Mnemosyne
Free, open-source spaced repetition flashcard software developed since 2003 as both a learning tool and a research project into long-term memory. Written in Python with Qt interface, supports rich media including images, audio, video, and LaTeX.
Score generated by AI agents based on publicly cited evidence and reviewed by the project maintainer. Not independently validated. Last assessed 2026-10-30. Score revised 2026-10-30: 5 → 7.
Score History
Timeline events are AI-curated from public reporting. Score trajectory is derived from documented events.
Peter Bienstman, now a photonics professor at Ghent University, began developing Mnemosyne in 2003 as an offshoot of the MemAid neural network spaced repetition project. The software was conceived as a free, open-source alternative to proprietary tools like SuperMemo, with a dual purpose: practical flashcard learning and academic memory research. Development rested on a single volunteer with no formal governance, but there was nothing to extract and no users to lock in.
Version 0.9 launched on February 8, 2006, introducing the core SM-2-based spaced repetition engine with basic flashcard functionality. The project collected anonymized research data from volunteering users from 2006, establishing its dual identity as both learning tool and research platform. Written in Python with a simple interface, it gained early traction among Linux users and LaTeX-savvy academics.
Version 1.0, released December 27, 2007, brought USB portability, a plugin system and automatic XML backups, and the project matured into a usable tool with a growing community. Its first research dataset was shared publicly via BitTorrent in 2009, and the Mnemogogo plugin brought reviews to Java phones in 2010. Anki, born in October 2006, developed alongside Mnemosyne as another open-source spaced repetition tool.
Version 2.0 on June 16, 2012 was a complete rewrite introducing card types, hierarchical tagging, the openSM2sync bidirectional sync protocol, graphical statistics and a card browser. Version 2.1 added a media-bundled card-sharing format, a second research dataset was published in 2014, and version 2.3 added an Android review client. This was a major investment of volunteer effort that kept the project technically viable.
Version 2.4 on December 5, 2016 migrated the codebase to Python 3 and PyQt5, and the project moved from Launchpad to GitHub in October 2017. Releases kept coming at a slow volunteer pace: Anki import (2017), Google Translate and text-to-speech (2019), a retention threshold (2021), PyQt6 with Flash support removed (2023) and pip installation (2023), with patch packages in October 2024. The first-run wizard shipped with research-log upload pre-checked, and the Android companion app drew mixed reviews.
On May 13, 2025 Peter Bienstman said he was no longer really developing Mnemosyne and that Android plugins were unsupported. Small contributions continued, including March 2026 PyPI releases and an HTTPS sync option merged in April 2026. In May 2026 he announced he had stepped down and that Lauta had taken over. By August 2026 he said the new owner was not replying and that Google would delist the Android app without updates, leaving the software intact but its upkeep in doubt.
Alternatives
Open-source spaced repetition software with shared decks, add-ons, and mobile support. Can import Mnemosyne 2.0 .db files. Free on desktop and Android; the official iOS app costs US$24.99. Offers extensive customization, with a more complex interface.
Commercial flashcard platform with shared study materials, mobile apps, and interactive study modes. Offers free accounts and paid Plus subscriptions. The displayed US annual prices are $35.99 for Plus and $44.99 for Plus Unlimited; paid plans include ad-free studying.
The original spaced repetition software lineage associated with Piotr Wozniak, whose SM-2 algorithm Mnemosyne adapts. The classic Windows product offers incremental reading; SuperMemo 20 is listed at US$68. It has a steep learning curve. The separate supermemo.com service offers language courses.
Dimensional Breakdown
Summaries below were written by AI agents based on the cited evidence. They are editorial interpretations, not independent research findings.
Dimension History
Timeline (29 events)
Peter Bienstman begins Mnemosyne development from MemAid
Peter Bienstman, now a photonics professor at Ghent University, started developing Mnemosyne as an offshoot of the MemAid neural network spaced repetition project created by David Calinski. The project aimed to create a free, open-source alternative to proprietary tools like SuperMemo while simultaneously serving as a platform for academic memory research.
Anonymized repetition-history collection documented from 2006
Mnemosyne was collecting repetition-history data for memory research by 2006. Research uploads are user-selectable: the desktop getting-started wizard lets users accept or decline anonymous log uploads.
Mnemosyne v0.9 first public release with SM-2 algorithm
Mnemosyne version 0.9 launched as the first public release, implementing a modified version of the SM-2 spaced repetition algorithm with adjustments for early and late repetitions. Written in Python, the software supported Windows, Linux, and macOS from the start.
Version 1.0 introduces plugin system and USB portability
Mnemosyne 1.0 released with a plugin system enabling extensibility, USB portability support for running from removable drives, and automatic XML backups. The plugin architecture allowed community contributions like cloze deletion card types and third-party mobile integrations.
First research dataset released publicly via BitTorrent
The Mnemosyne Project released its first accumulated research dataset via BitTorrent, making anonymized spaced repetition learning data available to any researcher. This demonstrated the project's commitment to data democratization and academic transparency, with no attempt to commercialize the collected data.
Mnemogogo plugin enables mobile review on Java phones
Version 1.2.2 added support for Tim Bourke's Mnemogogo plugin, which enabled Mnemosyne card review on Java-based mobile phones. This was an early community-driven mobile solution that exported scheduled cards to J2ME devices for review, with results synced back to the desktop.
SourceForge profiles Mnemosyne and Peter Bienstman
SourceForge published a feature titled 'When you think flash card, think Mnemosyne' on February 10, 2010, profiling Peter Bienstman. He said existing flashcard applications had been difficult to use and that he also wanted to collect data for memory research. The article previewed version 2.0 as a complete rewrite with N-sided cards, multiple categories per card, graphical statistics and components replaceable through plugins.
Version 2.0 complete rewrite with sync protocol
Mnemosyne 2.0 was a ground-up rewrite with a new modular architecture consisting of swappable components. It introduced card types (including N-sided cards), hierarchical tagging, a powerful card browser, graphical statistics, and the openSM2sync bidirectional synchronization protocol. The new design gave plugin writers maximum flexibility.
Version 2.1 adds media-bundled card sharing format
Mnemosyne 2.1 introduced a new .cards import/export format that bundles media files (images, audio, video) with card content for sharing. It also added tab-separated text export. This improved data portability and made community deck sharing more practical.
January 2014 research dataset made available for public download
Gwern Branwen announced a public download of a January 27, 2014 dump of Mnemosyne's accumulated review logs, obtained from Peter Bienstman and processed into a SQLite database. The logs were subsequently used for academic analysis of review scheduling, including a 2014 Stanford CS229 project.
Version 2.3 merges webserver and adds Android client
Mnemosyne 2.3 merged the webserver functionality into the main program and introduced an Android client that runs a webserver on the device for browser-based review with full sync protocol support. Backup creation time was also reduced by a factor of 2 or more. This was the first official mobile solution after the community-driven Mnemogogo plugin.
Version 2.4 migrates to Python 3 and PyQt5
Mnemosyne 2.4 upgraded the entire codebase from Python 2 to Python 3 and from PyQt4 to PyQt5, ensuring the project no longer depended on libraries becoming unsupported. The migration improved startup performance and added high-DPI display scaling support.
Version 2.5 introduces configurable study modes
Mnemosyne 2.5 added study modes allowing users to customize card selection strategies, such as studying only unlearned cards. The cramming scheduler was migrated into this new framework. High-DPI display scaling was enabled by default.
Project migrates from Launchpad to GitHub
The Mnemosyne Project moved its code repository and bug tracker from Launchpad to GitHub, retaining Launchpad only for translations. This modernized the development workflow and made the project more accessible to potential contributors familiar with GitHub's pull request model.
Version 2.6 adds Anki database import support
Mnemosyne 2.6 introduced import of Anki databases (both .anki2 and .apkg formats), significantly improving data portability and reducing switching costs for users moving between the two major open-source spaced repetition platforms.
PNAS study cites Mnemosyne while testing a scheduler on Duolingo data
Researchers at the Max Planck Institute published 'Enhancing Human Learning via Spaced Repetition Optimization' in PNAS. It mentions Mnemosyne as an example of spaced repetition software; its large-scale natural experiment uses Duolingo data, rather than Mnemosyne's research logs.
Version 2.7 adds Google Translate and text-to-speech
Mnemosyne 2.7 integrated Google text-to-speech and Google Translate directly into the card editing workflow, allowing users to add pronunciation audio and translations via right-click menu. The release also transitioned to 64-bit architecture exclusively and bundled Python 3.7.
Version 2.7 re-uploads all older science logs
Mnemosyne 2.7 added more error checking when uploading science logs. To avoid files missing from the past, the release re-uploaded all of a user's older logs as well, a change stated in the public changelog.
Version 2.8 adds retention threshold to cap review load
Mnemosyne 2.8 introduced an option to stop showing cards once they reach a certain number of successful retention repetitions, preventing review loads from growing indefinitely over years of use. It also made storing the client sync password optional and warned that the client stores it in plain text; the server stores a password hash.
Version 2.9 adds card history reset and webserver API
Mnemosyne 2.9 added the ability to reset a card's learning history without deleting the card, refined the cramming scheduler, and introduced a webserver API call for querying scheduled card counts programmatically. Python 3.10 compatibility was improved.
Version 2.10 migrates to PyQt6 and removes Flash support
Mnemosyne 2.10 upgraded from PyQt5 to PyQt6 and Python 3.11, removed the deprecated Flash insertion capability, and switched audio/video processing from mplayer to native Qt modules. This ensured the project continued running on modern Python and Qt frameworks rather than depending on end-of-life libraries.
Version 2.11 adds pip installation and ruby/furigana support
Mnemosyne 2.11 introduced ruby/furigana annotation rendering for Japanese and other CJK language learners, enabled pip-based installation via 'pip install mnemosyne-proj' for easier macOS and Linux deployment, and transitioned the build system to Poetry and testing to Pytest. This was the last major desktop release as of early 2026.
Hacker News discusses Mnemosyne as viable Anki alternative
A Hacker News thread about Mnemosyne drew community discussion, with users noting the project's readable codebase, Debian repository availability, and sustainable simplicity. Critics pointed to the dated UI compared to AnkiDroid and limited mobile functionality. Several commenters noted that Mnemosyne actually predates Anki.
Android client v2.7.4 updated for Google Play policy compliance
The Mnemosyne Android client received a v2.7.4 update conforming to Google's new policies for targeted Android version, privacy requirements, and full-screen behavior. The update was a maintenance release to keep the app available on Google Play rather than adding new features.
Peter Bienstman says he is no longer developing Mnemosyne further
Answering a question about custom card-type plugins on Android, Bienstman said Android plugins were not supported, especially since Google closed off Mnemosyne's data directory to outside programs. He said he was not really developing Mnemosyne further anymore, but that someone could take over and implement it since the code is available.
Mnemosyne 2.11.1 published to PyPI
The Python Package Index lists mnemosyne-proj 2.11.0.3 as uploaded on March 6, 2026 and version 2.11.1 on March 13, 2026, the first packages since October 2024.
Optional HTTPS sync merged for Android and desktop
Pull request #308 from contributor ljfp, merged April 18, 2026, added an optional 'Use HTTPS' toggle to the sync dialog on Android and the Qt desktop client, defaulting to off, so sync can run behind TLS reverse proxies. The PR notes its SSL context disables certificate verification, and it also fixed the Android build tooling.
Bienstman steps down; new maintainer Lauta takes over
Replying in the project's user group, Peter Bienstman said he had stepped down and that Lauta, a new maintainer, had just taken over the project.
Former maintainer says successor unresponsive and warns of Android delisting
A long-time user asked about Mnemosyne's future after macOS warned that the Intel-based Mac app would stop opening in a future macOS version. Bienstman replied that after transferring all ownership to Lauta, his successor had not replied to his emails, and that Google would delist the Android app unless updates were made soon. He said it did not look good and that users should be able to import their cards into Anki.
Evidence (36 citations)
D1: User Value Erosion
D2: Business Customer Exploitation
D3: Shareholder Extraction
D4: Lock-in & Switching Costs
D5: Twiddling & Algorithmic Opacity
D6: Dark Patterns
D7: Advertising & Monetization Pressure
D8: Competitive Conduct
D9: Labor & Governance
D10: Regulatory & Legal Posture
Scoring Log (6 entries)
5→7. Since Feb 2026: PyPI 2.11.0.3/2.11.1 (Mar 2026); optional HTTPS sync merged (Apr 2026); Bienstman announced he had stepped down and Lauta took over (May 16, 2026); Bienstman said in Aug 2026 that after transferring all ownership the successor was not replying and Google would delist the Android app without updates; homepage showed a hosting-suspension notice by late Oct 2026 (Wayback shows it live Oct 12; not scored, date and permanence unknown). No pricing, ads, legal or regulatory events. D1 1→2 (event: May 2025 development stop, Android plugin breakage, Aug 2026 Android delisting warning). D9 2→3 (event: ownership transferred to a single successor who, per Bienstman, stopped responding; no governance or succession process). D6 0→1 (correction: fact audit flagged the default; shipped first-run wizard pre-checks 'Upload anonymous logs' while README says off by default; verified in the 2017 and current .ui files). D8 1→0 (recalibration: no anticompetitive conduct of any kind; imports rivals' formats). Eras: MemAid Origins kept (D9 3→2 for consistency with later single-maintainer eras); First Public Release re-dated 2006-02-01→2006-02-08, 1.x Plugin Era 2007-12-01→2007-12-27, 2.0 Architecture Rewrite 2012-06-01→2012-06-16 (to release dates); Python 3 Modernization re-dated 2016-12-01→2016-12-05 and relabeled 'Python 3 Maintenance Years' (absorbs the 2017-2024 releases, D6 1 from the pre-checked wizard verified from Oct 2017); assessment-dated 'Steady-State Maintenance' (2026-02-19) re-dated to 2025-05-13 (Bienstman stops development) and relabeled 'Uncertain Succession'. Earlier eras' D6 left at 0 because the pre-2017 upload default is not documented; D8 0 in all eras. Trajectory stable→worsening (stewardship breakdown in last 12 months). Category 'Learning' looks right.
Checked all 17 removed/trimmed candidates: 1 restored, 0 partly restored, 13 confirmed removed, 3 already present. RESTORED: SourceForge 2010-02-10 profile confirms Bienstman’s reasons for creating Mnemosyne and previews a complete v2.0 rewrite; read archived original at https://web.archive.org/web/20260214092442/https://sourceforge.net/blog/when-you-think-flash-card-think-mnemosyne/. ALREADY PRESENT: corrected MemAid/Bienstman history, January 2014 dataset and academic use, and contributor-community scale with bounded README governance observation. CONFIRMED REMOVED means no acceptable support for the dropped broader assertion, not proof of its opposite. Original URLs and Wayback checked before targeted searches; no failed fetch or search snippet used as evidence. Detailed dropped subclaims and verdicts are in work/restore-ledger.json. Scores unchanged.
Checked all 61 original timeline/evidence/milestone items plus 3 alternatives and all prose/fields. ACCEPT items counted as verified; all FIX/REVIEW decided by lead. Rescued Windows praise, the January 27 dataset dump, actual per-card metrics and SourceForge article existence. Two strict invented details: (1) 2019 PNAS experiment attributed to Mnemosyne data, contradicted by its Duolingo abstract (https://pubmed.ncbi.nlm.nih.gov/30670661/); (2) Mnemosyne dataset influence on Anki SM-2 modifications, traceable only to AI-generated Grokipedia. Upload-default inconsistency and rolling historical counts are NOT classified invented. 31 unsupported assertions or closely linked claim units are itemized in work/decision-ledger.md; repeated prose claims count once. The release-cadence overstatement is corrected without calling the entire genuine no-major-release claim invented. All scores/era boundaries unchanged.