heavily sample or aggregate data and shorten retention windows.
AI tools working against sampled or aggregated data may be unable to answer operator questions, or even provide convincing but inaccurate answers that make it more challenging to troubleshoot issues and get the insights you need.
Ultimately, full-fidelity data is a basic requirement for accurate answers. But as data volumes grow, it becomes harder to retain it cost-effectively and to extract valuable insights. Hydrolix handles the ingest and storage piece, allowing you to keep petabyte-scale data long-term—and immediately available for querying regardless of age—at a fraction of the cost of traditional platforms. Meanwhile, an agentic approach handles the challenge of extracting insight from that data.
By combining long-range CDN data with agentic AI, you can ask questions that go well beyond performance metrics over the last 24 hours. The same logs that surface cache hit rates and rebuffering events also encode audience behavior, geographic demand patterns, content consumption trends, and anomalies that can indicate security threats or unauthorized distribution.
The following image from the repository shows how you can ask a question about your CDN data and get a visualization and summary. In this case, the question is “give me a full health check,” which results in a visualization of average bitrate by country with a natural language summary included.
Now imagine being able to ask any question of your data over a span of years. You can understand long-term trends and surface patterns on traffic, performance, and security that would’ve been challenging or even impossible to uncover before.
Check out the GitHub repository Deploying Hydrolix CDN Insights with Amazon Bedrock AgentCore.
Using AgentCore With Hydrolix CDN Insights
Building a system that operates reliably at enterprise scale, with proper session isolation, access controls, memory persistence, and observability, requires substantial infrastructure work. Which teams own that work, and do they even have the bandwidth to build out these systems?
Amazon Bedrock AgentCore addresses the infrastructure layer that makes agents production-ready. It provides a managed execution environment with complete session isolation, support for long-running workloads, fine-grained access policies, and monitoring through Amazon CloudWatch. Developers connect to it through standard frameworks. The CDN Insights with AgentCore solution uses the Strands Agents SDK, and AgentCore handles the operational complexity underneath. With this solution, you know longer have to worry about building a complete system from scratch.
Rather than a single general-purpose agent, the system uses a multi-agent approach with specialized subagents, each with its own domain expertise:
- A General Data Analyst agent for broad traffic exploration and ad-hoc queries.
- A CDN Infrastructure Expert specialized in cache hit/miss analysis, origin server latency, error rates, and edge location (POP) performance.
- A Viewer Experience Expert focused on Quality of Experience metrics using CMCD player telemetry, including buffer starvation, bitrate adaptation, throughput, startup performance, and geographic QoE breakdowns.
An orchestrator routes each user question to the appropriate specialist, which then uses the Hydrolix MCP Server to execute SQL queries and formulate answers. Results are returned as natural language responses, tabular data, and automatically generated chart visualizations.
The system also maintains memory across sessions using AgentCore Memory, so context from earlier in a conversation, or from previous sessions, informs subsequent answers. This turns a one-off query tool into something closer to an operational knowledge layer, all without additional engineering and IT work from your teams.
From CDN Data to Operational Intelligence
The CDN Insights with AgentCore reference solution is available on GitHub, with deployment instructions for both the Amazon Bedrock AgentCore backend (via AWS CDK) and the React web application frontend (via AWS Amplify). It’s designed as a ready-to-deploy starting point that can be extended with additional specialized subagents as your use cases grow.
If your team is already storing CDN or other data in Hydrolix, you can quickly add in an agentic intelligence layer with AgentCore. If you’re evaluating whether Hydrolix is the right foundation for your CDN analytics needs, this reference solution illustrates exactly what becomes possible when you combine a data platform built for massive scale and long-term retention with a modern agentic AI stack.
You can get answers to CDN performance questions in seconds, in plain language, making it easier to detect patterns in cache behavior, viewer experience, origin health, and traffic anomalies before they escalate. And you can also get a deeper understanding of how end users are engaging with your content over time, along with insights on bot traffic, performance, and security.
For engineering leaders, faster incident resolution means lower mean time to recovery and fewer escalations. For product and content teams, better QoE visibility translates to viewer retention. For security and operations, you can use long-term data to detect unauthorized distribution, bot traffic, and other threats.
Next Steps
- Check out the repository Deploying Hydrolix CDN Insights with Amazon Bedrock AgentCore.
- Learn more about CDN Insights.
- Get started with Hydrolix in AWS Marketplace.
