Keep a read it later list with OpenClaw

Most people who use a read it later app have a few hundred saved articles they never got back to, sitting in a queue they open less and less, and that queue recently lost its most familiar name. Mozilla shut Pocket down on July 8, 2025, disabled exports on November 12, and deleted the remaining user data after that. Omnivore, the open source favorite, had already gone offline in November 2024 after ElevenLabs hired the team and wound down the hosted service. A lot of people are now looking for a new home for the same pile.

Most of the apps that remain want a subscription. Readwise Reader is the choice for heavy highlighters who want one inbox for articles, PDFs, and newsletters, and it runs about $10 a month on the annual plan with no permanent free tier. Instapaper kept the simple version of the idea alive and still has a free tier, with Premium at $5.99 a month or $59.99 a year. Matter offers a generous free library and charges $8 a month or $60 a year for Premium features like HD text to speech and newsletter sync.

The polished readers excel at typography and highlights; the weak spot for many users is whether anything ever pushes the oldest saves back to the surface. Your Operator.io agent can keep the list in a file you own and, more to the point, push the backlog at you until it gets read.

How it works

You send the reading list prompt, and the agent stands up the queue and the daily nudge together. From then on, every link you hand it runs through the same short loop:

• It opens the page, writes a one line summary itself, tags the topic, and adds a row to a CSV in its workspace with the title, URL, source, and the date you saved it.
• It checks that file before saving, so the same article never lands in the list twice.
• Once a day it messages you a few unread pieces worth getting to, favoring the ones you saved a while ago so older saves resurface instead of sinking under everything newer.
• When you tell it you have finished one, it marks the row read and stamps the date, which keeps the daily picks fresh instead of looping you through the same five links.

The one line summary is what makes the list usable. A column of bare URLs tells you nothing six weeks later, while a row that says in a sentence what each piece is lets you pick tonight's read at a glance.

The prompt

This is the instruction the agent acts on:

Be my reading list. Create a spreadsheet in your workspace, a CSV with columns
for the title, the URL, the source, the date I saved it, a one line summary, a
topic tag, whether I have read it, and the date I read it. When I send you a
link, check the file first so you never save the same article twice, then open
it, write the one line summary and a tag yourself, and add it as unread. Set up
an automation that messages me here once a day with a few unread pieces worth
getting to, favoring the ones I saved a while ago so nothing rots at the bottom
of the list, and when I tell you I have read something, mark it read and stamp
the date. Before you start, ask me which topics I save most, how many items to
resurface a day, and what time to send them.

You can send this exact prompt to your agent straight from the prompts library instead of retyping it.

Using it day to day

The whole thing runs off links you were going to send anyway. You read half of something on your phone, paste the URL to OpenClaw, and it files the rest with a summary so future you knows what it was. The daily message arrives in your channel at the time you picked, with a few things to read and a sentence on each, and you answer it the way you would answer a friend who asked what you have been reading.

The agent opens each link from its own host rather than from your browser, so it is not carrying your logged in sessions. A public article summarizes cleanly. A piece behind a paywall or a sign in often comes back thin, written from whatever the page shows a logged out visitor, and an internal page that only loads inside your company network will not open for it at all. Save those to read yourself rather than handing them over, and save links you trust, since the agent will fetch whatever URL you give it.

Because the history is a file the agent controls, you can also steer it in plain language. Ask it for only the short pieces on a night when you have ten minutes, pull everything you ever saved about a topic before you write about it yourself, or tell it to clear out anything older than six months that you are clearly never going to read. It works from the CSV rather than guessing, so "what have I not gotten to from last month" is a real question with a real answer.

Tagging pays off once the list passes a few dozen rows. A tag like "security" or "product design" lets the daily nudge pull from one lane when you are in the mood for depth on that subject. You can also ask the agent to stop saving certain domains if they always turn out to be clickbait, which keeps the CSV from becoming a graveyard of listicles.

What happened to the read later category

Pocket's shutdown was unusual only in the scale of its user base. Mozilla's support article documents the timeline: new signups stopped in May 2025, the service went dark in July, exports ended in November, and the API was disabled the same day. Anyone who missed the export window lost saves that had lived in Pocket for a decade.

Omnivore followed a different path. The open source repo still exists and can be self hosted, but the cloud instance millions relied on is gone. Users had until mid November 2024 to export before deletion. For people who chose Omnivore specifically because it was open source, the lesson was harsh: hosted "open source" still depends on someone's servers staying online.

That leaves a split market, and the live options sort out like this:

Raindrop.io sits alongside these, treating bookmarks as collections with a reader mode.

Why a file beats another read later app

The dedicated apps are a pleasure to read in, and if you live for highlights and a polished reader, one of them is probably worth the money. Readwise Reader bundles highlight sync to Notion and Obsidian, Ghostreader summaries, and PDF handling for about $10 a month billed annually. Instapaper Premium adds full text search, permanent archive, and text to speech for $5.99 a month. Those products earn their price when reading and annotating is the core of your job.

The failure mode they share is backlog gravity. Saving is frictionless; reading is not. Without a deliberate resurfacing strategy, the queue grows monotonically until you guilt archive three hundred items or export and start over. Pocket users felt that when the export deadline forced the issue. The agent prompt builds resurfacing into the daily automation: favor older unread rows, cap how many it sends, mark read with a date so the picker has fresh candidates tomorrow.

The CSV trades the polished reader for ownership and a nudge loop. Your list is a file in the agent workspace, holding the exact articles you saved, each summarized and tagged. There is no separate subscription beyond the OpenClaw host you already run, and nothing to export later because it was a file the whole time. If you moved off Pocket and have not settled anywhere, that combination, portable data plus a daily ping aimed at the bottom of the queue, is the practical replacement.

Wallabag remains the self hosted alternative if you want a full reading app under your control. It parses articles, stores them offline, and syncs to mobile clients, but you operate the server and backups yourself. The agent approach skips the server entirely: the "reader" is whatever browser you already use when you tap the link in the daily message, and the agent's job is cataloging, summarizing, and resurfacing.

To start, open the prompts library and send the reading list to your agent. It asks which topics you save most and when you want the daily nudge, and from there it keeps the queue in whatever channel you already use.