Download the PHP package ols/php-fts without Composer

php-fts

A self-contained full-text search engine written in pure PHP.
No extensions. No external services. No dependencies. Just files.

Who is this for?

php-fts is designed for projects where deploying a dedicated search service is not an option — shared hosting, small VPS, or simply situations where you want to keep your stack minimal and portable.

If you have access to Elasticsearch, Meilisearch or Typesense and the infrastructure to run them, use those. They are more powerful and built for high-traffic, large-scale workloads.

If you don't — or if you'd rather not — php-fts gives you solid full-text search with ranked results, filters, and tolerant matching, with nothing to install and nothing to configure beyond a directory path.

It is a good fit if:

• You are on shared hosting (OVH, Infomaniak, o2switch, etc.)
• You want zero infrastructure overhead
• Your dataset is in the range of hundreds to tens of thousands of documents
• You index offline or on a schedule, and serve searches at runtime

It is not a good fit if:

• You need real-time indexing under heavy concurrent write load
• Your dataset is in the millions of documents
• You need geo search or multi-tenant isolation

Language support

php-fts is designed for Western Latin languages (French, English, Spanish, German, Portuguese, Italian, Dutch, and similar).

The tokenizer normalizes input by transliterating accented characters and common diacritics to their ASCII equivalents (é→e, ü→u, ñ→n, etc.), then indexes the result as trigrams.

This approach has limitations you should be aware of:

• Non-Latin scripts — Cyrillic, Arabic, CJK (Chinese, Japanese, Korean), Hebrew, Thai, and others are not supported. Characters outside the handled ranges are dropped during normalization, which means documents in those scripts will not be searchable.
• Some Latin languages — Languages with more complex orthographies, such as Turkish, may not normalize correctly. For example, the Turkish dotless ı or the ğ are not handled.
• Unicode normalization — Real-world input sometimes contains characters in NFD or NFKD form (e.g. diacritics encoded as separate combining characters). php-fts handles this: if the ext-intl extension is available, NFC normalization is applied before transliteration. Otherwise, combining diacritical marks are stripped directly. Either way, fête (NFD) is correctly indexed as fete.

If your content is primarily in Western European languages and ASCII-compatible, php-fts will work well. For multilingual or non-Latin content, a dedicated search engine with proper Unicode support is a better fit.

Features

• Full-text search with trigram indexing — tolerant to typos and partial matches
• BM25 + IDF scoring — industry-standard relevance ranking (same algorithm as Lucene / Elasticsearch)
• Per-document score — exposed in results, usable to build facet counts, sorting, or custom ranking
• Field boosting — weight some fields (e.g. title) more than others
• Filters — exact match, comparisons, range, in, not in, contains on array fields
• Combined AND / OR filtering — flexible condition logic
• Bulk insertion — up to 2.4× faster than individual inserts, single lock for the whole batch
• Soft delete with tombstones — fast deletes, cleaned up on compaction
• Atomic update — soft delete + re-insert in a single lock
• Compaction — rebuilds index files cleanly, removes deleted documents
• Fragmentation monitoring — know when to compact
• Binary file storage — portable across servers, no rebuild needed
• O(1) trigram lookup — fixed-size index (~810 KB), no tree traversal
• No extensions required — runs on any standard PHP 8.1+ installation

Requirements

• PHP 8.1 or higher
• Read/write access to a directory for index files
• ext-intl — optional, but recommended for robust Unicode normalization

Installation

Via Composer

Manual install — if you are not using Composer, copy the src/ directory into your project and include the autoloader:

Quick start

API Reference

Open / Close

Insert

Supported field types: string, int, float, bool, array of strings.

Search

Each result:

The score field is available on every result and can be used to build facet counts, custom sorting, or relevance thresholds.

Highlighting

Pass highlight: true to get a highlights field on each result, containing one snippet per string field. No overhead when disabled (default).

When excerpt is false, the full field value is returned with all matches wrapped in the specified tags.

Filters

Both and and or are optional, but at least one must be present.
When both are used: all AND conditions must pass and at least one OR condition must pass.
A document missing a filtered field is excluded from results.

Update / Delete

Maintenance

Index files

Files are fully portable — copy them between servers without rebuilding.

Scoring

Relevance is computed using BM25 + IDF:

• BM25 — term frequency saturation (a word appearing 10x doesn't score 10x higher) and document length normalization. Parameters: k1 = 1.5, b = 0.75 (standard Lucene defaults).
• IDF — a trigram present in every document contributes little; a rare trigram contributes a lot.
• The final score is normalized between 0 and 100.

Benchmark

Benchmarks were run on two environments:

• Windows 11 — local machine, NVMe SSD, PHP 8.3
• Linux (OVH shared hosting) — standard shared plan, PHP 8.3

Insertion

Insertion is an offline operation — indexing is typically done via a scheduled job, not at request time.
Always prefer insertBulk() in production: it acquires a single lock for the entire batch and is consistently ~2x faster.

Index size

Search — Linux shared hosting, 10 000 documents

200 queries, 10 distinct queries in rotation (including typos and out-of-corpus queries).
Measured with hrtime() on a live shared hosting environment under normal load.

Example application

The gif below shows one possible use of php-fts — a product search interface with filters and ranked results, built on top of a fake shoe catalogue.

It is just an illustration. php-fts is an engine, not an interface. You can use it to power a product search, a documentation search, an admin filter, a CLI tool, or anything else that needs full-text matching over a set of documents.

To run it locally:

No database. No external service. The filters, scores, and result counts are all computed by the engine.

License

MIT — see LICENSE.