Is your feature request related to a problem? Please describe.
I think Anki and modern SRS algorithms could become much more personalized by analyzing the long-term learning behavior of each individual user.
People do not learn in exactly the same way. Two users can use the same deck and the same SRS algorithm but have very different retention, forgetting patterns, response times, difficulty distributions, and optimal review workloads.
At the moment, a large amount of valuable learning data is already being generated through reviews, but it is difficult for users to turn that data into a comprehensive model of their individual learning behavior and then use it to improve their SRS configuration.
I would like Anki to provide a built-in way to analyze this data and optionally use an external AI as an analysis and optimization tool.
Describe the solution you’d like
I would like to propose a new optional feature: Personal Learning Profile + AI-Assisted SRS Optimization.
Anki could analyze the user’s historical review data and generate a structured, privacy-conscious export containing measurable learning characteristics such as:
-
Retention and recall rate
-
Again / Hard / Good / Easy distribution
-
Response time
-
Lapse and relearning frequency
-
Review workload
-
New-card workload
-
Card difficulty
-
Deck-specific performance
-
Tag-specific performance
-
Performance at different review intervals
-
Short-term vs. long-term retention
-
Daily and weekly consistency
-
Changes in performance over time
-
Cards that are repeatedly forgotten
-
Possible over-review or under-review patterns
-
The effectiveness of different scheduling configurations
This could be exported as a machine-readable JSON “Learning Profile”.
An optional built-in AI integration, or an AI-compatible export/import workflow, could then allow an external AI to analyze the profile.
The AI would be asked to answer questions such as:
-
Which parts of the current scheduling strategy appear effective?
-
Where is the user consistently struggling?
-
Is the current review workload unnecessarily high?
-
Are some cards or categories being reviewed too frequently?
-
Are long intervals causing excessive forgetting for this particular user?
-
Should the target retention or other scheduling parameters be adjusted?
-
Which changes are supported by the user’s historical data?
The AI could then return a strictly validated configuration containing only parameters that Anki explicitly allows to be changed.
Before applying the configuration, Anki could show:
Current value → Proposed value → Difference
and require user confirmation.
Anki should also keep configuration versions so the user can later compare whether a change actually improved retention, reduced workload, or simply increased repetition.
For example:
Configuration A
Retention: 91%
Average reviews/day: 95
Configuration B
Retention: 92%
Average reviews/day: 79
This would allow the system to evaluate scheduling changes based on actual long-term results rather than assuming that a theoretically different configuration is automatically better.
The important concept is that the SRS should remain evidence-based, but the parameters could become personalized to the individual’s observed learning behavior.
The AI should not diagnose the user’s brain or make medical/psychological claims. It should analyze observable learning data only.
The existing Anki/FSRS scheduling system should remain fully functional without AI. AI would be an optional optimization layer rather than a requirement.
Privacy would also be important: the export should contain only learning-related data, should be anonymizable, and should never automatically send anything to an external service without the user’s explicit action.
I am personally interested in this idea and would be happy to discuss it further or provide more details about the concept. I am a Persian-speaking user, and my spoken/written English is still quite limited, so direct English communication can be difficult for me. If anyone from the Anki team would like to discuss the idea with me, I can be contacted on Telegram:
Telegram: @IsaoSANdesuka
Describe alternatives you’ve considered
I have considered manually analyzing my Anki statistics and manually changing SRS parameters, as well as building a separate application that reads Anki data and performs this type of analysis.
However, I think this capability would be much more valuable if it were integrated directly into Anki.
Anki already has the most important asset for this idea: a large and detailed history of real user learning behavior.
Instead of only asking:
“Which cards are due?”
the system could also help answer:
“How does this particular user learn best, based on their own long-term data?”
I believe this could make Anki’s already powerful SRS system significantly more personalized while keeping the existing workflow intact.