SaaS discovery is now AI-native.

Paste a URL. An agent crawls the product, screenshots it, and writes the listing in about thirty seconds, then re-reads it weekly.

Submit a URLBrowse the flow
01ABOVE λ

Ordinary liquid: launch-day noise, ranked by whoever showed up.

SCROLL TO COOL THE FIELD ↓
No popularity ranking

Discovery is measurement. Nothing is ordered by votes, clicks, or traffic, so there’s nothing to game.

Listings stay alive

Weekly re-crawls, free. Real change resurfaces the product.

Claims carry evidence

Every verdict links to the Atlas data behind it: verified, failed, or unverifiable.

THE FLOW

Today’s activity.

Real events in the order they happened: listed, claimed, updated, λ crossed.

superfluidity.ai
TODAY
3:32 AM
ScoutLayer materially updated
Find contacts behind any website
3:31 AM
KinoPipe materially updated
FFmpeg as a service, with typed tools for agents over MCP…
2:33 AM
TripMode materially updated
Controls Mac data usage on slow or expensive networks
2:32 AM
MindMac materially updated
Native macOS ChatGPT client with multi-provider AI support
2:32 AM
BeaverCheck materially updated
Free website audit tool with 100+ checks in one scan
2:31 AM
2:31 AM
Licentric materially updated
Software licensing API with SDKs for validation and…
1:42 AM
BetterCmdTab materially updated
A fast, native window switcher and app launcher for macOS
1:41 AM
Batteries materially updated
Track your Apple and Bluetooth devices' battery levels…
1:41 AM
AirParrot 3 materially updated
Mirror and stream your Mac or PC screen to Apple TV,…
1:40 AM
Termique materially updated
Cross-platform SSH client with synced hosts, keys, and AI
1:40 AM
Orvanta materially updated
Orchestration layer for AI agents that act on real systems
LOAD EARLIER EVENTS ↓
Products observed
418
Crawls / 24 hours
117
Material diffs today
—
Median ingest
50s
The thirty-second demo

You paste.
It reads.
You publish.

Atlas renders your site in a real browser, captures the pages that matter, extracts facts against a schema, and drafts the listing. You get an editable draft; nothing is auto-published.

01
You get an editable draft

Atlas writes it; nothing is published until you have read every field and said so.

02
Failure costs nothing

If the crawl fails, nothing is charged for it and the job says so itself.

03
You own the words

Every field is editable before publish, and after it.

JOB 8f21c4 · INGESTQUEUED
https://driftnote.app0.0s elapsed
  1. ●Queuedwaiting for a worker···
  2. ·Crawling—
  3. ·Reading—
  4. ·Writing—
  5. ·Ready for review—
DRAFT · EVERY FIELD EDITABLENOTHING IS PUBLIC UNTIL YOU PUBLISH
Driftnote
Notes that file themselves
Publish to the flowRewrite the description
What we are replacing

The story is what happens after launch.

The old ritual
  • A podium ranked by upvotes
  • Founder headshots and hype adjectives
  • “Please upvote” in your DMs at 6am
  • A listing that dies in 24 hours
  • Copy written by whoever wanted it most
Atlas
  • A flow of real events, in order
  • The product screenshot is the face
  • No votes to game
  • Listings re-crawl themselves, weekly, free
  • Claims measured, with evidence attached
AI-native by construction

Ask in your words and get products, with sources.

Retrieval runs over the catalog we measured ourselves — full text and vectors, fused. The answer can only cite what came back. When nothing fits, it says so.

ASK THE CATALOGGROUNDED IN MEASURED PRODUCTS

Three products in the catalog do this. Murmur is the closest match — it records a memo, transcribes it, and drafts the changelog entry for you to approve. Marginal and Sundial each cover part of it; nothing in the catalog does it for a self-hosted repo yet.

M
MurmurFREEMIUM
Voice memos in, changelogs out
changelogstranscription
Records a memo, transcribes it, drafts the entry for approval.
FRESH 2d
Mg
MarginalPAID
Release notes from commit history
release notesgit
Same output, but it reads git rather than your voice.
1 UNVERIFIABLE · FRESH 21d
S
SundialOPEN SOURCE
Cron jobs you can actually read
cronopen source
Transcribes audio notes, but writes tasks, not changelogs.
FRESH 9d
GROUNDED IN 3 OF 1,284 OBSERVED PRODUCTS
Grounded answers only

“Nothing in the catalog does this yet” is a valid answer. We never invent a product to fill the page.

Nothing to pay to read

Browsing, filters, comparisons and every product page are free and stay that way. Only the machine work is metered.

Readable by machines too

The same catalog is exposed over REST and MCP, so the assistant you already use can answer from it.

Readings

Every claim opens to
its evidence.

A reading is a measurement with evidence: Atlas checks uptime, re-reads pricing, tests features. Verdicts are verified, failed, or unverifiable, each with its data.

0
readings verified
30d
before re-verification
0
claims taken on trust
MURMUR · READINGSMEASURED 2d AGO
✓Uptime 99.97% over 30 daysVERIFIED
EVIDENCE
1,438 checks / 30d · 0 failures over 200ms
last checked 14:02 UTC · 118ms
EXPIRES IN 26 DAYS · RE-VERIFIED ON REFRESH
✓Free tier exists as stated
✓Demo runs end to end in 40s
?“Works with any git host”UNVERIFIABLE
the lambda point

The lambda point.

In physics, λ is where helium turns superfluid and flows without friction. Here it is the moment a product first sustains real outbound traffic. It is earned from actual visitors, and there is no way to buy it or ask your friends for it.

No product has crossed λ yet. When one does, this is where its real outbound clicks will appear.

Distribution

Readable by people and by agents.

The catalog is a first-class machine surface: a keyed REST API, and an MCP server so the assistant answering “what should I use for this?” reads ground truth.

REST · /api/v1READ ENDPOINTS WORK UNKEYED
$ curl www.superfluidity.ai/api/v1/search?q=changelog
{
"slug": "murmur",
"tagline": "Voice memos in, changelogs out",
"pricing_model": "freemium",
"readings": [{ "kind": "uptime", "verdict": "verified" }],
"freshness": 0.88, "last_crawled_at": "2d"
}
MCP · /mcpSTREAMABLE HTTP · READ-ONLY
search_productsnatural language + filters
get_productfacts, pricing, readings, evidence
get_alternativesnearest products by embedding
compareside by side, from measured facts

Every public page ships JSON-LD, and the catalog describes itself in llms.txt for crawlers that prefer prose.

Credits

You pay for machine work.
Browsing stays free.

100 free credits a month, and packs that never expire. Monthly credits spend first, holds release when a job fails, and every movement is on a ledger you can read. No countdown timers; expiry is stated plainly.

MONTHLY
100

Free, every month. They spend first and do not roll over.

PURCHASED
100 from $9

No expiry, ever. Spent only once the monthly ones are gone.

What things cost
ask the catalog1
refresh a listing10
regenerate copy10
deep verification pass15
ingest a URL → listing25
claim your productfree
weekly re-crawlsfree
browsing, filters, pagesfree

Deleting a listing is free and permanent: one confirmation, no retention offer.

Put your product where
the measuring happens.

About thirty seconds to a listing. Claiming and deleting are free.

Submit a URLBrowse the flow