Nagare,feature by feature.

Add a competitor, watch the first reading

Type their site. Nagare maps it first (robots.txt, the sitemap, the pages that matter) and asks which to follow. Then it reads it, and each step shows as it lands: their site, their GitHub, their API documents, their past year from Common Crawl, and last the year before as a report.

DevCycle (devcycle.com) was added during the recording: 19 pages mapped and read by the HTTP crawler, its GitHub organisation found with 29 public repositories.

A battlecard for each competitor

Written after each crawl: who they are in one line, quotes in their own words, where you stand from the feature matrix, their pricing next to yours. Every claim links to the page it came from.

LaunchDarkly, 284 pages read.

Every change, with its diff

Each crawl is compared with the last one. A change keeps its page, its diff, its category and its severity; noise such as dates and counters is set aside.

The year before you started

The timeline draws each key page as a thread. Before tracking began, the hatched zone holds what Common Crawl captured, always marked archived and less reliable. Messaging lists each headline period, the year in one look, and the archive's coverage month by month.

The year before, month by month

Once Nagare has asked Common Crawl about a competitor, it writes the year before as a report: for each month, what changed on each competitor’s home and pricing pages, each change cited with its capture date. History, never news, and never in a brief.

Orbital’s: October 2025 to September 2026, twelve months.

Published, not promoted

Pages their site publishes that no other page links to: docs ahead of the marketing, a plan not listed yet. Grouped by what is worth a look first.

Your catalog against theirs

Each competitor's cell is proved by a quote from their site, and "not found" is never "no". A cell opens its history: when it was first read, what changed, the page that says it.

Orbital’s catalog against the five; the cell opened here is GrowthBook’s.

Pricing, side by side, and over time

Every plan aligned by tier, each price proved by its page. Changes are dated, a removed plan stays visible with the day it went, and Common Crawl's captures show their prices from before you started.

The words they lean on

Follow a theme and its words, and see for each product the share of pages that carry them, month by month, and since when. Before tracking, Common Crawl's captures fill the months, hatched.

These four themes were added for the tour.

Built, not pushed

Where each feature of your catalog appears on their site: built in docs, API or changelog, announced on the blog, packaged on pricing, pushed on the home page. Built but not pushed is the signal.

Is it really a competitor?

The customers they show, from their stories and the logos on their pages, with the day each appeared or disappeared, the ones you both show, and their audience next to yours.

What moves before the marketing

Off their site: funding, open roles by team, ads and links. Their public GitHub, read through GitHub's own API. The OpenAPI documents they publish, operation by operation, each change dated.

What their customers say

Their PeerSpot reviews, read through Firecrawl’s Alexandria for the product their site links to, or the one you set: the rating and its history rebuilt from each review’s date, what reviewers value and ask for in their own words, and what they say of the price.

LaunchDarkly: 4.1 / 5 on 19 reviews.

Where they may be heading

A trajectory reads six months of dated signals and writes hypotheses. Each one says how many signals it rests on and of how many kinds, places them on a time axis and lists them; every source is grouped by kind, the archive apart. Hypotheses, never predictions.

LaunchDarkly: 6 hypotheses, 30 sources.

The brief

At the cadence you choose, from daily to monthly: what moved across your competitors, what it may mean and what to watch, each sentence cited. It ends with what changed on your own site.

Your product first

What you do, who you sell to and your plans, read from your knowledge file and confirmed by you; then your feature catalog, plan by plan. Every competitor is read against it, and what only they offer comes back as a suggestion.

In your own AI tools

A read-only MCP server lets Claude Code or your agents ask Nagare directly, with 20 tools on a workspace API key, each answer with its evidence. Every site also exports as one clean llms-full.txt.