Full-text search

As personal vaults grow from dozens to thousands of documents, the utility of a knowledge base depends almost entirely on the speed and precision of its retrieval engine. If searching for an architectural recommendation or a customer feedback quote takes several seconds or requires typing an exact file name, the friction discourages capturing detailed information in the first place.

Many modern productivity apps offload search indexing to remote cloud databases. When you type a query, your search terms and note contents are transmitted across the internet to multi-tenant clusters.

Oyma takes the uncompromising local approach. Powered by an embedded SQLite FTS5 (Full-Text Search) engine proven in src/main/search/searcher.ts and architectural decision ADR 0014, Oyma indexes your entire archive directly on your Mac, delivering sub-10-millisecond query results across tens of thousands of documents.

High-performance local indexing with SQLite FTS5

SQLite FTS5 is an industrial-strength full-text indexer optimized for raw search performance and minimal disk overhead. Instead of building a resource-heavy search service or running a persistent Java runtime, Oyma leverages SQLite’s proven virtual table architecture directly within the macOS application process.

Whenever you create, edit, or rename a Markdown document:

1. Background tokenization: An incremental filesystem watcher detects changed blocks, tokenizing headings, paragraphs, and metadata without interrupting your typing.
2. Porter stemming and normalization: English words are stemmed (so searching for “recording” naturally matches “record” and “recorded”) while preserving exact character casing when phrase quotes are supplied.
3. BM25 relevance ranking: Search results are ranked using the Okapi BM25 statistical algorithm, scoring document relevance based on term frequency, inverse document frequency, and document length.
4. Instant snippet extraction: Matched tokens are highlighted in real-time context snippets, showing you the exact sentence where your term occurs before you even click to open the note.

Because the SQLite database resides alongside your vault metadata on high-speed NVMe storage, queries execute in single-digit milliseconds, returning results as quickly as you can press keys on your keyboard.

Advanced query operators for precision retrieval

Simple keyword matching is rarely enough when searching through years of project history and hundreds of meeting transcripts. Oyma equips power users with expressive query syntax:

• Exact phrase matching: Wrap terms in double quotes ("zero-bot architecture") to locate precise verbatim phrases without loose word scattering.
• Boolean operators: Combine terms using uppercase boolean logic:
  • Ollama AND Llama3: Notes containing both keywords.
  • Whisper OR Transcription: Documents touching either topic.
  • Security NOT Cloud: Exclude irrelevant articles from targeted audits.
• Prefix matching: Append an asterisk (token*) to match variations like “token”, “tokenize”, “tokenization”, and “tokens”.
• Folder scoping (path:): Constrain searches to a particular directory hierarchy, such as path:meetings/2026 or path:specs/infrastructure.
• Tag scoping (tag:): Isolate notes labeled with specific YAML frontmatter tags, such as tag:interview or tag:architecture.

These operators can be combined fluidly, allowing you to execute laser-focused queries like path:meetings tag:client "budget approved" with instant results.

Zero cloud footprint: Total search privacy

Your search queries reveal the most intimate contours of your thinking. What you search for exposes active vulnerabilities, internal confidential investigations, salary reviews, and emerging patent applications.

Because Oyma‘s search engine is strictly local:

• No query logs are generated on remote telemetry servers.
• No third-party analytics scripts observe your search cadence or frequency.
• The search index operates completely offline; if you disconnect your network or turn on Airplane Mode, search performance remains 100% identical.

For organizations with stringent security postures or legal discovery firewalls, local search ensures that full-text indexing complies inherently with strict data retention and non-disclosure standards.

Incremental indexing and minimal disk overhead

Large knowledge bases often degrade in performance when search indexing processes consume excessive CPU cycles or bloat system RAM.

Oyma‘s SQLite FTS5 implementation avoids these common pitfalls through disciplined architectural boundaries:

• Asynchronous background queue: File system modifications are queued and batched during brief idle moments in the user interface. Typing a sentence in the active document never competes with search indexing operations for CPU priority.
• Compact index footprint: Thanks to prefix token compression and SQLite’s optimized B-tree structure, the search index typically consumes less than 5% of the total size of your text archive. A vault containing ten thousand notes produces an index of just a few megabytes.
• Zero persistent daemon overhead: The search index lives entirely within the application runtime. There are no background helper daemons running when the application is closed, ensuring zero phantom battery drain on your MacBook.

Integrating search with meetings and knowledge networks

Fast search transforms the way you interact with meeting documentation. When you capture calls using zero-bot meeting recording and synthesize outcomes with local AI, every transcript snippet, speaker quotation, and action deadline becomes instantly searchable.

You no longer have to remember which specific client sync or executive meeting covered a particular topic. A rapid 3-character query surfaces the exact discussion, complete with clickable wikilinks to the broader project context.

Experience lightning-fast, sovereign information retrieval with Oyma.