neuron-core/neuron-ai

What it solves

Neuron is a PHP framework designed to simplify the creation and orchestration of AI agents. It provides the necessary infrastructure for building "agentic" applications—software where the AI agent drives the reasoning, action, and interface—rather than simply adding AI features to a traditional web app.

How it works

Neuron provides a set of core classes and components that developers can extend to build specialized agents. It abstracts the complexity of LLM providers, memory management, and tool integration. Key architectural features include:

• Agent & RAG Classes: Ready-to-use implementations for common patterns like tool calling, structured output, and Retrieval Augmented Generation.
• Workflow: A flexible system for creating custom agentic flows, supporting event-driven workflows, asynchronous execution, and human-in-the-loop interruptions for validation or correction.
• MCP Connector: Integration with the Model Context Protocol (MCP) to easily connect agents to external tools without manual implementation.
• Provider Abstraction: A unified interface to switch between various LLM providers (e.g., Anthropic, OpenAI, Gemini, Mistral) with minimal code changes.

Who it’s for

PHP developers and software houses who want to build production-ready AI agents and standardize their architecture on a dedicated agentic framework.

Highlights

• Multi-Provider Support: Compatible with a wide array of LLM and embedding providers.
• Human-in-the-Loop: Built-in support for interrupting automated processes for human approval or context.
• Structured Output: Ability to force agents to output data in a specific PHP class format for downstream automation.
• Observability: Integration with Inspector for monitoring and debugging non-deterministic AI agent executions.
• Toolkits: Pre-built tool collections, such as the MySQLToolkit for database interaction.