Neuron AI - PHP Agentic Framework

19/06/2026

Watch as an AI agent goes through its reasoning process. It identifies the current date and time, searches for information, and then calculates a square root using a dynamically retrieved tool. See how AI can adapt its approach to solve problems. Full video: https://www.youtube.com/watch?v=qYmidHAXEYM&t=1s

18/06/2026

How do you build a hyper-personalized customer chatbot without overloading the prompt context or shifting your stack to Python? 🌐

When building autonomous assistants for e-commerce or commercial platforms, the real challenge isn't connecting to an LLM: it is making that LLM smart enough to know your customer’s specific history, likes, and past orders without hallucinating.

Amitav Roy just published a practical technical video showing exactly how to bridge this gap and build a highly personalized chatbot natively in PHP using Neuron AI.

He demonstrates how to structure an agent capable of dynamically accessing user data to deliver real business value:

🔹 Customer Context Injection: Feeding the agent with specific user profiles, past purchases, and user behavior parameters.

🔹 Targeted Recommendations: Designing the logic so the chatbot can evaluate past data and suggest actual products the user wants.

🔹 Smart Offer Awareness: Empowering the system to trigger specific, relevant promotional offers based on user preferences in real time.

This isn't about chasing temporary artificial intelligence hype: it is about solving concrete production problems. By using a framework-agnostic tool like Neuron AI, you can build these intelligent logic layers directly on top of your existing PHP infrastructure, maintaining total control over token consumption and data isolation. 🚀

If you want to see how to transform raw customer data into actionable context for an autonomous agent, Amitav’s video provides a clear blueprint.

Watch the complete breakdown here:
👉 https://www.youtube.com/watch?v=Dk2QV1NA0xE

We are here to help you scale production-ready applications.

In this video, I am going to show you how we can make a chatbot whi...

18/06/2026

Watch the agent's ex*****on unfold on the inspector dashboard. See how it requests the current date and performs a search for calculations.
Full video: https://www.youtube.com/watch?v=qYmidHAXEYM&t=1s
*****on

15/06/2026

See how an AI agent uses tools to get the current date and calculate the square root of the day number. It's all about sending the right prompts and letting the tools do the work.
Full video: https://www.youtube.com/watch?v=qYmidHAXEYM&t=1s

12/06/2026

This tool search middleware's core function is to inject a special tool into the agent's default tool list. It retrieves the full list of tools available to the model, allowing it to search and select the right tools for a given task based on a query string.
Full video: https://www.youtube.com/watch?v=qYmidHAXEYM&t=1s

11/06/2026

Taking the stage at the International PHP Conference in Berlin was about one clear mission: challenging the assumption that production-grade AI belongs exclusively to Python. 🌐

These intense days have been fully focused on deep architectural discussions, code implementations, and connecting with the international community. Seeing our slides up on that stage and presenting the core concepts of Neuron AI to fellow engineers validates the exact reason we built this framework.

Scaling an open-source project and pushing a whole ecosystem forward doesn't happen in a vacuum. It requires shared challenges, continuous learning, and strong technical relationships.

Berlin was also the perfect opportunity for real, face-to-face networking. We are glad we could grab a physical copy of “Anatomia di una mente artificiale” by Enrico Zimuel.

Enrico hosted our team back in February at the PHP User Group (PUG) in Turin during the launch of Neuron v3.

Seeing recognized leaders in our community publish high-quality technical literature that grounds AI into solid software engineering principles is exactly what developers need to move past the temporary hype and focus on production-ready systems. 📈

The global landscape is shifting. PHP developers are realizing they can build, deploy, and scale autonomous agentic systems natively, leveraging the exact stack they already master. 🚀

A huge thanks to the team for the incredible effort during this international tour, to everyone we crossed paths with in Berlin, and to Enrico for his ongoing engineering contribution to the ecosystem.

If you are a PHP developer looking to integrate stable AI workflows without changing your stack or rewriting your codebase from scratch, let's connect.

You can check out Enrico’s book here: 👉 https://amzn.eu/d/07Gnmf5O

And dive into the Neuron AI architecture on GitHub: 👉
https://github.com/neuron-core/neuron-ai

11/06/2026

Learn how the tool search middleware works, accepting a pool of tools for model searches. This allows you to easily manage and access various tools, like those within the calendar toolkit, without them being added by default to every request.
Full video: https://www.youtube.com/watch?v=qYmidHAXEYM&t=1s

05/06/2026

Exploring the growing world of AI agents. As a frontend developer, I've found ChatGPT invaluable for coding assistance for years. Now, I'm integrating AI agents into site features, developing entire applications and software. Excited to see what Neuron brings to the table!
Full video testimonial: https://youtu.be/kW3RXaNC0LM

04/06/2026

Why should your AI agent care which LLM provider answers its calls? 🌐

In a well-designed system, it shouldn't. But in reality, hardcoding your application to a single provider creates a fragile architecture: you are stuck with their pricing, their rate limits, and their latency spikes.

When we designed v3 of Neuron AI, we spent a lot of time on the unsexy part of the infrastructure: the Unified Messaging Layer. We wanted a single, native PHP representation for messages and tools that every provider (OpenAI, Anthropic, Gemini) could understand.

That early architectural decision is now paying dividends.

Because the agent is completely decoupled from the underlying LLM, we just released a new component: neuron-core/router. It allows you to route single inference calls to different providers dynamically, and the agent doesn't even notice the switch.

This solves three massive headaches when running agents in production:

🔹 Token Budgeting: You can route requests to premium models (like Anthropic) only when complex tools are active, and offload simpler validation steps to faster, cheaper models.

🔹 Native Multimodality: Dynamically shift the single inference call to Gemini when your workflow detects a video file, keeping everything inside the same agent lifecycle.

🔹 Rate Limit Management: Use the built-in RoundRobinRule to distribute the load across multiple accounts or providers without adding external proxy layers.

In software engineering, clean plumbing isn't an academic preference: it is what keeps your system production-ready as the market changes. 🚀

If you want to see the exact code implementation, understand how the RouterProvider implements the standard interface, and check the built-in routing rules, the deep dive is live on our blog.

Read the architectural breakdown:
👉 https://inspector.dev/routing-inference-calls-between-providers-in-neuron-ai/

The router is completely framework-agnostic. You can inspect the source code and contribute to the package directly on GitHub:
👉 https://github.com/neuron-core/router

04/06/2026

I'm a front-end developer helping to integrate features and develop entire pages and applications. This event has been incredibly helpful with its interesting topics that will aid future self-improvement.
Full video testimonial: https://youtu.be/kW3RXaNC0LM