Anki FSRS explained: what it is, how to enable it, what to set

FSRS (Free Spaced Repetition Scheduler) is an alternative to Anki's legacy scheduler, which is based on SM-2. It uses your review history to estimate when you are likely to forget each card and schedules the next review so you recall it with the probability you choose, called desired retention.

In practice, you stop hand-tuning multipliers and intervals and mainly decide on one number: what share of cards you want to remember when they come up. FSRS works out the intervals from there. It is still spaced repetition; what changes is how the spacing is calculated.

According to the Anki manual, FSRS works in Anki 23.10 or later, AnkiMobile 23.10 or later, AnkiWeb, and AnkiDroid 2.17 or later. The manual warns that if you study in a client that does not support it, things will not work correctly. Before you enable it, update every device you review on.

This check prevents the most frustrating problem: reviewing on your phone with an old version and finding inconsistent intervals after syncing. If one of your devices cannot be updated, sort that out before switching schedulers.

You enable FSRS in the FSRS section at the bottom of the deck options, and the manual notes that the setting is global: you cannot use it in some decks and not others. Before turning it on, the manual recommends removing any custom scheduling code you may have added.

It also asks you to keep learning and relearning steps under one day, for example 10m or 30m. The reason is that FSRS handles day-based intervals; if your steps already reach a day or more, the two systems overlap. Check those fields in every preset you use.

Desired retention is the probability of remembering a card when it comes up for review. The default is 90%. Raising it shortens intervals and gives you more reviews; lowering it means fewer reviews but more cards forgotten before you see them. The manual suggests keeping it below 97%, because the workload climbs sharply as you approach 100%.

One way to picture it, as our own example: at 90%, out of 200 due reviews you would expect to forget about 20. With higher retention that number falls, but the total number of reviews rises. Retention is set per preset, so you can use a higher value for a deck tied to a near exam and a lower one for a maintenance deck.

FSRS has parameters that adapt to how you remember, fitted by an optimizer that analyzes your review history. The manual says re-optimizing about once a month is enough and mentions an optional health check.

With little data the fit is less reliable: the manual lists having few reviews, fewer than a few hundred, as a common reason FSRS performs poorly. If you are just starting out, you can keep the default parameters and optimize later, once your history has grown.

Switching schedulers in the middle of exam season adds uncertainty exactly when you need it least. If you can, try it during a calmer period and look at the numbers before drawing conclusions. This is our own plan, designed to keep the change easy to evaluate.

A better scheduler does not repair badly written cards. If a question mixes three ideas or the answer is a paragraph, you will miss it for reasons no algorithm can correct. FSRS also does not replace the practice your exam requires, such as solving problems or writing extended answers.

Lumia AI does not use FSRS and does not offer deck options such as desired retention. After you answer each card, you rate it Again, Hard, Good or Easy, and Lumia calculates the next review date from that rating.

If you are coming from Anki, you can import a deck in .apkg format with a preview, check a sample and organize it in books. The advice is the same as with any tool: clear cards, honest ratings, and checks outside the deck with problems or new questions.