The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →MemPalace is real, open-source software publicly associated with Milla Jovovich and developer Ben Sigman—but its original “perfect” benchmark headlines were methodologically overstated. The project’s 100% LongMemEval figure measured whether a relevant session appeared in the top five results, not whether an AI answered the benchmark correctly end to end. Its original LoCoMo result used a top-50 retrieval setting that could return every candidate session. Current project documentation reports lower, more clearly qualified retrieval results.
What Milla Jovovich’s project actually is
MemPalace is a local-first memory and retrieval system for AI conversations and project files. It keeps source material, indexes it, and lets a user search old context or load relevant material into a new session. The public repository describes commands for mining files, searching memories, and running a “wake-up” workflow: MemPalace on GitHub.
The project’s public story, dated April 6, 2026, presents Jovovich as associated with the project alongside technical collaborator Ben Sigman: the MemPalace project story. Available evidence does not establish which modules Jovovich personally wrote, how much code was generated by Claude Code, or the precise division of labor. It is therefore inaccurate to describe her as having single-handedly coded the system.
How the “memory palace” design works
MemPalace uses a hierarchy intended to make a large archive navigable:
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minute#1 Best Overall
- Get NVMe solid state performance with up to 1050MB/s read and 1000MB/s write speeds in a portable, high-capacity drive(1) (Based on internal testing; performance may be lower depending on host device & other factors. 1MB=1,000,000 bytes.)
- Up to 3-meter drop protection and IP65 water and dust resistance mean this tough drive can take a beating(3) (Previously rated for 2-meter drop protection and IP55 rating. Now qualified for the higher, stated specs.)
- Use the handy carabiner loop to secure it to your belt loop or backpack for extra peace of mind.
- Help keep private content private with the included password protection featuring 256‐bit AES hardware encryption.(3)
- Easily manage files and automatically free up space with the SanDisk Memory Zone app.(5). Non-Operating Temperature -20°C to 85°C
- Wings: broad domains or top-level areas.
- Rooms: topic or project groupings.
- Halls and drawers: additional organizational layers for stored material.
Under that metaphor are ordinary retrieval components. The project describes local vector search, ChromaDB, SQLite metadata, and retained source text. The metaphor is an information-architecture choice; it is not evidence that the software reproduces human memory.
Its central design decision is to preserve conversations or files rather than first asking an LLM to compress everything into a small list of “important facts.” That creates an inspectable source record, while derived indexes can be rebuilt or searched later.
Why AI memory is difficult
Chat sessions are commonly isolated. Summaries can remove alternatives, dates, or qualifications, and extracted facts can become stale. A memory system must handle at least four different jobs:
- Storage: retain source conversations or files.
- Retrieval: find sessions or passages related to a question.
- Reasoning: reconcile retrieved information and produce an answer.
- Persistent context: automatically bring useful information into a future session.
MemPalace mainly makes a case for retaining more source material and improving retrieval. That can reduce information loss during ingestion, but it does not by itself prevent irrelevant passages, contradictory versions, privacy exposure, or answer-generation errors.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Rank #2
- Solid state performance with up to 800MB/s read speeds in a portable drive. (Based on internal testing; performance may be lower depending on host device, interface, usage conditions and other factors. 1MB=1,000,000 bytes.)
- Back up your content and memories on a storage solution that fits seamlessly into your mobile lifestyle.
- Take it with you on your adventures—up to two-meter drop protection means this durable drive can take a beating. (Based on internal testing.)
- Secure it to your belt loop or backpack for extra peace of mind thanks to the tough rubber hook.
- From Sandisk, a brand professional photographers trust to take on assignments.
What the launch benchmarks claimed
Early project messaging promoted 100% results on LongMemEval and LoCoMo, strong comparisons with commercial memory products, local operation without a required subscription, and very high compression using the project’s AAAK format. The detailed objections and the project’s subsequent qualifications are documented in the benchmark issue.
| Claim or configuration | Reported result | What it measures | Important qualification |
|---|---|---|---|
| LongMemEval raw | 96.6% R@5 | Whether the gold session was retrieved in the top five | No LLM; not end-to-end answer accuracy |
| LongMemEval hybrid, held-out | 98.4% R@5 on 450 questions | Session-retrieval recall | More defensible, but still not QA accuracy |
| Original LongMemEval headline | 100% | Retrieval-style recall | Included targeted fixes for failing questions |
| LoCoMo raw | 60.3% R@10 | Session retrieval without reranking | Current repository figure |
| LoCoMo hybrid | 88.9% R@10 | Session retrieval without reranking | Current repository figure |
| Original LoCoMo headline | 100% | Reranked result after top-50 retrieval | Top-50 exceeded the candidate-session counts in the criticized run |
The current benchmark documentation records the 450-question held-out split and the raw baseline: MemPalace benchmark documentation.
Why the LongMemEval “100%” was disputed
It tested retrieval, not answering
The criticized runner concatenated user turns for each session, embedded those sessions, retrieved the top five, and checked whether a labeled gold session appeared in the set. It did not generate a final answer or apply LongMemEval’s answer judge. That is recall_any@5, not conventional end-to-end LongMemEval QA.
A retriever can find the right session while an agent answers incorrectly. It can also return several plausible sessions while failing temporal or multi-hop reasoning. For example, if a user changed a software preference six months later, retrieving both statements is not enough; the system must identify which one is current.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Rank #3
- MADE FOR THE MAKERS: Create; Explore; Store; The T7 Portable SSD delivers fast speeds and durable features to back up any endeavor; Build your video editing empire, file your photographs or back up your blogs all in an instant
- SHARE IDEAS IN A FLASH: Don’t waste a second waiting and spend more time doing; The T7 is embedded with PCIe NVMe technology that brings fast read and write speeds up to 1,050/1,000 MB/s¹, making it almost twice as fast as the T5
- ALWAYS MAKE THE SAVE: Compact design with massive capacity; With capacities up to 4TB, save exactly what you need to your drive – from large working files to game data and everything in between
- ADAPTS TO EVERY NEED: Whether using a PC or mobile phone, count on the T7 for extensive compatibility²; It’s a true team player when it comes to heavy-duty application usage or file-saving
- HI RESOLUTION VIDEO RECORDING: Record Ultra High Resolution (4K 60fs) videos directly onto the T7 Portable SSD with your favorite camera or mobile devices; Supports iPhone 15 Pro Res 4K at 60fps video and more³
The perfect result used question-specific logic
The benchmark issue describes fixes aimed at particular failures, including a quoted-phrase boost, a person-name boost, and pattern matching around phrases such as “I still remember” and “when I was in high school.” The project’s benchmark notes characterize this as teaching to the test. Benchmark tuning is not automatically illegitimate, but a tuned result should not be presented as an untouched, general-purpose perfect score.
The cleaner held-out number was lower
The more useful comparison is 96.6% R@5 for raw semantic retrieval and 98.4% R@5 on a 450-question held-out hybrid split. Those are retrieval results, not proof that the system answered every question correctly.
Why the LoCoMo “100%” was disputed
The criticized LoCoMo run used top_k=50. The issue reports that the relevant conversations contained roughly 19–32 candidate sessions. When a retriever may return 50 items from a pool smaller than 50, it can return the entire pool. An LLM reranker then chooses among all available sessions rather than solving a difficult top-10 retrieval problem: benchmark methodology discussion.
The current repository presents 60.3% R@10 for raw retrieval and 88.9% R@10 for its hybrid, no-rerank configuration. LoCoMo also contains questions whose answers may not be present in the conversation, so any complete-set score requires explicit handling of unanswerable cases; see the related project issue.
Rank #4
- Capacity Display Variance: 500GB external ssd often appears as around 465GB on Windows. MacOS can show full 500 GB capacity. This is binary calculation difference and doesn’t affect SSD hard drive actual physical storage
- 1050 MB/s Speed: Instantly access to your files with blazing-fast 10Gbps external SSD read up to 1050MB/s and write up to 1000MB/s. LED Light indicates USB SSD instant activity
- Data Security: Solid state drives S.M.A.R.T. health diagnostics and adaptive TRIM optimizing data block management ensures consistent write speeds and extends the longevity of the portable SSD
- USB-C & USB-A Cable: Both cables featuring rapid USB 3.2 Gen2, this USB SSD effortlessly bridges devices, enabling seamless cross-platform file transfers and backup between computers, smartphones, tablets and iPhone
- Always Fast: No slowdowns for large file transfers. With SLC caching (25% of current available capacity allocated as high-speed cache), this external SSD delivers steady 10Gbps for transfers within the cache capacity
What “local and free” means
MemPalace can run as a local open-source retrieval system without a required cloud subscription in its no-LLM mode. Local does not mean costless: users still supply storage, compute, backups, security, and maintenance. LLM reranking may require a hosted or paid model, depending on configuration. The benchmark issue says the 100% runs used paid Claude calls, while the no-API mode produced lower scores.
Local storage limits transmission to a vendor, but it does not eliminate risk. A laptop, database, backup, or embedding cache can still expose sensitive conversations. Retaining everything also preserves incorrect or confidential material that a summarization pipeline might have discarded.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Can developers try it?
The repository currently gives this installation path:
git clone https://github.com/MemPalace/mempalace.git
cd mempalace
uv sync --extra devAn editable alternative is:
pip install -e ".[dev]"Example workflows shown by the project include:
mempalace mine ~/projects/myapp
mempalace mine ~/.claude/projects/ --mode convos --scope --wing per-project
mempalace search "why did we switch to GraphQL"
mempalace wake-upTo reproduce the documented LongMemEval runner, the repository gives:
Best Value
- NEARLY 2X FASTER THAN OUR PREVIOUS GENERATION(8) – move 1,000 high-res photos in under 60 seconds(6) with up to 2000MB/s transfer speeds(2).
- IP65 RATING AND UP TO 3M DROP PROTECTION(3) – protects against spills and drops.
- POCKET-SIZED – fits easily in pockets and small bags.
- SPACE TO OWN YOUR AI CONTENT – speed and capacity to download your high-res clips and photo edits.
- 256-BIT AES ENCRYPTION(4) – helps keep private files secure with password protection.
uv run python benchmarks/longmemeval_bench.py
/path/to/longmemeval_s_cleaned.jsonIntegration names and plugin manifests can change, so developers should follow the current repository guide rather than assume that every Claude Code, Gemini CLI, MCP, or local-model path remains identical.
What survives the benchmark controversy
- Keeping verbatim source material can reduce losses caused by aggressive fact extraction.
- A local-first design can be useful for privacy-sensitive workflows and researchers.
- Separating source records from indexes makes inspection and re-indexing easier.
- Publishing scripts and result files allows others to reproduce or challenge measurements.
But the evidence does not show that the palace hierarchy itself caused the reported gains. An independent reproduction issue notes that the raw LongMemEval runner created a fresh ChromaDB client and did not exercise the wings, rooms, halls, or palace code paths: independent reproduction concerns. Useful ablations would compare flat storage, hierarchy, keyword and temporal boosts, embedding models, and LLM reranking.
How to evaluate an AI-memory system fairly
- Use the same dataset version, question subset, and candidate corpus.
- Report retrieval recall and final answer accuracy as separate metrics.
- Fix retrieval depth, reranker, judge model, rubric, and prompt.
- Include multi-hop, temporal, contradictory, adversarial, and unanswerable questions.
- Publish per-question failures, not only an aggregate percentage.
- Measure latency, token use, storage growth, API cost, and offline behavior.
- Test deletion, permissions, backups, prompt injection in stored memories, and accidental exposure of sensitive context.
Competitor benchmark repositories such as Mem0’s memory benchmarks illustrate why datasets, question counts, top-k settings, and configurations must be defined before numbers are compared. MemPalace’s retrieval figures cannot be fairly ranked against another product’s end-to-end QA accuracy.
Bottom line: an interesting project, not a proven perfect system
MemPalace is neither a fake product nor a demonstrated industry-beating memory platform. It is a real open-source experiment with a credible local-retrieval baseline and a clear “keep the source text” philosophy. The original marketing blurred retrieval, reranking, and end-to-end answering, while the LoCoMo setup made the retrieval challenge unusually permissive. The current 98.4% held-out LongMemEval R@5 and 88.9% LoCoMo R@10 figures are more informative—but they still answer a narrower question than whether an agent consistently remembers and responds correctly.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
