Free tools Windows power users keep installed
One-click scans. No signup required.
Keep Markdown files as the durable source of truth, then add search indexes and embeddings as disposable layers that can be rebuilt. Start with ordinary full-text search, measure where it misses your real questions, and add local semantic or hybrid retrieval only when it demonstrably improves those results.
This design gives you readable files, editor independence, and a clear privacy boundary. It also prevents a search plugin, vector database, or AI model from becoming the only place your knowledge exists.
As an Amazon Associate I earn from qualifying purchases.
What a local-first knowledge base actually is
A local-first knowledge base is a normal directory of files on your computer. The notes remain usable if an indexing program is removed, an embedding model is replaced, or an AI assistant is unavailable. Search databases, extracted chunks, vectors, and reranking results are generated copies of those files, not the canonical record.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Obsidian describes this storage model precisely: “Obsidian stores your notes as Markdown-formatted plain text files in a vault.” Its documentation explains that the vault is an ordinary folder that can be managed with other editors and file managers. See Obsidian’s data-storage documentation for the supported layout and file-handling details.
#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
Local-first does not automatically mean offline. A stack may keep your notes and index on disk while downloading a model once, calling a remote inference endpoint, or allowing an optional remote URL. Treat locality as a property of the complete data path:
- Where the original Markdown and attachments are stored.
- Where extracted text, chunks, metadata, and vectors are written.
- Where your question is processed.
- Where embedding, reranking, and query-expansion models run.
- Whether any plugin, server, or model configuration sends data elsewhere.
Audit each path instead of trusting a feature label such as “local” or “private.”
Choose a durable file layout before adding AI
Begin with a directory your operating system can back up and open without a special application. If you use Obsidian, that directory is called a vault and can be placed wherever the operating system permits. Keep one vault root rather than nesting one vault inside another: Obsidian warns that local links may not update correctly across vault boundaries.
A small structure is easier to maintain than a taxonomy designed before you have notes. For example:
knowledge-base/
├── 00-inbox/
├── projects/
├── topics/
├── meetings/
├── references/
├── index/
└── assets/Use 00-inbox for uncategorized captures, stable topic or project folders for material you revisit, and a few overview notes in index. The names are a starting convention, not a universal rule. Retrieval quality depends more on clear content and useful queries than on an elaborate folder hierarchy.
Keep notes portable
- Write in Markdown and avoid storing the only copy of important information in a proprietary database.
- Use relative links for local attachments and keep images, PDFs, and other assets in a predictable directory.
- Give notes descriptive filenames and headings. Headings become useful boundaries when a semantic index splits a long file into chunks.
- Use frontmatter or properties for fields you actually filter on, such as status, date, project, or owner. Do not create dozens of properties that you never query.
Portability is a practical recovery feature. If an index is corrupted, you should be able to open the folder in another editor, rebuild the index, and continue working.
Back up the source, not just the index
Sync and backup solve different problems. Obsidian lists Obsidian Sync, Dropbox, iCloud, OneDrive, Git, and other services as ways to synchronize a vault in its storage documentation. They have different conflict, retention, and privacy behavior; synchronization is not automatically an independent backup.
Rank #2
- 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
Keep at least one recoverable copy of the Markdown directory and attachments. An optional portable external SSD can hold a second copy, but no particular capacity, speed, or model is required by this workflow. Choose a backup method and schedule appropriate to how much work you could recreate after device loss. Generated indexes and vectors can be omitted from the backup if your chosen tool can rebuild them reliably.
Establish the keyword-search baseline
Do not begin with embeddings. First learn what the built-in search can already answer and record the misses. Obsidian’s Search documentation describes full-text queries, exact phrases, Boolean expressions, file and path filters, tags, properties, section matching, block matching, and sorting by file name or dates.
| Need | Baseline query pattern | What it controls |
|---|---|---|
| Exact wording | "incident response" |
Finds the phrase rather than either word independently. |
| Restrict to a folder | path:projects/ |
Limits results to a project subtree. |
| Restrict file content | content:retention |
Searches note text for a term. |
| Combine concepts | (migration OR upgrade) AND database |
Requires a logical relationship among terms. |
| Find a property | Property or tag filters documented by Search | Uses structured metadata instead of prose alone. |
| Find a section or block | Section and block operators documented by Search | Targets a heading or smaller Markdown block. |
Block-level matching can be slower because every file must be parsed. That delay is a useful reminder that more precise retrieval has a processing cost even before semantic indexing.
Make a small question set
Write 15 to 30 questions you genuinely ask, then list the note paths you expect each question to return. Include different forms of the same idea: a term you remember exactly, a paraphrase, an acronym, and a question whose answer is spread across several notes. Run these questions through keyword search and record which expected notes appear near the top.
This baseline prevents a common mistake: adding a vector index because semantic search sounds advanced when a path filter, property, or better heading would solve the problem with less maintenance.
How semantic and hybrid retrieval add value
Keyword search matches the words in a query. Semantic search converts text into numerical embeddings and retrieves passages whose meaning is similar, even when vocabulary differs. A reranker or query-expansion model may then reorder or broaden the candidates. Hybrid search combines lexical scores with vector similarity, preserving exact matches while recovering useful paraphrases.
The added layer normally performs these operations:
Rank #3
- 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.
- Scan Markdown files and identify changed documents.
- Split each document into chunks, often at headings or paragraph boundaries.
- Generate an embedding for each chunk and store it with the note path and metadata.
- Run lexical and/or vector retrieval for a question.
- Optionally rerank the top candidates or expand the query.
- Return snippets and paths to you or to an AI assistant.
Every step introduces state that can become stale, consume disk space, or expose information to a process you did not intend to authorize. Keep the source files independent so replacing any step remains possible.
Documented local implementation patterns
The following projects illustrate different architectures. They are examples to evaluate, not a universal ranking.
| Option | Retrieval and storage | Documented requirements or limits |
|---|---|---|
| QMD Semantic Search for Obsidian | Local vector and semantic search with BM25 fallback and automatic indexing. | Separate QMD CLI; Bun is listed in its requirements; desktop filesystem access; internet connection for the initial model download. The project’s model table lists roughly 300 MB for an embedding model, 640 MB for a reranker, and 2.2 GB for query expansion when those components are downloaded. |
| Obsidian Semantic Search MCP | Scans Markdown, splits on headings, embeds chunks with Ollama, and stores note data, full-text rows, and vectors in SQLite. Supports hybrid vector plus SQLite FTS5 and keyword-only modes. | The reviewed repository describes read, search, index, and status tools without note write or delete tools. It can read the filesystem without Obsidian running. The page labels version 0.3.1 an early preview; schemas and ranking can change before 1.0. Configuration permits a remote URL, so inspect settings rather than assuming every deployment is offline. |
| Vault Knowledge Base (OKB) | An Obsidian plugin wrapping a separate local obsidian-kb CLI. The CLI combines lexical, semantic, and graph-aware related-note discovery and can expose retrieval through MCP. |
Requires Obsidian Desktop and the locally installed CLI according to the community listing; it does not support Obsidian Mobile. |
For project-specific details and changes, consult the QMD repository, the semantic MCP repository, and the OKB listing. Model sizes and installation requirements are version-sensitive.
QMD-based indexing
QMD is appropriate when you want semantic retrieval inside a desktop Obsidian workflow and accept a separate command-line dependency. Its documented BM25 fallback is valuable when an embedding component is unavailable or an exact term matters. Budget disk space for the components you select: the listed sizes total several gigabytes if embedding, reranking, and query expansion models are all downloaded. The initial download means the setup is not a first-run offline install.
Read-only semantic MCP
The MCP pattern separates retrieval from note editing. A server can index a folder, answer search and read requests, report status, and leave writes and deletes unavailable. That narrower permission boundary is useful when an AI agent needs context but should not silently rewrite your files. The documented implementation stores full-text data and vectors in SQLite and can use keyword-only retrieval when semantic search is not helping.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Vault Knowledge Base
OKB adds graph-aware related-note discovery alongside lexical and semantic methods through a local CLI. Its desktop-only requirement matters if your normal workflow includes a mobile app. Treat the plugin and CLI as two maintained components: both must remain installed and compatible for the workflow to function.
Build the system in a controlled sequence
1. Create and inspect the source directory
- Create a normal directory for Markdown notes and attachments.
- Open it in your chosen editor or vault interface.
- Create a few folders and overview notes, but avoid designing an elaborate taxonomy before you have real material.
- Confirm that a text editor and file manager can open the files outside the primary application.
2. Put backup and recovery first
- Choose a synchronization or backup method and verify that it preserves both Markdown and attachments.
- Restore a sample note to a separate location so you know the process works.
- Decide whether generated indexes belong in backups; normally they are rebuildable and need not be treated as authoritative.
3. Run the lexical test set
- Write down representative questions and expected note paths.
- Use exact phrases, Boolean combinations,
file:,path:,content:, tags, properties, section, and block filters where appropriate. - Record misses, irrelevant results, and slow queries.
- Improve headings, filenames, or metadata when a simple structural change solves a miss.
4. Select a semantic layer for a specific gap
Choose a documented implementation only after you can name the gap: paraphrased questions, inconsistent terminology, cross-note discovery, or an agent that needs read-only retrieval. Check desktop versus mobile support, whether the editor must run, model downloads, disk use, refresh behavior, and permission scope before installing.
Rank #4
- 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
5. Install and configure without guessing
Follow the selected project’s current README and requirements rather than copying commands from an older guide. QMD requires its separate CLI and lists Bun; the MCP server requires its own runtime and Ollama-based embedding setup; OKB requires its CLI alongside the desktop plugin. Keep the index directory separate from the source tree when the tool permits it.
6. Audit locality before indexing private notes
- Inspect the configured model endpoint and embedding provider.
- Check where query text, extracted snippets, vectors, and logs are written.
- Review included and excluded paths.
- Confirm whether the AI interface has read-only tools or can write and delete files.
- Run a test with a non-sensitive folder and inspect network and filesystem behavior according to your operating system’s controls.
7. Build the first index and verify freshness
Index a small subset first. Open a newly added note, edit an existing heading, and delete a test file; then run the tool’s documented refresh or status action. Confirm that the changed path appears, the deleted path disappears, and snippets show current text. Different tools use different triggers, so do not assume that saving a file immediately updates every index.
Crashes, 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 minutePC 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 & 118. Compare keyword, semantic, and hybrid modes
Run the same question set through each available mode. Inspect both the top results and the misses. Keep semantic or hybrid retrieval only when it improves the questions that matter enough to justify its model storage and maintenance.
Measure whether semantic search is actually better
Use your own vault as the test corpus. For each question, define the relevant note paths before looking at results. The semantic MCP project documents an evaluation command that reports Recall@K and MRR:
- Recall@K: the proportion of expected relevant notes found within the first K results.
- Mean Reciprocal Rank (MRR): how high the first relevant result appears, averaged across questions.
These metrics are useful controls, not universal benchmarks. A vault about code, research, or meetings will have different vocabulary and relevance judgments. Include hard cases such as synonyms, abbreviations, short queries, and notes that share many common words.
| Test observation | Likely decision |
|---|---|
| Keyword search already returns the expected note first. | Keep the simpler lexical path for that question. |
| Semantic search finds paraphrased notes but adds many unrelated passages. | Try hybrid weighting, better headings, or narrower filters before adding a reranker. |
| Hybrid improves Recall@K and MRR on repeated real queries. | Keep it, document the model and index version, and schedule refreshes. |
| No meaningful improvement, or gains occur only on contrived questions. | Remove the extra layer or leave it disabled until the corpus changes. |
The MCP project specifically recommends avoiding additional ranking complexity or embedding cost unless hybrid search measurably beats the keyword baseline. That is sensible engineering: a more elaborate pipeline is justified by retrieval results, not by the presence of an AI label.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Keep the complete data path local when that matters
A useful audit records each artifact and its destination:
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.
| Artifact | Questions to answer |
|---|---|
| Markdown and attachments | Which folder contains the authoritative copy? Is it included in backups? |
| Chunks and metadata | Are snippets copied into a database or log? Is the database encrypted or excluded from sync? |
| Embeddings | Which model creates them, and where does that model run? |
| Query text | Does the search UI, MCP server, or model endpoint transmit it remotely? |
| AI responses | Are retrieved snippets sent to another service to generate an answer? |
| Configuration | Can a remote URL be enabled, and is it disabled or intentionally controlled? |
The documented semantic MCP configuration allows a remote URL even though its example workflow uses local Ollama embeddings and a local index. Read the actual configuration in your deployment. QMD explicitly requires internet access for its first model download, so describe that setup as local model use after download, not as permanently disconnected operation.
Use the narrowest useful permissions
If an assistant only needs to find and quote notes, expose search and read operations without write or delete access. A write-capable agent can reorganize, overwrite, or remove files, so it deserves a separate review and backup plan. Exclude private directories unless they are intentionally part of the corpus, and verify that exclusions survive a full rebuild.
Maintenance: indexes are derived data
Plan for refreshes when notes change, exclusions change, models change, or an index format changes. The documented MCP design describes action-driven indexing before a search after changes and a full rebuild after an embedding-model or index-format change. Other tools may refresh automatically, on a schedule, or only when manually triggered.
- Keep a note of the index location, model name, and tool version used to create it.
- Run a small incremental refresh after ordinary edits and test status output.
- Perform a full rebuild after changing embedding models or index schemas.
- Delete and recreate the derived index if it contains stale paths or corrupted vectors.
- Review excluded paths and traversal protections after upgrades.
Because an index can contain note snippets and vectors, treat it as sensitive derived data even though it is not the source of truth.
Organization helps at scale, but it is not an accuracy guarantee
Folders, headings, properties, and consistent naming reduce the amount of material a retriever must inspect. They also make lexical filters and chunk boundaries more useful. Do not infer that organization alone makes an AI answer correct.
A 2026 preprint, Filesystem-Based Memory for LLM Agents: Organization, Evolution, and Sustainability, reports that organized stores roughly halve retrieval cost when the material is large. In its growth study, organization eroded for all but the strongest management agent, and measured organization did not by itself improve answer quality. Those findings apply to the paper’s benchmarks and experimental setup, not as a promised speedup for an individual Markdown vault; read the paper and its limitations before generalizing them.
Troubleshoot the common failure modes
The right note is not found
- Run the exact phrase and a path or property filter to establish whether the file is indexed at all.
- Check that the folder is included and that an exclusion rule is not too broad.
- Inspect headings and chunk boundaries; a key fact split away from its context may rank poorly.
- Add a synonym or acronym to the note if it is genuinely useful, then rebuild or refresh the index.
Results are stale
- Check the index status and last-refresh time.
- Trigger the tool’s documented incremental action.
- Rebuild after changing an embedding model or index format.
- Confirm that the search process can read the same filesystem path where the editor saves notes.
Search is slow
- Remember that block search may parse every file.
- Limit broad queries with
path:, properties, or date filters. - Reduce the indexed corpus by excluding caches, generated files, and intentionally private folders.
- Measure whether reranking or query expansion improves results enough to justify its latency and disk use.
Model storage is unexpectedly large
Check which optional components were downloaded. The QMD listing gives separate approximate sizes for embeddings, reranking, and query expansion; installing all three can require several gigabytes. Remove unused components only through the project’s supported process so the index does not reference missing models.
Recommended Free Tools
Mobile access is incomplete
Desktop filesystem access is a stated requirement for QMD, and the OKB listing says it does not support Obsidian Mobile. A desktop semantic index may still coexist with mobile note editing through synchronization, but mobile search behavior depends on the specific application and is not guaranteed by the desktop index.
A practical stopping rule
Stop at lexical search when it answers your real questions with acceptable effort. Add semantic retrieval for a defined miss, validate it on known notes, and retain it only when Recall@K, MRR, or visible result quality improves enough to pay for model downloads, indexing time, and maintenance. Keep Markdown readable outside the tool, keep backups independent, and treat every index and AI permission as replaceable infrastructure rather than as your knowledge base itself.
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.
