Amazon Nova Act: Open-Source Price Tracking Agent on GitHub

Amazon just released Nova Act — an open-source browser automation SDK that replaces entire teams of manual price monitors using AI automation. If your company employs even 5 people to track competitor prices daily, you're likely spending $250K–$500K per year on work that can now run unattended, around the clock. Available now on GitHub, Nova Act is AWS's most practical open-source AI agent yet.

The real cost isn't just salaries. Manually tracked prices are always stale — competitor pricing changes hourly, and every decision made on outdated data translates directly into lost revenue or missed margin. Nova Act is AWS's open-source answer to this problem. To get started with AI automation tools like this, visit the AI for Automation setup guide.

Why Traditional Price Scripts Break — and Cost You More

Legacy automation tools like Selenium or rules-based RPA (Robotic Process Automation — software that mimics human clicks and form fills using fixed, hardcoded instructions) rely on "element selectors" — exact references to specific positions on a webpage. The moment a retailer redesigns their product page, rotates a banner, or launches a flash sale, those selectors break silently. Your data stops updating, and nobody notices until pricing decisions go wrong.

Nova Act uses natural language commands instead. You describe what you want — "find the current price of Product X" — and the agent figures out how to retrieve it, even if the page layout changed overnight. This makes agents dramatically more resilient to the real-world messiness of modern ecommerce sites: short-lived promotions, rotating hero banners, and dynamic JavaScript content that would shatter a rigid script.

10 Parallel Browser Sessions: What AI Automation Delivers for Your Team

The standout capability is parallel execution (running multiple tasks simultaneously). Using Python's ThreadPoolExecutor (a built-in Python tool for running multiple tasks at the same time across processor cores), Nova Act can run up to 10 concurrent browser sessions — each crawling a different competitor page at the exact same moment.

In practice, a task that used to occupy a 3-person team for half a workday now completes in minutes. The architecture follows a map-reduce pattern (split one big job into many smaller ones, run them in parallel, then combine results into a single report), distributing crawl work across lightweight instances that merge into one structured, validated dataset.

• Navigation: Agents browse multi-step pages using plain English instructions — no hardcoded element paths required
• Extraction: Pydantic models (Python data structures that enforce strict data types and validate input) define exactly what to pull — price, currency, product name, availability status
• Error handling: ActErrors surface problems explicitly, enabling retry logic, fallback flows, and detailed logging instead of silent data gaps that corrupt your pricing model

4 Industries Where AI Automation Kills Manual Price Monitoring

AWS identifies at least 4 industries where teams routinely lose hours to manual monitoring every single day:

• E-commerce: Product pricing, promotional tracking, inventory availability across dozens of retailers simultaneously
• Insurance: Premium comparisons, policy coverage differences across competing carriers
• Financial services: Loan rates, credit card offers, fee structures across competing banks and fintechs
• Travel & hospitality: Flight prices, accommodation rates, seasonal pricing shifts that can change multiple times per day

The underlying workflow is identical across all four: humans copying numbers from websites into spreadsheets. Nova Act automates the browsing and extraction step entirely — leaving your team to act on insights, not gather them.

Running Your First Price Intelligence Agent — 5 Steps

Nova Act requires Python 3.10 or later and a free API key from nova.amazon.com. The SDK integrates with 3 major code editors — VS Code, Cursor, and Kiro — converting natural language prompts directly into working automation code inside your editor without switching tools.

# 1. Verify Python 3.10+
python --version

# 2. Install the SDK
pip install nova-act

# 3. Set your API key
export NOVA_ACT_API_KEY="your-api-key-here"

# 4. Clone the sample price comparison solution
git clone https://github.com/aws-samples/amazon-nova-samples.git
cd amazon-nova-samples/use-cases/price-comparison

# 5. Install dependencies and run
pip install -r requirements.txt
python price_intelligence_agent.py

The sample solution in the Amazon Nova Samples GitHub repository is production-oriented — not a toy demo. It includes parallel crawling logic with max_workers=10, typed Pydantic schemas for structured output, and full ActError handling out of the box.

What Nova Act Won't Do — Honest Limitations

Nova Act is powerful, but three constraints are worth knowing before you commit:

• Captchas require a human: AWS explicitly recommends human-in-the-loop (a human steps in manually to solve the puzzle) rather than automated captcha solving — for both reliability and ethical reasons. The SDK supports pausing execution for human intervention via AgentCore Browser Tool's UI takeover feature.
• One browser per instance: Each Nova Act instance controls exactly one browser. Reaching 10 parallel sessions means spinning up 10 separate instances — adding operational complexity for teams without Python experience.
• Intermediate Python required: This is not a no-code tool. Pydantic schemas (data validation structures), ThreadPoolExecutor (parallel processing), and custom error handling logic are all required to build production-grade agents that run reliably at scale.

Real-Time Debugging Without Context-Switching

For teams deploying across large product catalogs, a common pain point is diagnosing agent failures: when something breaks quietly, you're jumping between log files, screenshots, and code across multiple tabs. Nova Act's AWS Console integration solves this by surfacing detailed traces (step-by-step execution logs showing exactly what the agent did), live screenshots captured during the crawl, and S3-stored artifacts (saved result files in Amazon's cloud storage) — all in one dashboard, in real time.

You can explore the full codebase and deploy the sample agent today at the Nova Act GitHub repository. If your pricing team is still running manual checks in 2026, this is the week to change that — AWS's sample code takes under an hour to set up and run against real competitor URLs. For more automation workflows like this one, explore the step-by-step guides on AI for Automation.

Related Content — Get Started | Guides | More News