README

A tool that intercepts PHP errors, warnings, and exceptions and provides practical help and advice generated by AI (local models or external APIs). No static text: details and suggestions are created on-the-fly by artificial intelligence based on the error message and context.

Screenshots:

Supports local AI backends (e.g. Ollama/LocalAI) and APIs (e.g. OpenAI, Anthropic, Google Gemini).
HTML, text or JSON output.
Simple configuration via environment variables or by instantiating the Config.

Requirements

PHP >= 7.4
Composer
(Optional) a local AI backend (Ollama/LocalAI) or an API key (OpenAI, etc.)

Installation

composer require raffaelecarelle/php-error-insight

Configuration

You can configure the tool via environment variables or through code.

Supported environment variables:

PHP_ERROR_INSIGHT_ENABLED: true/false (default: true)
PHP_ERROR_INSIGHT_BACKEND: none|local|api|openai|anthropic|google|gemini
PHP_ERROR_INSIGHT_MODEL: model name (e.g. llama3:instruct, gpt-4o-mini, claude-3-5-sonnet-20240620, gemini-1.5-flash)
PHP_ERROR_INSIGHT_API_KEY: API key (required for api/openai/anthropic/google backends)
PHP_ERROR_INSIGHT_API_URL: service URL (optional override; e.g. http://localhost:11434 for Ollama, https://api.openai.com/v1/chat/completions for OpenAI, https://api.anthropic.com/v1/messages for Anthropic, https://generativelanguage.googleapis.com/v1/models for Google Gemini)
PHP_ERROR_INSIGHT_LANG: language for AI prompt (it, en, ...; default: en)
PHP_ERROR_INSIGHT_OUTPUT: auto|html|text|json (default: auto)
PHP_ERROR_INSIGHT_VERBOSE: true/false (default: false)
PHP_ERROR_INSIGHT_TEMPLATE: path to a custom HTML template (optional)

Configuration examples:

Local backend (Ollama):

export PHP_ERROR_INSIGHT_BACKEND=local export PHP_ERROR_INSIGHT_MODEL=llama3:instruct export PHP_ERROR_INSIGHT_API_URL=[http://localhost:11434](http://localhost:11434)

API backend (OpenAI compatible):

export PHP_ERROR_INSIGHT_BACKEND=api
export PHP_ERROR_INSIGHT_MODEL=gpt-4o-mini
export PHP_ERROR_INSIGHT_API_KEY=sk-...
export PHP_ERROR_INSIGHT_API_URL=https://api.openai.com/v1/chat/completions

API backend (Anthropic Claude):

export PHP_ERROR_INSIGHT_BACKEND=anthropic
export PHP_ERROR_INSIGHT_MODEL=claude-3-5-sonnet-20240620
export PHP_ERROR_INSIGHT_API_KEY=api-key
# optional override
# export PHP_ERROR_INSIGHT_API_URL=https://api.anthropic.com/v1/messages

API backend (Google Gemini):

export PHP_ERROR_INSIGHT_BACKEND=google
export PHP_ERROR_INSIGHT_MODEL=gemini-1.5-flash
export PHP_ERROR_INSIGHT_API_KEY=api-key
# optional override
# export PHP_ERROR_INSIGHT_API_URL=https://generativelanguage.googleapis.com/v1/models

Usage (Vanilla PHP)

In your application's bootstrap, register the handler included in the examples, or use the helper provided by the package. A minimal example is available in examples/vanilla/index.php.

Quick example:

use ErrorExplainer\Config; use ErrorExplainer\Register;
require **DIR**.'/vendor/autoload.php';
$config = Config::fromEnvAndArray([ 'backend' => 'local', // none | local | api 'model' => 'llama3:instruct', 'language'=> 'en', 'verbose' => true, ]);
Register::install($config); // sets up error and exception handlers
// Generate an error to see the output strpos();

Output:

In HTML you'll see the page with stack trace and the "Details/Suggestions" section populated by AI.
In CLI you'll get text/JSON depending on configuration.

How it works

The tool intercepts errors/warnings/exceptions.
Builds a prompt with message, severity and location.
Sends the prompt to the configured AI backend.
Shows the AI response as details and practical suggestions.

Note: the tool no longer uses static translated texts for details/suggestions. If the AI backend is not configured or doesn't respond, those sections might remain empty.

Privacy and Data Sanitization

By default, prompts are sanitized before being sent to any AI backend to reduce the risk of leaking sensitive information (emails, tokens, private IPs, cookies, payment-like numbers, etc.). You can control this behavior via environment variables:

PHP_ERROR_INSIGHT_SANITIZE: 1|0 (default: 1 when unset)
PHP_ERROR_INSIGHT_SANITIZE_RULES: comma-separated list of rules to enable (secrets, pii, payment, network). Example: secrets,pii,network
PHP_ERROR_INSIGHT_SANITIZE_MASK: override the default mask string (default: REDACTED)

Technical details:

Sanitization happens inside Internal/Explainer just after the prompt is built and before any backend/API call.
The default sanitizer masks Authorization headers, JWT-like tokens, emails, phone numbers, Italian CF/IBAN, payment-like card numbers, private IPs, and Cookie headers.
You can inject your own AI client (AIClientInterface) if you prefer to handle sanitization externally.

For more details, see docs/sanitizzazione-dati-ai.md.

License

GPL-3.0-or-later

This project is licensed under the GNU General Public License v3.0 or later. See the LICENSE file for details.

raffaelecarelle / php-error-insight

Maintainers

Details