Someone on your team is probably checking a competitor’s pricing page again right now. Then Reddit. Then the careers page, in case there’s a hiring spike worth noting. Then a message to whoever manages the tracking spreadsheet.
This is how most companies still do competitive research, and it doesn’t scale. A person can only watch so many websites, forums, and job boards before something important slips through, usually right when it matters most.
RivalRadar was built to take this work off a person’s plate. Instead of analysts manually gathering intelligence, a set of AI agents does the watching, reads what they find using large language models (LLMs, AI systems trained to understand and generate text), remembers what they’ve learned over time, and turns it into a report someone can actually act on.
What Makes RivalRadar Different From a Typical Monitoring Dashboard
Most tools in this space are dashboards. They collect data from a few sources and display it, but a person still has to look at everything and decide what it means.
RivalRadar tries to do that second part too. It pulls signals from competitor websites, Reddit, Hacker News, RSS feeds, and hiring portals, then uses AI to filter out noise and figure out what’s actually worth flagging.
The bigger idea behind the platform is that a single signal rarely tells the whole story. A price change on a competitor’s website might not mean much by itself. But paired with a burst of new hiring, some frustrated comments on Reddit, and a recent changelog update, it can point to a real shift in that company’s strategy. RivalRadar connects signals like these using semantic search (matching content by meaning rather than exact keywords) and AI reasoning, rather than treating each one as an isolated event.
The Technology Behind RivalRadar
RivalRadar’s codebase is split into separate modules, one each for collecting data, enriching it with AI, storing it, coordinating the different pieces, and generating reports. Keeping these pieces separate makes the system easier to maintain and scale as more competitors or data sources get added.
The dashboard and interface run on Angular and PrimeNG, giving administrators and analysts a clean view of competitors, incoming signals, reports, users, and system settings.
The backend runs on ASP.NET Core Web API (Microsoft’s framework for building the services that power an application behind the scenes). The code is organized in layers, separating how it talks to the database from the business logic and the background jobs that run on a schedule. That separation keeps the system manageable as it grows.
For the AI itself, RivalRadar runs its language models locally using Ollama, a tool that lets AI models run directly on RivalRadar’s own servers instead of routing every request through an outside AI provider. This has a few practical benefits.
- Lower reliance on external AI services
- Lower running costs over time
- More control over how data is handled and stored
- Faster testing of new models
- No interruptions from external API outages or rate limits
The platform currently runs on Gemma4:e4b as its main reasoning model, nomic-embed-text to generate the embeddings (numeric representations of meaning, explained more below) used for search, and Qdrant as the database that stores those embeddings. Together, these let RivalRadar retrieve relevant context and reason about it entirely on infrastructure it controls.
A Real-Time View From the Dashboard
The dashboard is where most people using RivalRadar will spend their time day to day. It gives a real-time snapshot of what’s happening across every competitor being tracked.
At a glance, it shows the total number of signals collected, how many competitors are being monitored, how sentiment is trending, what each AI agent has been doing, how many reports have been generated, and how signals break down by source.
Charts summarize activity so nobody has to read through every individual signal to spot a trend. The sentiment view, in particular, makes it easy to tell whether the conversation around a competitor is turning more positive or more negative over time.
The dashboard also shows how the AI agents themselves are performing, including how many signals each one has generated recently, which makes it easier to notice if one source suddenly goes quiet or spikes.
Setting Up and Managing the Competitors You Want to Track
Before RivalRadar can watch anyone, someone has to tell it who to watch. Administrators add, edit, and manage competitors through a dedicated screen in the platform.
Each competitor record includes basic details such as company name, website, a short description, status, which agents are active for that company, and when the record was created.
One useful detail here is that agents can be turned on or off individually for each competitor. Not every company needs to be monitored the same way. Some don’t publish an RSS feed. Some don’t keep an active careers page. Some barely come up on Reddit. Rather than running every agent against every competitor regardless of fit, administrators can enable only the sources that actually apply.
New competitors also inherit a default set of agents based on settings configured for the whole workspace, so nobody has to set this up manually every time.
How the Five Monitoring Agents Work Together
At the center of RivalRadar is a set of five specialized AI agents, each responsible for one type of source.
- WebWatcher Agent, for competitor websites
- Reddit Agent, for community discussion
- HackerNews Agent, for developer and startup conversation
- JobBoard Agent, for hiring activity
- RSS Feed Agent, for blogs, changelogs, and release notes
Each agent runs independently of the others. If one runs into an error or a source goes down temporarily, the rest keep working without interruption.
Every agent follows the same basic sequence when it runs. It connects to its source and pulls in raw data. Then it cleans the data into a consistent format, checks whether it’s actually relevant to the competitor in question, analyzes the sentiment, and turns it into a structured signal that gets stored. It also generates an embedding for that signal (the numeric fingerprint that lets RivalRadar find it again later by meaning, not just keywords) and saves that to the platform’s long-term memory. The whole run is logged from start to finish, so there’s a record of what happened and when.
That shared process is what turns scattered, messy data from five very different types of sources into a consistent stream of structured intelligence.
The WebWatcher Agent Tracks Website Changes
The WebWatcher Agent keeps an eye on a competitor’s website for changes that actually matter, not every small edit. It watches high-value pages such as pricing, features, products, changelogs, careers, blog, and solutions pages.
Because many modern websites load their content dynamically, the agent uses Playwright, a browser automation tool. It runs a headless (invisible, with no on-screen window) version of Chromium so it can render pages the same way a person’s browser would, including single-page apps that build their content on the fly.
Once a page loads, RivalRadar cleans up the content and compares it against the version it saved last time. A change-percentage calculation decides whether the difference is meaningful enough to flag. Minor edits, like a typo fix or a tweaked button label, get ignored. A new pricing tier or a materially different feature list gets turned into a signal.
The agent also extracts different information depending on the page type. It pulls plans and pricing from pricing pages, job listings from careers pages, and announcements from blog pages. Every signal also gets an importance score, based on the type of page, how big the change was, and how strategically relevant it looks. Pricing and changelog updates tend to score higher, since they’re often the first visible sign of a bigger shift in a competitor’s direction.
The Reddit Agent Reads Community Sentiment
Community discussion is one of the most useful, and most overlooked, sources of competitive intelligence. The Reddit Agent is built around that idea.
It searches Reddit using several strategies at once, combining company names, product names, domain references, comparison phrases, and problem-related search terms. It also checks competitor-specific subreddits when they exist.
One of its more useful features is relevance checking. Company names often show up online in contexts that have nothing to do with the actual company. Instead of trusting keyword matches alone, RivalRadar uses an LLM to judge whether a given post genuinely relates to the competitor being tracked. That cuts down on the noise a keyword-only tool would collect.
For discussions that get real engagement, the agent also pulls top comments, upvotes, downvotes, and comment counts. Sentiment is analyzed by the AI in context rather than through simple keyword matching, which helps it pick up on sarcasm and more indirect ways people express frustration or approval.
The HackerNews Agent Watches the Developer World
Hacker News tends to surface certain kinds of news earlier than almost anywhere else. That includes pricing backlash, developer frustration, outages, new AI product launches, and side-by-side comparisons between competing tools.
The HackerNews Agent runs multiple search combinations and ranks what it finds by points, comment count, general activity, and how recent it is. This source tends to matter most for AI companies, developer tools, SaaS products, infrastructure providers, and API-based services, though it can surface something useful for almost any tech company.
Like the Reddit Agent, everything it finds goes through the same AI relevance check before it gets stored, so a company name showing up in an unrelated thread doesn’t get treated as a real signal.
The JobBoard Agent Reads Between the Lines of Hiring
Hiring activity often hints at where a company is heading before they say so publicly. The JobBoard Agent monitors listings on Greenhouse and Lever, two of the most widely used job board platforms. It uses browser automation to take snapshots of job postings and compares them against earlier snapshots over time.
It’s looking for new roles, expansion trends, growth in specific departments, geographic expansion, and hiring focused on AI. A jump in enterprise sales postings can hint at a push into a new market. A wave of AI engineering roles can point to where a company is putting its product investment. A spike in support hiring often means their customer base is growing faster than expected.
By comparing each new snapshot against the last one it saved, the agent can spot these shifts as they happen rather than waiting for someone to notice them manually.
The RSS Feed Agent Follows Product Announcements
Not every company advertises where its RSS feed lives. Rather than relying on a single, fixed URL, the RSS Feed Agent tries several common patterns, such as /rss, /feed, /blog/rss.xml, and /changelog/rss, to find feeds that aren’t obviously linked anywhere.
Each entry it finds gets turned into a structured signal with a title, a short summary, a link back to the source, a sentiment reading, and a timestamp. Over time, this builds a running stream of product announcements, release notes, and other updates without anyone having to check a competitor’s blog by hand.
Why RivalRadar Remembers What It Learns
One of the more advanced things RivalRadar does is remember. After each signal is collected, RivalRadar converts it into an embedding, essentially a numeric representation of what that piece of content means. It generates these using the nomic-embed-text model and stores them in a vector database called Qdrant, a database built specifically to search this kind of numeric data quickly.
This lets the system pull up related historical context, spot patterns that repeat over time, and search by meaning rather than exact wording. It also enables what’s called Retrieval-Augmented Generation, or RAG, where the AI looks up relevant past information before generating new analysis instead of relying purely on what happened this week.
In practice, that means a report isn’t just a summary of the last few days. If a competitor has raised prices twice in the past year, or if the same customer complaint keeps resurfacing, or if their engineering hiring has been climbing steadily, RivalRadar factors that history into what it generates rather than treating each week as a blank slate. The result reads more like ongoing analysis than a series of disconnected weekly snapshots.
How Raw Signals Become an Executive Report
Collecting signals is only useful if someone can act on them, which is where the reporting side of RivalRadar does most of its work.
A component called the Synthesis Agent pulls together the current signals, related historical context from vector memory, importance scores, and sentiment patterns. It feeds all of that into a structured prompt for the AI to work from. From there, RivalRadar automatically generates an executive summary, key insights, strategic implications, recommendations, a watchlist, and a set of priority actions.
Reports can be viewed on the dashboard, sent as formatted HTML emails, or posted to Slack, depending on how a team prefers to receive them. Each one uses priority labels, CRITICAL, HIGH, MEDIUM, and IMMEDIATE, so the most time-sensitive findings stand out instead of getting buried in a long list of updates.
Keeping the System Reliable
Running AI agents around the clock only works if someone can tell when something breaks. RivalRadar includes a dedicated observability module. It tracks when each agent ran, whether it completed successfully, how long it took, how many signals it produced, and any error messages along the way. That makes it straightforward for an administrator to spot and fix problems before they affect the intelligence being generated.
Because each agent runs independently, the underlying workload can be spread across multiple workers or servers as monitoring needs grow. Planned improvements include distributed task queues, Kubernetes-based orchestration (a system for managing containerized applications across many servers), real-time data streaming, and more advanced scheduling for the agents.
The Bigger Picture
Most competitive intelligence tools are still fundamentally dashboards. They pull data together and leave a person to figure out what it means. RivalRadar takes a different approach. It treats monitoring, validating, remembering, and reporting as tasks AI agents can handle on their own, with people stepping in for the strategic decisions rather than the manual digging.
That’s the real shift here. Autonomous agents, browser automation, vector memory, and LLM reasoning aren’t just interesting pieces of technology on their own. Put together the way RivalRadar does, they turn scattered, easy-to-miss signals into something a product team or executive can actually use without spending hours pulling it together by hand.
Teams looking to build something similar face many of the same building blocks, whether they’re building a full competitive intelligence system or a narrower agentic workflow for a different problem. That usually means local or hosted LLMs, vector search, browser automation, and orchestration across several moving parts. Development partners with hands-on experience across that stack, such as IT Path Solutions, can help shorten the distance between a concept like this and a working system.
Frequently Asked Questions
Does RivalRadar still work for a competitor that doesn’t have an RSS feed or an active careers page?
Yes. Agents are configured individually per competitor, so if a company doesn’t publish an RSS feed or keep its job listings current, that specific agent can be turned off without affecting monitoring for that company’s other sources.
How does RivalRadar avoid mistaking an unrelated mention of a company name for a real signal?
Every post the Reddit and HackerNews agents collect goes through an AI relevance check before it’s stored. If a company name shows up by coincidence in a discussion that has nothing to do with that company, it typically gets filtered out rather than logged as a signal.
Why does RivalRadar run its own AI models instead of using a cloud AI provider?
Running models locally through Ollama cuts down on dependence on outside AI services, keeps costs more predictable, gives the team more control over how data gets handled, and avoids interruptions from third-party rate limits or outages.
Will RivalRadar flag every small change on a competitor’s website?
No. The WebWatcher Agent calculates how significant a change actually is before turning it into a signal, so something like a typo fix or a reworded button label gets ignored, while a new pricing tier or feature change gets flagged.
Can RivalRadar scale up to track a large number of competitors at once?
That’s how it’s designed. Because each agent runs independently, the workload can be spread across multiple workers or servers, and further scaling work, like distributed task queues and container-based orchestration, is already planned.
Where do the reports RivalRadar generates actually show up?
Reports appear on the dashboard, can be sent as formatted HTML emails, or posted straight to Slack. Each one carries a priority label, like CRITICAL or HIGH, so the most urgent findings don’t get lost among routine updates.