NeuralPGN: A Local-First Chess Opening Trainer Powered by the FSRS Spaced Repetition Algorithm

When Spaced Repetition Meets Chess

For dedicated chess enthusiasts, memorizing an opening repertoire has always been a tough nut to crack. You spend hours studying a variation, then completely blank on the critical move in an actual game weeks later — it's a frustration almost every player knows. Traditional review and puzzle-drilling methods lack scientific memory scheduling, which naturally limits their effectiveness.

NeuralPGN, an open-source project that recently appeared on Hacker News, was built to solve exactly this problem. It's a "local-first" chess opening trainer that brings the FSRS spaced repetition algorithm to chess study, wrapped in a lightweight desktop experience built with the Tauri framework.

FSRS: Turning the Forgetting Curve Into Your Training Rhythm

What Is FSRS, and Why Is It Better Than SM-2?

FSRS (Free Spaced Repetition Scheduler) has emerged as a dark horse in the memory-learning space in recent years. To understand its value, you need to know what it replaces.

The historical roots of SM-2: SM-2 (SuperMemo Algorithm 2) was designed by Polish researcher Piotr Woźniak in 1987 and is the foundational algorithm behind modern spaced repetition software. Anki has used SM-2 since its launch in 2006 and built a massive user base on top of its open-source ecosystem. SM-2's core mechanic calculates the next review interval using a fixed formula based on the user's self-rating (0–5). It's simple and effective, but lacks adaptability for individual differences — a limitation that drove the research behind next-generation algorithms like FSRS.

FSRS was proposed by Chinese developer Jarrett Ye (叶峻峣) in 2022 and was officially integrated into Anki as an optional scheduler in 2023. Its core breakthrough is a memory model based on neural network fitting — the DSR model (Difficulty, Stability, Retrievability) — which uses three parameters to quantify the current memory state of each card. Trained on hundreds of thousands of real user review sessions, FSRS can more accurately predict the probability of forgetting and supports personalized parameter tuning, making its scheduling genuinely fit each learner's actual forgetting curve rather than applying a one-size-fits-all multiplier.

What This Means for Chess Players

Applying FSRS to chess opening training is a clever idea. Every opening variation and every critical move can be treated as a "memory card." The system continuously tracks your recall of the correct move in a specific position and schedules a review right at the moment you're about to forget — maintaining the longest possible retention with the minimum amount of practice.

Opening knowledge has a classic "use it or lose it" quality. An amateur player might have dozens of opening systems in their repertoire, but without periodic review, memory quickly crumbles. The core value of FSRS is that it doesn't waste your time on moves you already know cold — it precisely directs your training resources to the memories that are on the verge of slipping away, genuinely maximizing learning efficiency.

Technical Architecture: Tauri + Local-First, Lightweight and Private

Why Tauri Instead of Electron?

NeuralPGN is built with the Tauri framework. To appreciate this choice, it helps to understand the alternative.

Electron was launched by GitHub in 2013. By bundling a full Chromium browser engine with a Node.js runtime, it lets web developers quickly build cross-platform desktop apps — VS Code, Slack, and Discord are all built on it. However, a typical Electron app's installer often exceeds 100MB, and its runtime memory usage easily surpasses 200MB. It has long been criticized as a "resource hog."

Tauri released its 1.0 version in 2021 with a fundamentally different approach: it reuses the operating system's built-in WebView (WebKit on macOS, WebView2 on Windows, WebKitGTK on Linux), while backend logic is implemented in Rust. This reduces the final bundle size to 3–10MB, cuts memory usage by roughly 50–80%, and Rust's memory safety guarantees also provide stronger security.

For a training tool that needs to run persistently and be invoked frequently, this means faster startup times and lower system resource consumption — a noticeably better user experience than a bloated Electron app.

The Three Benefits of a Local-First Architecture

"Local-first" is another core design principle of NeuralPGN. This concept was formally articulated by the Ink & Switch research lab in their 2019 paper of the same name. The central thesis is that user data should be stored on the local device first, with cloud sync as an optional capability rather than a mandatory dependency — a direct response to the loss of data sovereignty under SaaS subscription models, where users risk losing their data if a service shuts down. Knowledge management tools like Obsidian and Logseq have risen to prominence in recent years with local-first as a core selling point, demonstrating the concept's strong influence in the indie developer community.

NeuralPGN fully embraces this philosophy: all game data, training records, and memory states are stored locally on your device, require no cloud upload, and work perfectly offline. This delivers three key advantages:

Privacy: Your opening repertoire and training data are entirely under your own control and are not collected by third parties.
Offline access: Training is never interrupted, whether you're on a plane or in an area with poor connectivity.
Data freedom: PGN is a universal open format that supports free import and export, completely eliminating the risk of platform lock-in.

As data privacy becomes an increasingly important concern, local-first architecture is becoming the preferred choice for a growing number of indie developers.

PGN: The Power of an Open Standard

NeuralPGN uses PGN (Portable Game Notation) as its core data format, a choice worth noting in its own right. PGN is the universal open standard for chess, established by Steven Edwards in 1993. It stores game information as plain text, including tag pairs (event, date, player names, ELO ratings, etc.) and move sequences in Standard Algebraic Notation (SAN). Its openness makes it the universal export format for major platforms like Lichess and Chess.com. Choosing PGN means players don't need to reorganize their existing research — existing repertoires can be imported directly — and it completely eliminates the platform lock-in risk associated with proprietary data formats.

Who Is NeuralPGN For?

NeuralPGN is primarily aimed at three types of users:

Serious players building systematic opening repertoires: Those who need to maintain large opening libraries will find that FSRS significantly reduces the burden of memorization.
Privacy-conscious technical players: The local-first and open-source nature of the tool is a natural fit for this audience.
Heavy PGN users: Those who can seamlessly import their existing research into the training system.

A typical workflow looks something like this: import your PGN opening library → the system breaks variations down into training units → complete daily reviews scheduled by FSRS → the system tracks your mastery and dynamically adjusts the plan.

A Broader Takeaway: The Promise of Cross-Domain Algorithm Transfer

As a project that just debuted on Hacker News, NeuralPGN hasn't yet attracted a large community following, but the product logic behind it is worth paying attention to.

The potential of cross-domain algorithm transfer: Taking a mature cognitive science algorithm (FSRS) and applying it to the vertical use case of chess training is the kind of "cross-domain combination" that often produces surprisingly useful tools. The same idea could easily extend to language learning, instrument practice, programming knowledge retention, and any other domain that requires maintaining memories over the long term.

Tech stack trends for indie developers: Tauri + Rust + local-first is becoming the go-to stack for a new generation of indie desktop apps. It strikes a strong balance between development efficiency, application performance, and user privacy — making it especially friendly for small teams and solo developers.

Of course, as an early-stage project, NeuralPGN also faces real challenges: lowering the barrier to preparing PGN data, polishing a more intuitive training interface, and building an active user community — all areas that will need ongoing attention.

Conclusion

NeuralPGN is a small but elegant open-source project. Rather than piling on features, it stays focused on one clear, specific pain point: helping chess players memorize openings efficiently with a scientifically-grounded algorithm. For players who want to systematically improve their opening knowledge while keeping their data private, this local-first, algorithm-driven chess training tool offers a compelling new option worth trying.

It also reaffirms a timeless truth: the most genuinely useful tools are usually born from a deep understanding of a specific problem — not from blindly chasing the latest technology trend.

Key Takeaways

NeuralPGN is an open-source chess opening trainer that applies the FSRS spaced repetition algorithm to chess study
FSRS, developed in 2022 and integrated into Anki in 2023, uses a DSR memory model to more accurately predict forgetting and personalize review scheduling compared to SM-2
Built with the Tauri framework, NeuralPGN is significantly more lightweight than Electron-based alternatives (3–10MB bundle, 50–80% less memory usage)
A local-first architecture ensures all data stays on your device — no cloud dependency, full privacy, and offline access
PGN support as the core format eliminates platform lock-in and lets players import existing repertoires directly
Best suited for serious players with large opening repertoires, privacy-conscious technical users, and heavy PGN users