PHPackages                             kabdullah27/php-token-squeezer - PHPackages - PHPackages  [Skip to content](#main-content)[PHPackages](/)[Directory](/)[Categories](/categories)[Trending](/trending)[Leaderboard](/leaderboard)[Changelog](/changelog)[Analyze](/analyze)[Collections](/collections)[Log in](/login)[Sign up](/register)

1. [Directory](/)
2. /
3. [Utility & Helpers](/categories/utility)
4. /
5. kabdullah27/php-token-squeezer

ActiveLibrary[Utility & Helpers](/categories/utility)

kabdullah27/php-token-squeezer
==============================

General-purpose AI token optimization library for PHP — compress context, reduce costs, boost speed.

v1.0.1(1w ago)03↓100%MITPHPPHP ^8.2

Since May 28Pushed 6d agoCompare

[ Source](https://github.com/kabdullah27/php-token-squeezer)[ Packagist](https://packagist.org/packages/kabdullah27/php-token-squeezer)[ RSS](/packages/kabdullah27-php-token-squeezer/feed)WikiDiscussions main Synced 1w ago

READMEChangelog (1)Dependencies (5)Versions (3)Used By (0)

TokenSqueezer 🗜️
================

[](#tokensqueezer-️)

**General-purpose AI token optimization library for PHP & Laravel.**

Compress context, cut token usage by up to 80%, and call any AI provider through one fluent API for *any* domain.

```
composer require yourusername/token-squeezer

```

---

Features
--------

[](#features)

- **Fluent chain API** — readable, intuitive, testable
- **Multi-provider** — OpenAI, Claude, Gemini, Kimi, Mimo, Ollama, or your own driver
- **Smart compression** — 4 built-in modes (including RTK log compression) + plugin support
- **Caveman Mode** — force models to output highly compressed replies to save output tokens
- **Auto-caching** — context-hash keys, pluggable drivers (array / file / redis / laravel)
- **Schema enforcement** — define expected JSON keys, auto-fill missing
- **Token monitoring** — usage tracking, cost estimation, latency stats
- **Zero Laravel dependency** — works in plain PHP too
- **Dry-run / inspect** — preview prompt + compression without calling AI
- **Fallback Chain** — auto-retry across multiple providers on failure
- **Rate Limiter** — per-provider sliding window, prevent 429 errors
- **Batch Processing** — analyze multiple items with partial-failure safety

---

Quick Start
-----------

[](#quick-start)

### 1. Install

[](#1-install)

```
composer require kabdullah27/php-token-squeezer
```

### 2. Laravel Setup (auto-discovery works automatically)

[](#2-laravel-setup-auto-discovery-works-automatically)

```
php artisan vendor:publish --tag=token-squeezer-config
```

Add to your `.env`:

```
TSQ_PROVIDER=openai
OPENAI_API_KEY=sk-...
TSQ_CACHE_DRIVER=laravel
```

### 3. Plain PHP Setup

[](#3-plain-php-setup)

```
use TokenSqueezer\TokenSqueezer;
use TokenSqueezer\CompressMode;

TokenSqueezer::configure([
    'default_provider' => 'openai',
    'providers' => [
        'openai' => [
            'api_key' => getenv('OPENAI_API_KEY'),
            'model'   => 'gpt-4o-mini',
        ],
    ],
    'cache'   => ['driver' => 'file'],
    'monitor' => true,
]);
```

---

Usage Examples
--------------

[](#usage-examples)

### Trading / Financial Analysis

[](#trading--financial-analysis)

```
$result = TokenSqueezer::analyze()
    ->context([
        'symbol'       => 'BTCUSDT',
        'price'        => 104500,
        'rsi'          => 74,
        'macd'         => 'cross_up',
        'trend'        => 'bullish',
        'volume_spike' => true,
    ])
    ->compress(CompressMode::AGGRESSIVE)
    ->schema(['trend', 'risk', 'action'])
    ->temperature(0.1)
    ->maxTokens(80)
    ->cache(ttl: 60)
    ->via('openai')
    ->run();

// ['trend' => 'bullish', 'risk' => 'medium', 'action' => 'wait breakout confirmation']
```

---

### E-commerce Product Scoring

[](#e-commerce-product-scoring)

```
$result = TokenSqueezer::analyze()
    ->context([
        'product'   => 'Nike Air Max 90',
        'price'     => 1200000,
        'stock'     => 3,
        'rating'    => 4.7,
        'reviews'   => 842,
        'category'  => 'sneakers',
        'discount'  => 15,
    ])
    ->compress(CompressMode::BALANCED)
    ->schema(['score', 'label', 'recommendation'])
    ->temperature(0.2)
    ->maxTokens(100)
    ->cache(ttl: 600)
    ->via('gemini')
    ->run();
```

---

### Customer Support Ticket Triage

[](#customer-support-ticket-triage)

```
$result = TokenSqueezer::analyze()
    ->context([
        'subject'  => 'Cannot login after password reset',
        'tier'     => 'premium',
        'age_days' => 2,
        'tags'     => 'auth, reset, blocked',
    ])
    ->compress(CompressMode::BALANCED)
    ->prompt('Triage this support ticket: {{context}}. Assign priority and team.')
    ->schema(['priority', 'team', 'estimated_resolution_hours'])
    ->temperature(0.1)
    ->maxTokens(80)
    ->cache(ttl: 120)
    ->run();
```

---

### Content Moderation

[](#content-moderation)

```
$result = TokenSqueezer::analyze()
    ->context([
        'text'        => substr($userComment, 0, 200),
        'user_age'    => 17,
        'platform'    => 'forum',
        'lang'        => 'id',
    ])
    ->compress(CompressMode::MINIMAL)
    ->schema(['safe', 'category', 'action'])
    ->temperature(0.0)
    ->maxTokens(60)
    ->via('claude')
    ->run();
```

---

### SEO / Content Scoring

[](#seo--content-scoring)

```
$result = TokenSqueezer::analyze()
    ->context([
        'title_len'    => 62,
        'meta_len'     => 148,
        'h1_count'     => 1,
        'word_count'   => 1450,
        'keyword_density' => 1.8,
        'readability'  => 'grade_8',
        'images_alt'   => false,
    ])
    ->compress(CompressMode::AGGRESSIVE)
    ->schema(['seo_score', 'issues', 'priority_fix'])
    ->temperature(0.1)
    ->maxTokens(120)
    ->cache(ttl: 3600)
    ->run();
```

---

### HR / Resume Screening

[](#hr--resume-screening)

```
$result = TokenSqueezer::analyze()
    ->context([
        'role'          => 'Backend Engineer',
        'years_exp'     => 4,
        'skills'        => 'PHP, Laravel, MySQL, Redis, Docker',
        'education'     => 'S1 Informatika',
        'english_level' => 'intermediate',
        'location'      => 'Jakarta',
        'salary_expect' => 15000000,
    ])
    ->compress(CompressMode::BALANCED)
    ->schema(['fit_score', 'strengths', 'gaps', 'proceed'])
    ->temperature(0.2)
    ->maxTokens(150)
    ->via('openai', 'gpt-4o-mini')
    ->run();
```

---

### System logs cleanup (RTK & Caveman Mode)

[](#system-logs-cleanup-rtk--caveman-mode)

```
$result = TokenSqueezer::analyze()
    ->context([
        'system_logs' => "error: connection failed\nerror: connection failed\nerror: connection failed",
    ])
    ->compress(CompressMode::RTK) // 💡 Deduplicates repeating logs & comment lines (RTK logic)
    ->caveman(true)               // 💡 Instructs AI to reply in extremely brief "caveman" format
    ->via('mimo')                 // 💡 Uses Xiaomi Mimo provider
    ->run();
```

---

Fallback Chain
--------------

[](#fallback-chain)

Automatically try the next provider if the primary one fails (timeout, 429, API error).

```
TokenSqueezer::analyze()
    ->context(['symbol' => 'BTC', 'rsi' => 74])
    ->compress(CompressMode::AGGRESSIVE)
    ->schema(['trend', 'risk'])
    ->via('openai')
    ->fallback('claude', 'gemini')  // tried in order if openai fails
    ->run();
```

- Only falls back on `TokenSqueezedException` (network/API errors)
- Rate-limited providers in the chain are **skipped** automatically
- Throws `FallbackExhaustedException` if all providers fail (contains per-provider error details)

```
use TokenSqueezer\Exceptions\FallbackExhaustedException;

try {
    $result = TokenSqueezer::analyze()
        ->context($data)
        ->via('openai')
        ->fallback('claude', 'gemini')
        ->run();
} catch (FallbackExhaustedException $e) {
    // $e->errors = ['openai' => 'connection failed', 'claude' => '...']
    logger()->error('All AI providers failed', $e->errors);
}
```

---

Rate Limiter
------------

[](#rate-limiter)

Prevent hitting provider rate limits. Sliding window algorithm, in-memory per process.

**Configure via `.env`:**

```
TSQ_OPENAI_RATE_LIMIT=60    # max 60 requests
TSQ_OPENAI_RATE_WINDOW=60   # per 60 seconds
TSQ_CLAUDE_RATE_LIMIT=40
TSQ_CLAUDE_RATE_WINDOW=60
```

**Or in config:**

```
'providers' => [
    'openai' => [
        'api_key'     => env('OPENAI_API_KEY'),
        'rate_limit'  => 60,  // 0 = no limit (default)
        'rate_window' => 60,
    ],
],
```

When the limit is hit, the provider is skipped in the fallback chain, or `RateLimitException` is thrown (with `retryAfterSeconds` property) if no fallback is available.

---

Batch Processing
----------------

[](#batch-processing)

Analyze multiple context arrays with one call. Each item runs sequentially with full support for compression, caching, fallback, and rate limiting.

```
$results = TokenSqueezer::batch([
    ['symbol' => 'BTC', 'rsi' => 74, 'trend' => 'bullish'],
    ['symbol' => 'ETH', 'rsi' => 55, 'trend' => 'neutral'],
    ['symbol' => 'SOL', 'rsi' => 82, 'trend' => 'overbought'],
])
->compress(CompressMode::AGGRESSIVE)
->schema(['action', 'risk'])
->temperature(0.1)
->maxTokens(80)
->via('openai')
->fallback('claude')      // fallback applies per item
->cache(ttl: 300)
->run();

// $results:
// [
//   ['index' => 0, 'result' => ['action' => 'hold', 'risk' => 'medium'], 'error' => null],
//   ['index' => 1, 'result' => ['action' => 'buy',  'risk' => 'low'],    'error' => null],
//   ['index' => 2, 'result' => null, 'error' => 'Rate limit exceeded...'],
// ]
```

**Stop on first error** (default continues on error):

```
TokenSqueezer::batch($items)
    ->schema(['score'])
    ->stopOnError()   // abort remaining items if any fails
    ->run();
```

**Filter results by success/failure:**

```
$results = TokenSqueezer::batch($items)->schema(['score'])->run();

$successes = array_filter($results, fn($r) => $r['error'] === null);
$failures  = array_filter($results, fn($r) => $r['error'] !== null);
```

---

Advanced Usage
--------------

[](#advanced-usage)

### Custom System Prompt

[](#custom-system-prompt)

```
TokenSqueezer::analyze()
    ->context(['risk_score' => 87, 'country' => 'ID', 'amount' => 500_000_000])
    ->system('You are a fraud detection engine. Be conservative. Always return JSON.')
    ->schema(['fraud_likelihood', 'flags', 'block'])
    ->temperature(0.0)
    ->maxTokens(80)
    ->run();
```

### Custom Prompt with {{context}} Injection

[](#custom-prompt-with-context-injection)

```
TokenSqueezer::analyze()
    ->context(['order_id' => 'ORD-9182', 'status' => 'pending', 'delay_days' => 4])
    ->prompt('Summarize this order situation: {{context}}. What should we tell the customer?')
    ->asText()      // plain text instead of JSON
    ->temperature(0.3)
    ->maxTokens(80)
    ->run();
```

### Inspect Without Calling AI (debug)

[](#inspect-without-calling-ai-debug)

```
$info = TokenSqueezer::analyze()
    ->context(['symbol' => 'ETH', 'rsi' => 68, 'trend' => 'bullish'])
    ->compress(CompressMode::AGGRESSIVE)
    ->schema(['trend', 'risk'])
    ->inspect();

// $info['estimated_reduction'] => "72% (from 124 to 35 chars)"
// $info['compressed_context']  => "SMBL=ETH|RSI=68|TRND=BULL"
// $info['prompt']              => [...]
```

### Custom Compressor Plugin

[](#custom-compressor-plugin)

```
use TokenSqueezer\Contracts\CompressorInterface;

class MyDomainCompressor implements CompressorInterface
{
    public function compress(array $context): string
    {
        // your custom logic
        return implode(' ', array_map(
            fn($k, $v) => strtoupper($k[0]) . ":{$v}",
            array_keys($context),
            $context
        ));
    }
}

TokenSqueezer::analyze()
    ->context($data)
    ->compress(CompressMode::CUSTOM)
    ->addCompressor(new MyDomainCompressor())
    ->schema(['result'])
    ->run();
```

### Custom Provider Driver

[](#custom-provider-driver)

```
use TokenSqueezer\Providers\ProviderFactory;

ProviderFactory::extend('deepseek', MyDeepSeekProvider::class);

TokenSqueezer::analyze()
    ->context($data)
    ->via('deepseek', 'deepseek-chat')
    ->run();
```

### Token Usage Monitoring

[](#token-usage-monitoring)

```
// After multiple requests:
$usage = TokenSqueezer::usage();

// $usage:
// [
//   'total_requests'      => 42,
//   'cache_hits'          => 31,
//   'cache_hit_rate'      => '73.8%',
//   'total_input_tokens'  => 1840,
//   'total_output_tokens' => 430,
//   'avg_latency_ms'      => 487,
//   'estimated_cost_usd'  => '$0.0004',
//   'by_provider'         => ['openai' => [...], 'claude' => [...]],
// ]

TokenSqueezer::resetUsage();
```

---

Compression Modes
-----------------

[](#compression-modes)

ModeWhat it doesReduction`MINIMAL`Normalize whitespace, stringify booleans~20%`BALANCED`+ strip stopwords, abbreviate common phrases~50%`AGGRESSIVE`+ encode to shortcodes, max density, drop all vowels in keys~75-80%`RTK`+ deduplicate repeating log lines, strip whitespace & comments~80-90%`CUSTOM`Only your plugins run — full controlYou decide---

Cache TTL Guide
---------------

[](#cache-ttl-guide)

Use CaseSuggested TTLReal-time data30–60 secPer-request scores2–5 minProduct catalog10–30 minStatic analysis1–24 hours---

Supported Providers
-------------------

[](#supported-providers)

ProviderDefault ModelNotes`openai``gpt-4o-mini`Cheapest, fastest`claude``claude-haiku-4-5-*`Great for structured JSON`gemini``gemini-1.5-flash`Free tier available`kimi``moonshot-v1-8k`OpenAI-compatible`mimo``mimo-v2.5`Low cost, OpenAI-compatible`ollama``llama3`Local, zero API costcustomYour driverImplement ProviderInterface---

Environment Variables
---------------------

[](#environment-variables)

```
TSQ_PROVIDER=openai
TSQ_CACHE_ENABLED=true
TSQ_CACHE_DRIVER=laravel    # array | file | redis | laravel
TSQ_CACHE_TTL=300
TSQ_CACHE_PREFIX=tsq:
TSQ_MONITOR=true

OPENAI_API_KEY=sk-...
OPENAI_MODEL=gpt-4o-mini

ANTHROPIC_API_KEY=...
CLAUDE_MODEL=claude-haiku-4-5-20251001

GEMINI_API_KEY=...
KIMI_API_KEY=...

MIMO_API_KEY=...
MIMO_MODEL=mimo-v2.5
MIMO_BASE_URL=https://api.xiaomimimo.com/v1/chat/completions

OLLAMA_URL=http://localhost:11434/api/chat
OLLAMA_MODEL=llama3
```

---

License
-------

[](#license)

MIT

###  Health Score

41

—

FairBetter than 87% of packages

Maintenance99

Actively maintained with recent releases

Popularity4

Limited adoption so far

Community6

Small or concentrated contributor base

Maturity47

Maturing project, gaining track record

 Bus Factor1

Top contributor holds 100% of commits — single point of failure

How is this calculated?**Maintenance (25%)** — Last commit recency, latest release date, and issue-to-star ratio. Uses a 2-year decay window.

**Popularity (30%)** — Total and monthly downloads, GitHub stars, and forks. Logarithmic scaling prevents top-heavy scores.

**Community (15%)** — Contributors, dependents, forks, watchers, and maintainers. Measures real ecosystem engagement.

**Maturity (30%)** — Project age, version count, PHP version support, and release stability.

###  Release Activity

Cadence

Every ~2 days

Total

2

Last Release

10d ago

### Community

Maintainers

![](https://www.gravatar.com/avatar/ba169e62ce7288e75eb8f03b72c49d0c9c3a14f0803829a9deb9ff1b83b682db?d=identicon)[kabdullah27](/maintainers/kabdullah27)

---

Top Contributors

[![kabdullah27](https://avatars.githubusercontent.com/u/33799089?v=4)](https://github.com/kabdullah27 "kabdullah27 (8 commits)")

---

Tags

laravelaitokenopenaioptimizationGeminiclaudellmkimimimo

###  Code Quality

TestsPHPUnit

### Embed Badge

![Health badge](/badges/kabdullah27-php-token-squeezer/health.svg)

```
[![Health](https://phpackages.com/badges/kabdullah27-php-token-squeezer/health.svg)](https://phpackages.com/packages/kabdullah27-php-token-squeezer)
```

###  Alternatives

[cognesy/instructor-php

The complete AI toolkit for PHP: unified LLM API, structured outputs, agents, and coding agent control

317117.1k1](/packages/cognesy-instructor-php)[helgesverre/toon

Token-Oriented Object Notation - A compact data format for reducing token consumption when sending structured data to LLMs

125103.5k19](/packages/helgesverre-toon)[rumenx/php-seo

AI-powered, framework-agnostic PHP package for automated SEO optimization. Intelligently generates meta tags, titles, descriptions, and alt texts using configurable AI providers or manual patterns.

113.7k](/packages/rumenx-php-seo)

PHPackages © 2026

[Directory](/)[Categories](/categories)[Trending](/trending)[Changelog](/changelog)[Analyze](/analyze)