Turn any page into a feed that shows only what fits you
Four taps from a listing site, a shop, a job board or a vendor advisory page to a feed in your reader. Your own criteria decide what gets through. No selectors, no code, your own AI key.
For people who check the same marketplace, job board, release list or advisory page every day and are tired of reading past the ninety percent that was never for them.
You open the same five tabs every morning. Most of what is new is not for you, and the one that was is already gone.
Listing sites have no feed. Shops announce new arrivals nowhere. Vendor advisory pages bury the one patch that matters under fifty that do not. The feeds that exist send everything, so you skim two hundred titles to find three. Site filters only know keywords; they cannot tell "two rooms with a balcony, no agency" from "two rooms, agency fee, ground floor".
One feed, in your reader, with only the entries that match what you wrote down as "good".
You describe a good entry once, in plain words. Every new entry from every page you follow is read against that description and scored. You subscribe to "60 or better" and forget the tabs. Change your mind, and the next entries are judged by the new rules; the old verdicts stay on record.
Four taps from an address to a feed you want to read
-
1 · Address
Paste a link
The page is fetched, in a real browser when it has to be, and you see what was found before you commit to anything.
-
2 · Entries
Confirm the list
The repeating blocks on the page are proposed as entries, with title, link, text and image already guessed. Tap to correct one. You never type a selector.
-
3 · Storage
Keep what it collects
Store entries in a collection with the shape you want: a job offer with a salary, a product with a price, an article with an author. Or skip it and just take the feed.
-
4 · Scoring
Say what good looks like
Write your criteria, set a threshold, paste your OpenAI-compatible key. Every new item is scored and the winners go to your reader or to Telegram.
Not a scraper. A reader that already did the reading.
A feed address for any page
Paste it into FreshRSS, Feedly, NetNewsWire, Reeder, Inoreader. Refreshed on your schedule, served from cache when the site is slow or down.
Scores you can argue with
Every verdict carries the reasoning and the version of your criteria that produced it. Edit the criteria and compare old and new scores side by side before you trust them.
Delivery where you already are
A threshold feed for your reader, a Telegram digest with the top entries of the day, or a Slack channel for a team, so a good one never waits for you to open the app.
One item, many boards
The same posting on three sites is one entry, scored once, listing every place it appeared.
Pages that fight back
Rendered in a real browser with a persistent profile, and a second solver steps in for challenge pages. When a site still refuses, you are told, not shown an empty feed.
Your key, your bill, your data
Scoring runs on your OpenAI-compatible key. You see exactly what each verdict cost, and nothing you follow is used for anything but your feeds.
The pages people stop refreshing
Finding a job
Every board in one feed. Your criteria say remote within the EU, Rails in the stack, a real salary; the model reads each posting and only the fits get through.
Finding a flat
Listing sites scored by what you wrote: two rooms, a balcony, no agency fee, under your budget. A good one reaches your phone before the viewing slots are gone.
Finding a plot of land
Parcels are described in ten ways on ten sites. Say the area, the access road, the zoning you need, and read one feed of parcels worth a call.
Security patches that matter
Vendor advisory pages for your stack, read against what you actually run. The one patch worth a rollout tonight reaches Slack; the fifty that do not stay out of it.
The scored results, wherever you already look
In your RSS reader
A feed address per collection, with a score threshold if you want one. FreshRSS, Feedly, NetNewsWire, Reeder, Inoreader.
On the Sift site
Every item with its score and the reasoning, searchable and filterable, on a phone or a desktop.
On Telegram
A digest of the day's top entries sent to your chat on the schedule you pick.
Into Slack by webhook
A channel for the team: only the entries above the line, posted as they arrive, with a link back to the item.
Built by someone who needed it every day
Sift started as the author's own way to follow [NUMBER] boards, from job listings to flats, without opening them, and has run every day since [YEAR]. Every piece of it exists because a real page broke, a real feed got noisy, or a real verdict was wrong and had to be explained. It is not a demo of an AI; it is a tool with an AI in one well-defined place.
Your feeds are plain RSS you can read in any reader. Deleting your account removes what it holds.
Pay for the plumbing. The AI runs on your own key.
Every paid plan is bring your own: your OpenAI-compatible key and, when a site needs one, your own proxy. The model bill goes to your provider at their price, typically [TYPICAL MONTHLY MODEL COST] a month for a few hundred items. Sift charges a flat fee for the fetching, the storage and the feeds.
Free
0Up to 5 RSS feeds to follow, one collection to store and score them in, refreshed hourly. Enough to stop refreshing the pages you check most.
Start freeReader
Most peopleUp to 100 boards to follow and store as items, as many collections as you like, scoring with your OpenAI-compatible key, threshold feeds, Telegram digests, criteria history and comparison, refresh every 15 minutes. Bring your own proxy for the sites that need one.
Start free, upgrade when it pays offIf in the first two weeks of Reader the feeds did not save you the tabs, write one line and the month is refunded. Cancel any time; your feeds keep working on the free plan.
The three doubts we hear most
- I could do this with a scraper and a script.
- You could, and for one page that is a fine evening. The part that costs months is the rest: pages that change layout, sites that block a headless browser, keeping a feed alive at 4 am, and explaining a score you gave three weeks ago. Sift is that rest, done once for everybody.
- I do not trust an AI to pick for me.
- Neither do we, which is why it never picks. It scores, with its reasoning attached, against words you wrote, and you set the line. Everything under the line is still in the collection, one filter away.
- My key, my bill: will this run up costs?
- Each item is scored once per version of your criteria, never on every refresh. You see the count and the cost per month in the app, and you can cap it. A few hundred items a month is typically [TYPICAL MONTHLY MODEL COST].
Yes if, no if
This is for you if
you follow a few pages closely and know in words what a good entry looks like: a flat worth a viewing, a job that fits, a patch that really needs a rollout, a paper worth an evening. You already use a reader, or you want one place instead of five tabs.
This is not for you if
you need to crawl thousands of pages, want data exports for a business, or want the AI to act on your behalf. Sift reads and scores; it does not apply, buy or send.
The practical ones
- Which readers work?
- Any reader that takes an RSS address: FreshRSS, Feedly, NetNewsWire, Reeder, Inoreader, Miniflux. The address carries your private token.
- Which AI providers can I use?
- Any OpenAI-compatible endpoint, with your own key pasted once in Settings. The key is stored encrypted and used only for your scoring.
- What if a site blocks it?
- Most pages render in a real browser with a persistent profile; challenge pages go to a second solver. A page that still refuses is reported as blocked in the app instead of pretending to be empty.
- Can I add a source from my phone?
- Yes. Adding a source is four screens designed for a thumb, and correcting a pick is a tap on the page, not a selector.
- Can I leave with my data?
- Your feeds are plain RSS, and deleting your account removes what it holds the same day you ask.