NowAware: An Agent That Notices What Matters
What if you didn't have to keep checking websites for important updates? NowAware is a personal AI agent designed to monitor selected public sources, identify meaningful changes, and explain when those changes require your attention.
NowAware: An Agent That Notices What Matters
The problem: Information changes, but people cannot watch everything
Important information is scattered across websites, public notices, announcements, documents, and feeds. A scholarship deadline changes. An official notice introduces a new requirement. A transport update affects a planned journey. A public announcement contains something a person has been waiting for.
The problem is not always finding information. Sometimes, the problem is knowing when information changes.
Most people repeatedly visit websites, refresh pages, or set reminders to check again later. Traditional alerts can help, but they often notify users about every update without explaining whether it matters.
I wanted to explore a different approach: an AI agent that understands what someone cares about, watches selected sources, and brings meaningful changes to their attention.
That idea became NowAware.
What is NowAware?
NowAware is a personal information-monitoring agent for students, families, professionals, commuters, and anyone who needs to keep track of changing information.
Rather than limiting the agent to a single industry, the user describes a goal in natural language.
For example:
- "Watch this scholarship announcement and tell me if the deadline changes."
- "Monitor official weather alerts for my selected area."
- "Track updates on this public notice and explain any new requirements."
- "Watch this announcement feed and tell me when a new opportunity matches my interests."
The agent translates the request into a monitoring rule, identifies the information to track, and checks the selected source according to a defined schedule.
When a potentially important change appears, it compares the new information with the previous version, assesses its relevance to the user's request, and prepares an explanation supported by the source.
The objective is not to generate more notifications. It is to make the notifications people receive more useful.
How I designed the agent
I structured NowAware around five stages.
1. Understand the user's goal.
The agent identifies what the user wants to monitor, which source contains the information, what kind of changes matter, and how the user wants to be notified. It confirms the monitoring rule before activating it.
2. Establish a baseline.
The system retrieves the selected public page or supported feed and stores a snapshot of the relevant information. This provides a reference for future comparisons.
3. Detect changes.
A scheduled task checks the source again and compares the latest content with the previous snapshot. The system separates content changes from formatting differences wherever practical.
4. Evaluate significance.
The agent uses the user's original goal and the detected differences to determine whether the change is worth reporting. It should distinguish an actual deadline change from an unrelated webpage redesign.
5. Explain and notify.
When a change meets the notification criteria, the agent provides a concise explanation, supporting source, relevant before-and-after details, and a suggested next step. If no meaningful change is detected, the system remains quiet.
How I built it with AWS
The prototype uses Python and the Strands Agents SDK to define the agent and its tools. I use Amazon Bedrock for model inference and natural-language interpretation.
The agent does not rely solely on the language model to remember what happened previously. The monitoring application maintains source snapshots and compares them with newly retrieved content.
Amazon EventBridge Scheduler invokes AWS Lambda functions on a defined schedule. Lambda performs the retrieval and comparison workflow, then invokes the agent when a potentially meaningful change needs interpretation.
Amazon DynamoDB stores monitoring rules, source snapshots, and notification history. Amazon SES delivers email notifications. A Streamlit interface provides the first-run experience, monitoring configuration, and a history of detected changes.
The main components have separate responsibilities: the agent interprets the user's intent, scheduled functions observe the sources, persistent storage maintains state, and the notification layer delivers the result.
For implementation details, I used the official documentation for the Strands Agents SDK , Amazon Bedrock , and Amazon EventBridge Scheduler .
The delightful detail: Quiet by default
The most important design decision is that NowAware should not treat every detected change as an important event.
Imagine a monitored page changes its layout, navigation, or formatting. That does not necessarily deserve a notification.
Now imagine the same page publishes a revised deadline or a new eligibility requirement. That may deserve immediate attention.
NowAware evaluates the change against the user's monitoring rule and explains why an alert was generated. The user can inspect the supporting source rather than having to trust an unexplained AI-generated message.
This design also creates a useful feedback loop: users can refine what they care about when an alert is irrelevant or when a meaningful change is missed.
How I evaluated it
I tested the monitoring workflow using controlled changes to sample public information pages and feeds.
The evaluation covered three scenarios:
- A meaningful change to a tracked value, such as a published deadline.
- An irrelevant change, such as a formatting modification.
- A retrieval failure or ambiguous change where the system should avoid presenting an unverified conclusion as fact.
For each case, I checked whether the system detected the change, used the previous snapshot correctly, produced a source-supported explanation, and handled errors without generating misleading notifications.
Actual test results: [Insert the number of successful tests, total scenarios tested, and any failures observed.]
The purpose of this evaluation is to measure whether the agent can distinguish meaningful changes from noise, rather than simply whether it can generate a convincing explanation.
Limitations and responsible design
NowAware is not an instantaneous monitor of the entire internet. Its detection speed depends on the configured checking interval, source availability, and retrieval reliability.
The initial prototype supports only the public URLs and feeds implemented for the demonstration. Source content is treated as untrusted data, and the agent must not follow instructions embedded in a webpage as though they were system instructions.
The application should also restrict allowed destinations, avoid private or authenticated pages in the initial prototype, minimize stored information, and provide transparent source links. For high-impact information, the user should verify important details against the original source before acting.
These constraints are part of making the agent dependable, not simply adding features.
Try NowAware
NowAware explores a simple idea: people should not need to repeatedly check every source that matters to them. They should be able to explain what they care about, let an agent monitor it, and receive a clear explanation when something genuinely changes.
The goal is not to know everything. It is to notice the things that matter to each person.
#agents
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article