Neuron AI is the first PHP agentic framework

Joined March 2025
Guide your coding agent. Don't follow it. 🧭 It writes the code. You decide what should exist: where the memory lives, which step needs a human yes, how a run resumes. Follow it: code that runs. Guide it: an app your team can keep building on. #PHP #AIagents
1
5
101
Want to try Neuron v4 without reading the docs first? Six steps on your machine, one question in a chat, and this is what you get at localhost:8000 👇 #PHP #AIagents
1
1
5
169
What you need: PHP 8.3+ with phpredis, Composer, Node 20+, Docker (or your own MySQL 8 and Redis), an Anthropic or OpenAI key. git clone the repo, cd into it docker compose up -d composer setup add your API key to .env php artisan db:seed composer run dev
1
1
68
Agent-written code passes the demo, good architecture or not. Phase 2 is where you find out if your team got an architecture it can build on, or something that just works but is a mess. The Kickstart: one session with the people who built Neuron, on your project. 👇 #PHP
1
2
129
Ask an AI agent for your KPIs and you usually get a paragraph of numbers. In this demo the agent builds the dashboard instead: KPI cards, a revenue chart, a top products table. Live, inside a Laravel app, with the agent written in PHP. 📊
1
3
96
How it works: the agent reads the DB schema, runs read-only SQL, then calls render_cards, render_chart or render_table. It picks the component, passes the data, and Vue components stream into the chat over AG-UI. Neuron v4 on Laravel 13.
1
39
1/ His agents never call each other. The workflow decides what runs next. Jules Langlois built a personal "Jarvis" with small PHP agents and a Neuron workflow, then wrote up how he did it. 🧵
1
1
2
68
7/ He's upfront that it's a prototype: "this prototype was built with AI coding assistance, and it is meant as a starting point, not production code." He tested it on Neuron 4.1 with simulated agents and in-memory adapters, failure and resume included.
1
13
Your agent's chat froze at 10 KB. No exception, no error in the console. Valerio, Neuron's CTO, hit this after moving an agent into a queue worker: one tool result (the content of a file) went over Pusher's 10 KB limit, and the message was simply rejected. 🧵
1
2
66
Every run ends with one of three terminal events: completed, interrupted, failed. The UI knows when to stop waiting. The chat history stays the source of truth, the stream is a convenience layer on top. To set it up, ask your coding agent to use the /neuron-streaming skill.
1
1
13