v0.6.0: Fuzzy Remix

Added

• Fuzzy Search Engine: mneme find now auto-corrects typos in search queries before executing. If a query term isn't found in the index, the engine substitutes the closest matching term and re-runs the search, printing a "Typo detected — did you mean…" hint to the user.
• internal/query/fuzzy.go: New package providing the core auto-correction pipeline — GetVocabulary, ExpandTokensWithFuzzy, and AutoCorrectQuery — which orchestrate candidate lookup, distance scoring, and query rewriting.
• Trigram Index (internal/utils/trigram.go): Character-trigram index over the full indexed vocabulary for fast O(1) candidate pre-filtering. Uses the Dice coefficient (TrigramSimilarityThreshold = 0.3) to short-circuit Levenshtein evaluation on dissimilar terms.
• Damerau-Levenshtein Distance (internal/utils/levenshtein.go): Replaced the standard Levenshtein implementation with the Optimal String Alignment variant of Damerau-Levenshtein, which counts adjacent transpositions (e.g., "uesr" → "user") as a single edit, improving correction accuracy for common finger-slip typos.
• Dynamic Edit-Distance Thresholding: Fuzzy matching applies a length-aware maximum edit distance — distance 1 for terms of 4–5 characters, distance 2 for terms of 6+ characters — to control false-positive corrections on short words.
• Compound Identifier Correction: Query terms that look like camelCase or snake_case identifiers are split into constituent parts, each part is independently corrected, and then recombined. This prevents over-eager fuzzy correction of technical identifiers.
• Fuzzy Constants (internal/constants/fuzzy.go): Centralised tuning knobs — FuzzyMaxEditDistance, FuzzyMinTermLength, TrigramSimilarityThreshold, and FuzzyScorePenalty — making it easy to adjust matching behaviour without touching algorithm code.
• Expanded Default Ignore List: Added target, build, and dist to the default crawler ignore list in internal/config/load.go to avoid indexing common build-output directories.
• Binary Download Instructions in README.md: Added a "Using a pre-built binary" section for users who do not have Go installed, covering Linux, macOS (Intel & Apple Silicon), and Windows.
• Updated Logo: Replaced the placeholder logo with a high-resolution asset (assets/logo.png).

Changed

• Find Command Fallback: Fixed the fallback path in internal/cli/find.go so that corrected search terms (not the original raw doc.MatchedTerms) are passed to display.FormatSearchResult on the first attempt. The original matched terms are only used as a fallback if the corrected highlight fails.
• Fuzzy Score Penalty: Fuzzy-matched tokens are penalised by a configurable factor (FuzzyScorePenalty = 0.8) so that exact matches always rank above auto-corrected results.
• Tokenizer Hardening: Updated internal/index/tokenizer.go to avoid splitting compound identifiers that are present as a single entry in the index, preventing ghost tokens that pollute fuzzy candidate lists.

Fixed

• allCorrected Never Reset: Fixed a logic bug in AutoCorrectQuery where allCorrected was initialised to true but never set to false when a query part matched exactly or had no suitable correction, making the recombine guard condition always hold. Now allCorrected is correctly set to false in both the exact-match and no-candidate branches.
• Damerau vs. Standard Levenshtein: Replaced calls to the old LevenshteinDistance helper with DamerauLevenshteinDistance and IsWithinDamerauDistance throughout internal/query/fuzzy.go, fixing cases where single-transposition typos were incorrectly counted as two edits and thus rejected.
• Config Ignore List Test: Updated internal/core/config_test.go to expect the expanded nine-entry default ignore list, aligning the test with the updated load.go defaults.

v0.5.0

Added

• Quoted Phrase Search: Use quotes in mneme find to search for exact phrases (e.g., mneme find "error handling" go). Multi-word quoted arguments are treated as phrase matches in snippet highlighting, while individual words are matched separately.
• Query Parsing Module: New internal/query/parse.go with ParseQueryInput for structured tokenization of mixed phrase and keyword queries, with full test coverage.
• Benchmark Suite: New internal/benchmark package with a build-tagged (//go:build benchmark) test suite that measures crawl, index, search, read, and write throughput across corpus sizes of 100–5,000 files. Results are printed as a formatted table. Run via make benchmarks.
• TESTING.md: Added documentation for the benchmark targets.

Changed

• Find Command UX: Improved Long description and Example fields for the find command with clear usage patterns for phrase vs. keyword search.
• Search Result Filtering: Results are now filtered to only include documents with actual text-matching snippets, reducing false positives from BM25 stemming.
• Code Cleanup: Removed unused exported functions, aligned comments, and applied CodeRabbit review suggestions across multiple packages.

Fixed

• Benchmark Isolation: Fixed benchmark tests leaking into the user's real config and data directories by saving and restoring constants.ConfigPath alongside constants.DirPath.
• Test Stability: Fixed failing tests caused by stray quote characters in query argument handling.

v0.3.0

Added

• Windows Support: Full cross-platform support for Windows, including platform-specific path handling and OS detection stored in the VERSION file for compatibility checks.
• Platform Abstraction Layer: New internal/platform package with platform-specific implementations (platform_unix.go, platform_windows.go) for portable path resolution and data directory management.
• CI/CD Pipelines: Added GitHub Actions workflows for automated testing (test.yml) and release builds (release.yml).
• Build System: Introduced a Makefile with standard build targets for streamlined local development and CI builds.
• CLI Enhancements: Added comprehensive descriptions and illustrative examples to all commands and subcommands for a better help experience.

Changed

• Configuration Restructure: Moved configuration constants to internal/constants for clearer package responsibilities.
• Crawler Hardening: The crawler now skips Windows system files and directories (e.g., $Recycle.Bin, System Volume Information) using a case-insensitive lookup.
• Storage Paths: Updated internal/storage/fs.go to use the new platform abstraction layer for resolving data directories, replacing hardcoded Unix paths.
• Documentation: Updated README.md with data store details and improved build instructions.

Fixed

• Windows Temp Dir Fallback: Resolved an issue where the Windows fallback path could cause permission errors by using the system temp directory instead.

v0.2.0: Inital Release

Added

• Pluggable Ingestor System: Introduced Ingestor interface to support multiple file sources, enabling future extensions beyond the local filesystem.
• Batch Indexing: Implemented LSM-tree inspired batch processing for index creation. This significantly reduces memory usage during indexing by writing numbered chunks and maintaining a manifest.json.
• BM25 & VSM Ranking: Integrated BM25 and Vector Space Model algorithms to improve search result relevance and ranking.
• Binary Segment Storage: Transititioned segment storage to Protocol Buffers for improved performance and reduced disk footprint.
• Tombstones & Cleanup: Added mneme clean command and a tombstone mechanism to safely manage and remove obsolete index files.
• Multilingual Tokenization: Enhanced tokenization logic to better support multi-lingual text analysis.
• Configurable Logging: Added log level configuration (defaulting to info) to reduce noise.
• Binary File Skipping: The crawler now automatically identifies and skips binary files (e.g., images, videos) to prevent indexing garbage data.
• Exclude Extensions: new configuration option exclude_extensions to explicitly ignore specific file types.

Changed

• Core Refactor: Moved primary data structures to internal/core to improve modularity and reduce circular dependencies.
• Progress Bar: Unified the progress bar implementation to be "sticky" at the bottom of the terminal, providing a cleaner UX.
• Search Highlighting: Improved snippet generation to accurately highlight partial matches in search results.
• Configuration: Cleanup of unused ranking weights in the configuration file.

Fixed

• Module Dependencies: Resolved indirect dependency issues in go.mod.
• Test Stability: Fixed flaky tests in config and logger packages related to environment permissions and default values.

What's Changed

• Ingestor implementation by @Ayanrocks in #1

New Contributors

• @Ayanrocks made their first contribution in #1

Full Changelog: https://github.com/Ayanrocks/mneme/commits/v0.2.0