Built, not pushed
Where each feature of your catalog appears on their site: docs and changelog, blog, pricing, home page. A feature built but not pushed is the signal.
Competitive intelligence for product teams · early access
The web forgets. Nagare doesn’t.
See where your competitors are heading.
Every version of their site, the year before you started, and the signals that move before the marketing.
Nagare reads your competitors’ public websites, their API documents and their GitHub, keeps every version and every deletion, and tells you what they may be preparing. Every sentence links to its proof.
For product managers and founders who own competitive strategy.
EXAMPLE · A NEW COMPETITOR’S FIRST READING
Reading flagstack.io: 312 pages so far…Looking for their GitHub organisation…Looking for their OpenAPI document, on their site then in their SDKs…Asking Common Crawl how their home page read last year…First reading done.Nagare read flagstack.io for the first time.
Reading flagstack.io: 312 pages so far…Looking for their GitHub organisation…Looking for their OpenAPI document, on their site then in their SDKs…Asking Common Crawl how their home page read last year…First reading done.Nagare read flagstack.io for the first time.
- ✓Their site· 312 pages read
- ✓Their GitHub account· flagstack, 4 repositories
- ✓Their API documents· 64 operations, from their SDK
- ✓Their past year· 9 captures, archived
Every version, and the year beforeAny page between any two dates, and a year of their home and pricing pages from a public web archive, marked as less reliable.
Signals before the marketingDocs, API documents, GitHub releases, pages nobody links to, open roles: where a launch shows first.
Every claim, citedNo “AI may be wrong” disclaimers. Each sentence links to the page, the change or the signal behind it.
Nagare in 77 seconds
Your own product and its feature catalog first, then your competitors, from a first reading to the brief: thirteen steps recorded in the console on a demo workspace, a feature-flag product against five real competitors, read from their public sites, their GitHub, their API documents and Common Crawl. No sound.
Hypotheses, never predictions
A trajectory report says what a competitor may be preparing. Each hypothesis rests on dated signals you can open: a feature in their docs that their home page does not mention yet, new operations in their API, a release of their SDK, a page published but linked from nowhere.
It says “may” and shows its evidence. You read the signals and decide.
WHAT A SENTENCE CAN CITE
EA change on their site, with its diff
PA page as Nagare read it, on a given day
SOperations added to their published API
GA release or a repository on their public GitHub
JA role they opened, and its team
CA page as Common Crawl archived it, marked less reliable
Day one already has a past
Add a competitor and Nagare asks Common Crawl, a non-profit archive of the public web, how their home and pricing pages read over the past twelve months.
Archived periods are always marked: fewer captures, coarser dates, less reliable than what Nagare reads itself. Nagare tells that year in one report, month by month. From the day you add them, it keeps every version.
12 monthsone capture a month at mostarchived, less reliable
What moves before the marketing
A feature often shows up in their docs, their API and their SDKs weeks before their home page. Nagare reads those places, dates what it finds, and says when it is only a hint.
Their API, operation by operation
The OpenAPI document they publish, or the one their public SDKs carry on GitHub. New, removed and deprecated operations, each dated, and cited in your brief.
Their public GitHub
Repositories and releases, read through GitHub’s own API. What existed when Nagare started is history; only what appears afterwards is news.
Published, not promoted
Pages their site publishes that no other page of theirs links to: a launch page waiting, a plan not listed yet. Nagare reads what the site publishes and never guesses addresses.
Is it really a competitor?
The customers they show, from their customer 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, in their own words.
Names only: Nagare never copies their logos.
Your week,
in two minutes.
A brief every day, week, two weeks or month, whatever you switch on. What moved across your competitors, on their sites, in their API and in their hiring, what it likely means, and what to watch next. It ends with what changed on your own site.
DDailyWWeekly2WEvery 2 weeksMMonthly
In the app, by e-mail and in Slack.
EXAMPLE · WEEKLY BRIEF · 17 OCT — 23 OCT 2026
Two competitors moved on pricing. Entitly removed its free plan and raised the entry tier to $49/moE1. Flagstack’s API gained approvals: three new operationsS1, a day after its Node SDK shipped themG1. Meterbase is going upmarket: SOC 2 Type IIE2 and two Enterprise AE rolesJ1.
● MAJOR · PRICINGFree plan removed from Entitly’s pricing page
● NOTABLE · PRODUCTFlagstack’s API gained approvals
● NOTABLE · HIRINGMeterbase hires 2 Enterprise AEs
Next to your product
Battlecards, pricing and the feature matrix put each competitor next to your product, each claim proved by their own page.
Live data: what Nagare read between 20 and 23 Oct 2026 for Orbital, a demo product, against five feature-flag tools.
WHERE YOU STAND · GROWTHBOOK
● They’re ahead · 2
A/B testsOn-premise deployment
● Even · 3Percentage rolloutsUser segmentsSSO (SAML)
On-premise deployment for regulated industries — a talking point, with its evidence.A battlecard for each competitor
Where they’re ahead, where you are, and three talking points with their evidence. Written after each crawl; yours to edit.
FIRST PAID PLAN · AS OF 23 OCT
You$39/moflat
GrowthBook$40/moper seat
Unleash$75/moper seat
Statsig$150/mo5M events incl.
Pricing, side by side
Every plan aligned by tier, entry plans compared, each price proved by its page, and every move dated.
OFF THEIR SITE · LATEST CHECKS, 20–23 OCT
Unleash raised a $35M Series B (Mar 2026)F1
LaunchDarkly lists 53 open roles on its careers pageJ1
Statsig is running 12 Google adsA1
Statsig links to GrowthBook from a comparison pageL1
Signals off their site
Open roles, funding rounds, Google ads, traffic, the links they publish and their PeerSpot reviews. Each one names its source.
FEATURE MATRIX · ORBITAL VS 5
YOULDSGGBFSUN
A/B tests◐✓✓✓✓✓
On-premise deployment✕··✓✓✓
Session replay✕✓✓···
“Self host on-premises or in your own cloud.” — flagsmith.com, 20 Oct✓ yes · ◐ partial · ✕ no · · not foundWhere you’re behind, in their own words
Your catalog from your own knowledge.md; each competitor’s cell proved by a quote from their site. A model never marks a feature absent.
In your own AI
Claude Code or your own agents ask Nagare directly, through its MCP server: competitors, changes, page history, briefs, battlecards, trajectories, the matrix and pricing. Read-only, on a workspace API key, every answer with its evidence and a link back.
Each site also exports as one clean llms-full.txt.
EXAMPLE · CLAUDE CODE, WITH NAGARE’S MCP SERVER
> What is Flagstack preparing?
Flagstack may be preparing an approvals workflow. Their docs have described approvals since 2 Oct, their API gained three approval operations on 21 Oct, and their Node SDK shipped them the day before.
P1 · /docs/approvalsS1 · Flagstack APIG1 · sdk-node v2.1.0
get_report · get_api_surface · get_git · get_lifecycle · get_customers · list_changes
How Nagare works
Four steps, and a thread that keeps every page’s history.
01
Declare
Paste a competitor’s site. Nagare maps its pages and shows its first reading as it happens: the site, their GitHub, their API documents, their past year.
02
Read politely
An identified crawler on your schedule, robots.txt honored, same site only. GitHub and Common Crawl through their own public APIs.
03
Keep every version
A new version only when content changed. Deletions recorded with the day they happened. The archive’s year kept apart and marked.
04
Tell you where they’re heading
Briefs, battlecards and trajectories: what moved, what it may mean, each sentence linked to its proof.
Nothing is reprocessed when nothing changed: the history stays honest, and your AI budget stays small.
Questions, answered
Is a trajectory a prediction?
No. It is a hypothesis with the dated signals behind it: a feature in their docs but not on their home page, new API operations, a release, a page published but not linked. You see each signal and decide what it means.
Where does the past come from?
From Common Crawl, a non-profit that archives a large part of the public web about once a month. Nagare asks it how a competitor’s home and pricing pages read in the twelve months before you added them, keeps at most one capture a month, and always marks it as archived and less reliable. It feeds comparisons and one report, the year before, month by month; never an alert or a brief.
Which pages does it read?
The public pages of the sites you declare: pricing, product, docs, changelog, careers, blog. A site that asks robots not to read a page is left alone. Beyond their site, Nagare reads the API documents they publish and their public GitHub, through GitHub’s own API. Never LinkedIn, never behind a login.
Is it real-time?
No, and on purpose. Nagare reads on a schedule you choose: every day, every week, or by hand with Crawl now. Major changes still send an alert as soon as they are read.
Where does AI come in?
Triage, the brief, the battlecard and the trajectory are written by a model, and every sentence carries the change, page or signal behind it. Reading sites, the archive, API documents, GitHub and customers involves no AI. There is no chat: your own assistant reads Nagare instead.
Can my own AI use it?
Yes. A read-only MCP server lets Claude Code or your agents query Nagare on a workspace API key, each answer with its evidence. Each site also exports as one clean llms-full.txt, regenerated only when its content changed.
What does it cost?
Nagare is in early access and its pricing is not decided yet. Join the waitlist and we will tell you first.
Does it see what they deleted?
Yes. When a page disappears, Nagare records the day it went and keeps its last version. A retired plan, a quietly removed customer logo: removals are often the loudest move a competitor makes.
Their last year is already there.
Add a competitor and Nagare reads its past year from the archive, marked as less reliable. From that day on, Nagare keeps every version itself: the sooner you start, the more of their history is yours.
EXAMPLE · FLAGSTACK.IO · SINCE 4 AUG
A year from the archive, then every week since, kept.
312pages kept
9archived captures
47changes
11weekly briefs