Neuron Ai
The PHP Agentic Framework to build production-ready AI driven applications. Connect components (LLMs, vector DBs, memory) to agents that can interact with your data.
Last verified:
What is Neuron Ai?
Neuron AI is a production-ready PHP framework for building and orchestrating AI agents, complex workflows, and RAG (Retrieval-Augmented Generation) systems directly within your existing PHP stack. It is the first agentic framework in the PHP ecosystem, designed for PHP developers who want to create AI-driven applications using the language and tools they already know.
Key features include support for AI Agents with built-in memory and tool calling, multi-agent workflows with state management and human-in-the-loop logic, complete RAG pipelines connecting PDFs/SQL/docs to vector stores like Pinecone or Elasticsearch, and built-in monitoring via Inspector.dev for full observability into agent thought processes and tool calls. The framework supports 11+ LLM providers including OpenAI, Anthropic, Gemini, Mistral, Ollama, and AWS Bedrock, allowing you to swap providers with a single line of code.
Neuron AI is designed for PHP developers, particularly those working with Laravel or Symfony frameworks, who need to build production-grade AI applications. It is ideal for teams creating customer service bots, data analyst agents, CLI agents, YouTube video summarizers, SQL query optimizers, travel planners, and deep research systems. The framework provides 100% PHPStan type coverage, fully typed methods/properties/return values for maximum IDE support, and is coding agent-friendly with integrated MCP server support.
Neuron Ai pricing
Pricing model: Freemium
Neuron AI is open-source and free to use under the MIT license. The framework itself has no cost - you install it via Composer with 'composer require neuron-core/neuron-ai'. Pricing is based on your needs since it's open source. However, you will pay for external services you integrate: LLM provider API costs (OpenAI, Anthropic, Gemini, etc.), vector store services (Pinecone, Elasticsearch hosting), embedding provider API costs (Voyage, OpenAI embeddings), and Inspector.dev monitoring if you use their paid tiers. There is no paid plan for Neuron AI itself - it remains free and open-source.
Neuron Ai pros
- 100% PHPStan type coverage across entire framework
- Fully typed methods, properties, and return values for maximum IDE support
- Support for 11+ LLM providers (OpenAI, Anthropic, Gemini, Mistral, Ollama, AWS Bedrock, Azure, Deepseek, Grok, HuggingFace, OpenAILike)
- Swap LLM providers with a single line of code without impacting implementation
- Built-in memory and chat history management automatically
- Multi-agent workflow support with state management and orchestration
- Native human-in-the-loop pattern for validation and correction
- Complete RAG pipeline with vector stores (Pinecone, Elasticsearch) and embeddings
- Built-in observability via Inspector.dev integration with one-line setup
- Modular agent toolkits (MySQLToolkit, CalculatorToolkit) for specialized capabilities
- MCP Connector for connecting to external MCP servers
- Structured output support for business process automation
- Seamless Laravel integration with Artisan commands (php artisan neuron:agent)
- Symfony service container compatibility with interface-based design
- No vendor lock-in due to modular interface-based architecture
- Data integration from CSV, JSON, SQL databases for unified agent models
- Agent and RAG can be used as standalone components in Workflows
- 1.2k GitHub stars with 291 releases and active maintenance
- MIT license for open-source freedom
- Video tutorials available for step-by-step learning from installation to advanced features
Neuron Ai cons
- Requires PHP 8.1 or higher, limiting older PHP projects
- Only supports PHP ecosystem (no Python, JavaScript, or other language support)
- Monitoring requires separate Inspector.dev account and API key setup
- LLM provider API keys must be managed separately (not included)
- Vector store services like Pinecone require separate paid subscriptions
- Embedding providers like Voyage require separate API keys and costs
- Learning curve for developers new to AI agents and agentic patterns
- Non-deterministic AI behavior makes debugging more challenging without monitoring
Frequently asked questions about Neuron Ai
What is Neuron AI?
Neuron AI is a PHP framework for creating and orchestrating AI Agents. It allows you to integrate AI entities into your existing PHP applications with a powerful and flexible architecture. It provides tools for the entire agentic application development lifecycle, from LLM interfaces to data loading, to multi-agent orchestration, to monitoring and debugging. It is the first agentic framework in the PHP world with a Workflow-First architecture.
How do I install Neuron AI?
Install the latest version by running the Composer command: 'composer require neuron-core/neuron-ai'. Neuron requires PHP ^8.1. After installation, you can create your first agent using the command 'php vendor/bin/neuron make:agent App\Neuron\MyAgent'.
Which LLM providers does Neuron support?
Neuron supports 11+ LLM providers: Anthropic, OpenAI (also as embeddings provider), OpenAI on Azure, Ollama (also as embeddings provider), OpenAILike, Gemini (also as embeddings provider), Mistral, HuggingFace, Deepseek, Grok, and AWS Bedrock Runtime. You can switch between providers with just one line of code without impacting your agent implementation.
How do I add tools to my agent?
Make your agent able to perform concrete tasks by adding tools or toolkits (collections of tools) in the tools() method. For example, you can add MySQLToolkit::make(\DB::connection()->getPdo()) to give your agent database reading capabilities. You can also use MCP Connector to connect to tools exposed by an MCP server instead of implementing tools manually.
How does memory work in Neuron agents?
Neuron automatically manages conversation memory, making agents context-aware. When you send multiple messages to an agent, it remembers what was said in previous interactions. For example, if you tell the agent your name and then ask 'Do you remember my name?', it will recall your name from the conversation history. You can learn more about memory in the official documentation.
How do I monitor my agents in production?
Neuron integrates with Inspector.dev for monitoring. After signing up at inspector.dev, set the INSPECTOR_INGESTION_KEY variable in your application environment file (e.g., 'INSPECTOR_INGESTION_KEY=nwse877auxxxxxxxxxxxxxxxxxxxxxxxxxxxx'). After configuring this, you will see the agent execution timeline in your Inspector dashboard with details of internal steps, tool calls, and RAG retrievals.
What is Neuron Workflow?
Workflow is a smart flowchart for AI applications that allows you to use all Neuron components (AI providers, embeddings, data loaders, chat history, vector store, etc.) as standalone components to create customized agentic entities. Unlike Agent and RAG classes which are ready-to-use implementations, Workflow lets you program your agentic system completely from scratch. It supports human-in-the-loop patterns, multi-agent streaming, and collaboration.
How do I implement RAG with Neuron?
To create a RAG, you need to attach additional components: a vector store and an embeddings provider. Create a RAG with 'php vendor/bin/neuron make:rag MyChatBot'. Your RAG class must implement provider() for the LLM, embeddings() for the embedding provider (e.g., VoyageEmbeddingProvider), and vectorStore() for the vector store (e.g., PineconeVectorStore). This connects PDFs, SQL, or docs to vector stores for agent context.
Does Neuron work with Laravel?
Yes, Neuron has excellent Laravel integration. It provides a neuron-laravel bridge with Artisan Commands like 'php artisan neuron:agent SalesAgent' to scaffold new agent classes. You can store chat histories directly in your database using Eloquent models and use the AIProvider facade to switch between LLM providers with a single line in your .env file. The framework feels like a natural extension of Laravel tools you already use.
What is structured output in Neuron?
Structured output allows Agents to understand natural language but output in a structured format like PHP classes, useful for business process automation and data extraction. Define the output structure as a PHP class with SchemaProperty attributes, then use agent->structured() with your message and class. The agent returns an instance of your class with parsed values, which you can use with downstream systems.