I Designed a 3 Layer Magento 2 AI Framework Open Source

I designed a three level framework for Magento 2 to allow any developers to implement AI based functionality. It allows you to add any AI Providers like ChatGPT, Gemini etc… and allows you to add Functionality without knowing what AI provider will be used under the hood.

I will open-source is soon but already happy to discuss about it. The key pont is this is a FRAMEWORK. You can write your own module if you are a developer or digital agency, top on the framework. You can have own AI Providers, Functions and so on. It’s yours.

AI Framwork fundamentals

The framework is as a three-level architecture:

1. AiCore Stable contracts, provider pool, executor, stream normalization
2. Provider modules OpenAI, Ollama, Anthropic, etc.
3. Feature modules Chat, SEO, recommendations, content improvement, each implemented as task/use-case classes

It does not use any 3rd party extension or solution which is important to keep it independent.

The execution layer in the AiCore

Inside AiCore, createed an internal orchestration service for the flow:

• AiExecutorInterface
• ProviderResolverInterface
• FailoverExecutorInterface
• TaskRunnerInterface

Flow:

1. Feature task builds a generic request
2. Resolver selects provider/model for store + feature
3. Executor calls sync or stream mode
4. Response is normalized
5. Task maps it back into domain output

The architecture with a bit more technical

1. Core framework module

Name: ClusterifyAI_AiCore

This module contains only the stable fundamentals:

• provider contracts
• request/response DTO contracts
• execution/orchestration services
• failover logic
• transport helpers
• streaming normalization
• config access
• logging / telemetry
• test doubles

If you are familiar with Laravel, then this is kind of the SDK.

2. Provider modules

Separate provider integrations into their own Magento modules:

• ClusterifyAI_AiProviderOpenAI
• ClusterifyAI_AiProviderAnthropic
• ClusterifyAI_AiProviderOllama

later: Gemini, xAI, Groq, etc or any provider even local ones, it’s up to you.

Custom provider approach: the framework knows how to discover and use providers, but each provider is installed separately and implements only the capabilities it supports. Also trying to avoid “monolit” design and keep it updateable etc….

3. Feature modules

Keep business use cases out of the core:

• ClusterifyAI_AiFeatureChat
• ClusterifyAI_AiFeatureSeo
• ClusterifyAI_AiFeatureRecommendations
• ClusterifyAI_AiFeatureContent

These modules should define task/use-case classes that encapsulate prompts, context gathering, schema expectations, and response mapping.

Failover and provider chains

The AiCore module now treats failover as a native concept, including arrays of providers/models.

In Magento, each feature can be configurable:

• primary provider/model
• fallback chain
• timeout
• max retries
• streaming enabled yes/no
• store scope

For example:

• Chat: local Ollama first, remote provider fallback
• SEO: cheaper model first
• Recommendations: structured-output-capable model first

Storage and conversations

Because chat is one of my goals, later I can create these database tables to ensure the conversations are safed and can be distributed as information.

• clusterify_ai_conversation
• clusterify_ai_message
• clusterify_ai_request_log

Testing strategy

Fake providers matter. For testing purposes we have fake providers.

• FakeTextProvider
• FakeStreamTextProvider
• fixture-based provider responses
• contract tests every provider module must pass

Also this way third-party providers can prove compatibility with the framework.

Update is coming in a few days and code will be released for review and discussions very soon.

If you have quiestion or need a hand then please message me.

Happy coding,