Building a free, zero-effort system to automate knowledge retention
Or more simply, how to memorize (mostly) everything you read for <10m a day
Author’s note: I recommend reading the entire piece, but if you just want to implement this system, skip to Section 3. If you do not already use Obsidian or Anki, it will take you ~20 minutes to implement.
1. Motivation and goals of building this system
I am foremost a researcher and reader.
Since my childhood, I have always been an avid reader, spanning biography books (I most remember Einstein’s and Neil Armstrong’s biographies) to fiction (Percy Jackson!).
In middle school, I began neuroscience research. Despite my preconceived notions of working with cool biological tissues, it mostly consisted of, well, more reading. If you were granted autonomy to design your own experiments end-to-end (and did not become some glorified assembly-line worker for the lab), then 80% of your job as a researcher would be reading and understanding the literature instead of hands-on wet-lab work!
Fast forward to today (now a college dropout…), I’m now mostly reading blogs.
That means zero mainstream news! I don’t have any subscriptions to WSJ, NYT, or any other junk like that. Call me pretentious, but I prefer high-beta news (news with strong divergence from the consensus) — I find that niche domain-expert writers from research papers or Substacks provide not only asymmetric information early, but provide differentiated and detailed takes.
This is important to me.
(A list of my favorite newsletters can be found here and a collection of my favorite reads of all-time can be found here.)
Oh, but what if something really important happened and you were left in the dark?
My reasoning is that for anything really important I should know, someone will tell me. Secondly, this acts as a very good selection filter against temporal information, and therefore any mainstream news that slips past the filter is evergreen. Temporal information only matters for a moment in time; I’m mostly interested in information that remains important for a long(er)-time.
(Additionally, I know someone who doomscrolls X all day and sends interesting news all day as his full-time job.)
Surprisingly, he isn’t brain-rotted and actually writes. His Substack is linked below.
However, the problem now is that I’m reading 10+ articles a day, sometimes up to 30! I am now constrained by biology: without photographic memory, I am simply not able to retain 90% of the information I read. (Hmm… maybe we can engineer photographic memory in humans…)
Why, then, am I reading so much?
I read to learn. I read to improve my understanding of the world. But if I forget 90% of what I read, then I should just read 1-3 articles a day instead with a 100% retention rate!
That didn’t seem right to me — with the birth of the internet and almost all information known to humanity in my pocket — there must be a method to read more and retain more.
A tangent here, but it is of great sadness to me that despite the invention of computers and software and limitless knowledge, there has not been an explosion of intelligence and knowledge work.
Anyway, after several years of unserious, on-and-off experimentation, I have settled on a working AND free AND zero-effort system to automate knowledge retention at scale.
I promise you, if you adopt this system, you will no longer leave memory to chance and external factors. You’ll be engineering memory at will with no effort.
You make memory a choice.
2. What does it look like in practice?
(Bolded words are tools/software).
I read something online.
I highlight pieces of information that are personally important/relevant to me using Obsidian Web Clipper.
I import these notes into my Obsidian (a note-taking app). Flashcards are automatically generated using an AI model on the Cerebras API for free.
I review the generated flashcards in Obsidian and run a command.
I review those generated flashcards on Anki for the rest of my life.
Boom. That’s it.
It relies on just two principles:
AI for automating card generation
Anki for automating spaced repetition
Principles of spaced repetition
Spaced repetition is a tried-and-true psychological method to memorize anything efficiently. Instead of memorizing by cramming (which all of us did in college, typically the day before a midterm), we memorize over time. This is highly efficient for permanently storing knowledge, as graphed by the Ebbinghaus curve:
By spreading out reviews and increasing the duration between reviews, you’re able to flatten the forgetting curve (or the half-life) for some knowledge unit, effectively extending the memorization period of that specific card/fact/knowledge unit.
Luckily, with software like Anki, the interval spread and the optimal review duration are automatically handled for you, so all of this is abstracted away so you learn in bliss.
With this slow-but-steady approach, you can not only scale to memorize more bits of information, but retain them for longer as well. This is the opposite of cramming: you can only remember a capped amount for a temporary period.
It is interesting to me that there seems to be no limit to the amount of information you can retain. I have been able to memorize the capital and flag of every country in the world, and am on my way to memorizing the 1st, 2nd, and 3rd largest cities of each country. I have also memorized every single element on the periodic table. And I will probably continue to learn more things as I age.
But don’t take it from me alone!
Many Jeopardy winners have used spaced repetition, for example. Anki is also popular among pre-med students when preparing for the MCAT.
So yes, this thing works, and it works really well.
ROI of spaced repetition
Okay, maybe I lied a little: reviewing Anki cards every day does require a bit of effort!
So, is it worth it?
My answer is an unequivocal yes, on the assumption that you are reading and highlighting information that is relevant and useful to you. In fact, I can estimate the ROI for doing this!
Let’s assume I generate, on average, ~10 flashcards (cards) per article. Reading 10 articles a day means 100 new cards a day. On average, it takes me 12 seconds to review a card, with an accuracy of 80%. Based on my 5-year experience with Anki and talking with other users, this is quite normal.

This means that learning all 100 cards will only take 20 minutes a day. This seems very reasonable!
In this quite famous blog, the author calculates that for an average card, you only need ~4-7 minutes of total review time over your life to remember a card.1 This corroborates well with my own data.
But let’s take a worst-case scenario: it takes you 10 minutes of total review time to remember a card for life! That means if memorizing a fact seems worth 10 minutes of your time over your entire life, just Ankify it.
This is really really important — if you believe that a fact will be worth 10 minutes of your time throughout your life, you can make memorizing it for the rest of your life a choice.
I have 60 years left to live (I am assuming I die at 80). That’s ~22,000 days I have left in this world. If you amortize the time cost over that period, that’s just 1.6s/day/card.2
Counterintuitively, that’s tiny!
Zooming out to my entire lifetime, the daily burden is almost nothing. So even under the worst-case assumption, the lifetime price of remembering everything I read boils down to the equivalent of a few minutes a day.
This is why I think of Anki not as “extra work” but as an infinitely compounding force multiplier on the time I already spend reading.
TLDR: the ROI is high.
Spaced retention + AI automation
However, some people still push back against Anki. The three common claims are that:
Making Anki cards wastes learning time, since it takes a disproportionate amount of time to make one than it takes to review one.
It is also hard to make good and effective cards, so much so that there are many separate guides on writing effective cards.
You shouldn’t memorize things you haven’t learned or understood yet.
These were the same problems I suffered from in the prehistoric pre-GPT era, but that era has long passed. AI can now solve each of these problems, so there is effectively no reason why you shouldn’t be using this system in 2025.
The system I detail below will:
Automatically generate Anki cards based on a template provided in the prompt.
Make good cards using context aggregated from the sources I provided above.
Help you understand information you don’t know very quickly.
3. A guide on replicating this yourself
Unfortunately, this only works using Obsidian as your primary note-taker instead of Notion or Logseq or Roam Research (looking at you, Kyle Harrison — it’s never too late to join the light side).
Fortunately, basically everyone I know who uses Obsidian thinks it’s much better than Notion or Logseq or Roam Research (except for Kyle Harrison…)
Still, even if you don’t plan on completely switching to Obsidian, it only takes ~20 minutes to set up this entire system, so you can still follow along!
Obsidian
First, install Obsidian. Then:
Open the settings page.
Open the “Community Plugins” page.
Turn off restricted mode and browse for this plugin: Export to Anki
Go to the settings of the plugin and click on Note Type Table. For the basic note type, add this in its custom regexp: ^(.*[^\n:]{1});{2}([^\n:]{1}.*)
In the same plugin settings page, set the Deck to any name you want (I have mine as “Obsidian”).
Copy the rest of my settings.
Note that if you already use Obsidian, fill in the Scan Directory setting, otherwise an entire vault scan will cause Obsidian to crash.
Obsidian Web Clipper
Next, install the Obsidian Web Clipper. Then:
Open the settings page of the Chrome extension.
Click Highlighter on the left and copy these settings:
Click Interpreter on the left. Enable it and enable automatic run.
Under providers, add a new provider and copy these settings. You can copy the Base URL here: https://api.cerebras.ai/v1/chat/completions
Note that you can use any provider you want, such as OpenRouter or Groq. However, Cerebras is the only provider I know that provides free API usage.
For the API key, make a free Cerebras Cloud account here (this is a referral code, you and I will both get +200K tokens of usage per day), click API keys on the left, and generate a new API key. Copy and paste that in the API key field.
Under models, add a new model and copy these settings. You can copy the Model ID here: qwen-3-235b-a22b-instruct-2507
Note that you can use any model provided by Cerebras. I find the Qwen-3 Instruct model to be quite good at following instructions with a large context length (64,000 tokens) while remaining sufficiently fast.
Make a new template on the left.
Name it whatever you want.
You typically want to set the behavior as: Create new note.
You can customize your note name, but I just use this (I recommend keeping the safe_name operator): {{title|safe_name}}
Add the folder path where you want the note to be saved. If you made a new Obsidian vault for this purpose, just leave it blank for now.
For the note context, copy and paste this:
{{content}}
# Anki
{{"Read the interpreter context. Finally, make effective Anki cards based on the interpreter rules and the content you read above."}}
For the interpreter context, copy and paste this:
You will create effective Anki cards for learning based on the following rules to help me recall facts from articles I read.
When making Anki cards, follow these principles:
1. Single idea per question and answer.
2. Each card should ask a specific question (no fill-in-the-blanks or cloze cards).
3. The question should be fully self-contained.
4. Make cards as concise as possible. Complete sentences not necessary. THIS IS SUPER IMPORTANT!!!
5. No bolding or italics formatting in markdown.
Note context:
Title: {{title}}
Highlighted facts: {{content}}
For each highlighted fact:
- If a highlight is too long or contains multiple ideas, break it up into multiple atomic Anki cards, but don't make redundant cards.
- Output each card in the format: “QUESTION?;;ANSWER.”
- Separate each card (“QUESTION?;;ANSWER.”) with two empty lines.
Anki
Finally, install Anki. Then:
At the bottom, create a new deck called Obsidian (or whatever you want, it has to match what you typed for the Obsidian plugin earlier).
Click the cog wheel next to that new deck to edit the deck settings.
Under Daily Limits:
Set new cards/day to 9999
Set maximum reviews/day to 9999
Under New Cards:
Set learning steps as 10m 1h
Set insertion order to Sequential (oldest cards first)
Copy these settings:
Enable FSRS and set the desired retention to 80% or 85%. Do not set it above 90% because it exponentially increases your review time.
Save the settings!
Open the Add-ons page, click Get Add-ons, and input: 2055492159
Click on AnkiConnect and click Config on the bottom right. Paste this in:
{
"apiKey": null,
"apiLogPath": null,
"ignoreOriginList": [],
"webBindAddress": "127.0.0.1",
"webBindPort": 8765,
"webCorsOrigin": "http://localhost",
"webCorsOriginList": [
"http://localhost",
"app://obsidian.md"
]
}
And… you’re done!
Walkthrough
Open something you want to read.
Open the Obsidian Web Clipper in the top-right (I use a keybind) and click the highlighter button.
Highlight away!
Clip highlights and click Add to Obsidian.
Ensure the Anki cards are in the right format (question first, followed by ?;; then answer). Review and adjust the generated cards if needed.
Open the Command Palette on Obsidian (CMD+P), type Scan Vault, and Enter. (Anki needs to be open.) When it’s finished, you should see these HTML ID comments underneath each card.
Review the cards and profit.
Do this for each article and review the cards every day. It’s okay to be backlogged on New cards, you should always get Learn and Due cards to 0 though.
PROFIT.
4. FAQs
Ask me in the comments or in DMs and I will answer them here!
How do I delete cards?
Type DELETE between each card and HTML comment and run the Scan Vault command again. The comments should disappear, indicating a successful delete. Do not only delete cards in the Anki interface, as the next time you run the command in Obsidian, it will regenerate those deleted cards.
Useful tip: In the Export to Anki setting, I changed “Delete Note Line” to “DD” for ease of use.
To delete cards in bulk, I use VSCode to regex match <!-- and insert DD above it.
This calculation was also inspired by Gwern’s analysis of this post.
This is simplified — Anki cards are often front-loaded, and I don’t take into account the differences in learning and reviewing schedules. The daily time/card also eventually decreases as the duration times stretch to decades.





















Great write-up. I’m tackling this problem too with Prepma: https://prepma.com
. It focuses on flashcards + spaced repetition + quizzes. How would you compare that approach to yours?
I was JUST thinking about this the other day, massive notion doc isn't cutting it...