Civic Canary

Inspiration

Nonprofits and community organizations often depend on government and public-benefit websites for accurate information about programs, eligibility requirements, documents, and services. These websites change frequently, and manually checking them is repetitive, time-consuming, and easy to overlook.

We built Civic Canary to automate that monitoring so teams only need to step in when an important change requires human judgment.

What it does

Civic Canary is an autonomous AI agent that monitors public-benefit websites for meaningful changes in policies, content, documents, links, and accessibility.

Instead of sending alerts for every small webpage update, Civic Canary verifies the source, collects supporting evidence, analyzes whether the change matters, stores the finding, and only surfaces it when human review is needed.

This helps nonprofit and community teams spend less time checking websites and more time supporting the people they serve.

How we built it

We built Civic Canary using the Strands Agents SDK as the main agent orchestration layer.

The system uses:

Strands Agents SDK for agent reasoning and tool orchestration
Amazon Bedrock for AI-powered analysis
Amazon Bedrock AgentCore Browser to inspect websites and collect evidence
Amazon EC2 to run the application and background monitoring worker
Amazon DynamoDB to store monitored websites, findings, scan history, and review decisions
Amazon S3 to store evidence and monitoring artifacts
Amazon CloudFront to deliver the web application
React and FastAPI for the frontend and backend

The main workflow is:

Monitor → Detect → Verify → Analyze → Store Evidence → Human Review

The agent handles repetitive monitoring automatically while keeping final decisions with a human reviewer.

Challenges we ran into

One of the biggest challenges was identifying which website changes were actually important.

Government websites often change timestamps, page layouts, menus, or other dynamic content without changing the information people rely on.

We also had to make website verification reliable, collect clear evidence, handle failures safely, and avoid duplicate or unnecessary findings.

Another challenge was balancing automation with safety. Civic Canary can monitor, verify, and analyze changes automatically, but it does not publish corrections without human approval.

Accomplishments that we're proud of

We are proud that Civic Canary goes beyond a traditional chatbot and performs real work in the background.

The agent can continuously monitor public resources, detect meaningful changes, verify them, save supporting evidence, analyze their importance, and only involve a human when action is needed.

We also built a complete AWS-based system using the Strands Agents SDK, Amazon Bedrock, AgentCore Browser, EC2, DynamoDB, S3, and CloudFront while maintaining a human-in-the-loop review process.

What we learned

We learned that building a useful AI agent requires more than an LLM.

A reliable agent also needs tools, persistent data, verification, error handling, structured workflows, and clear rules for when a human should take control.

Using the Strands Agents SDK also showed us how AI reasoning can work with AWS services to build an agent that runs continuously in the background instead of waiting for a user prompt.

What's next for Civic Canary

Next, we want to support more types of public websites, improve change detection, strengthen accessibility checks, and add better notification options.

We also plan to improve scalability, monitoring, and observability, while exploring deeper use of AgentCore for production deployment.

Our long-term goal is to make Civic Canary a reliable background assistant for nonprofits and community organizations, quietly monitoring important public information and only surfacing changes that truly need human attention.