README

The module connects your store to several large language model (LLM) providers at once — OpenAI, Anthropic Claude, Google Gemini, xAI Grok, DeepSeek, Gonka, and any OpenAI-compatible API — through a single interface, with automatic failover to a backup provider and cost tracking for every call. Yu_AiLlm doesn't do anything visible on the storefront by itself — it's the infrastructure layer other AI modules are built on: Yu_AiChat (the customer-facing chat assistant) and Yu_AiCatalogSearch (AI-powered catalog search).

Why it exists

Any module that needs an LLM would otherwise have to, on its own:

deal with each vendor's HTTP format separately — OpenAI has one message format, Anthropic another, Gemini a third, each with its own tool-calling format;
write its own failover logic for when the primary provider is unavailable;
store and encrypt API keys;
calculate the cost of every request by hand.

Yu_AiLlm takes care of all of this and gives other modules one simple interface: send a request, get a response — without knowing or caring which specific provider answered.

Features

Multi-provider. Seven providers out of the box: OpenAI, Anthropic Claude, Google Gemini, xAI Grok, DeepSeek, Gonka, plus a universal adapter for any OpenAI-compatible API (e.g. a self-hosted Ollama server). Each provider is enabled and configured independently.
Automatic failover. Providers are tried in the priority order set in the admin panel; if a provider is unavailable (rate limit, invalid key, network failure, timeout), the request automatically falls through to the next one in the list. A provider that just failed is temporarily excluded from the chain, so time isn't wasted on a vendor already known to be down.
Cost tracking. Every LLM call is priced in dollars, based on the token count and the price per 1M tokens set in the admin panel for that provider. Yu_AiChat's analytics consume this data. If no price is set for a provider, the module simply omits the cost instead of erroring out.
Unified temperature scale. Response creativity is set on a common 0–1 scale regardless of each vendor's native range — no need to remember that one provider uses 0–1 and another uses 0–2.
Secure key storage. All API keys are stored encrypted using Magento's own configuration encryption mechanism.

Requirements

PHP >= 8.1
Magento 2.4.x
guzzlehttp/guzzle (installed automatically as a dependency)

Installation

composer require yu-dev/module-ai-llm
bin/magento module:enable Yu_AiLlm
bin/magento setup:upgrade
bin/magento setup:di:compile
bin/magento cache:flush

The module doesn't do anything on the storefront by itself — it only provides the LLM layer for other modules (Yu_AiChat, Yu_AiCatalogSearch), which are installed separately and list it as a dependency.

Configuration

Stores → Configuration → AI → LLM AI

Shared Settings — the default temperature and token limit, used when a consuming module doesn't set its own.
A separate settings group per provider (OpenAI, Anthropic Claude, Google Gemini, xAI Grok, DeepSeek, Gonka, Custom):
- Enabled — turn this specific provider on.
- Priority — the lower the number, the earlier this provider is tried in the failover chain.
- API Key — stored encrypted.
- Model — the specific model for this vendor (e.g. gpt-5-mini, claude-haiku-4-5).
- Price: Input / Output (USD per 1M tokens) — used for the cost estimate; optional.

You can enable just one provider, or several at once — in which case automatic failover kicks in between them by priority.

Author

Yuriy Akishin:

📧 Email: yuriy.akishin@gmail.com
💼 LinkedIn: https://www.linkedin.com/in/yuriyakishin/
💻 GitHub: https://github.com/yuriyakishin

License

MIT

yu-dev / module-ai-llm

Maintainers

Package info

Statistics

Security