
300 users ask your agent at once. With general-purpose real-time servers, answers arrived more than 11 seconds late (p99, our benchmark). IQ-Stream: 2.4 ms. Included in Neuron IQ 👉 stream.neuron-iq.dev #PHP #AIAgents

Neuron AI is the first PHP agentic framework

300 users ask your agent at once. With general-purpose real-time servers, answers arrived more than 11 seconds late (p99, our benchmark). IQ-Stream: 2.4 ms. Included in Neuron IQ 👉 stream.neuron-iq.dev #PHP #AIAgents

Friday, 6 pm. A bug in production, and your customer's agent is down. Who do you call? Neuron IQ: priority fixes and a direct channel with the team that created Neuron 👉 neuron-iq.dev #PHP #AIAgents

7/ Local demo only: default passwords, no HTTPS. Runs on ai_neuron by Antonio De Marco of @nuvoleweb, still 1.0.x-dev, no stable release. Save this for the day your agent outgrows the request. Code: buff.ly/4oZ7VKW Talk: buff.ly/XcNH9ZK #PHP #Drupal #AIAgents

6/ That handler method is 21 lines with blanks and comments. Only that method. The price: a worker to keep alive and a hub to run. For whoever approves your stack: long agent runs shouldn't compete with page requests for PHP workers, and the agent stays in PHP.

5/ You need something in between: Pusher, Soketi, Mercure. The demo uses Mercure: ->setChannel(fn () => new MercureChannel($this->hub, 'neuron-demo/runs/' . $input['runId'], private: FALSE)); Neuron forwards the events there. The browser connects to the hub, not PHP.

4/ A separate process runs it: dr messenger:consume asynchronous That's Symfony Messenger, in Drupal through the sm module. The web worker is free as soon as the 202 goes out. Which opens the next problem: how does the output reach the browser?

3/ This is the whole controller action: $this->bus->dispatch(new AgentRun($agent, $request->toArray()), [new TransportNamesStamp('asynchronous')]); return new JsonResponse(NULL, 202); Look for the line that runs the agent. There isn't one 👀

2/ The web server, PHP-FPM and any proxy in front of them can each end the request on their own timeout. Raise them all and the run still holds a PHP web worker: 20 minutes of agent is 20 minutes that worker serves nobody else. So the agent has to leave the request.

1/ Your PHP agent starts taking 10, 20 minutes to answer. What happens to the HTTP request that started it? @Barbalerio walked through it at @drupalcampitaly. The problem is the same in any PHP app. The code below is from a @drupal demo. 🧵

valantic is building agentic applications in PHP with Neuron. To get up to speed faster, their team did a live Kickstart session with @Barbalerio , creator of Neuron. 🎥 Neuron IQ: direct support from the team that created Neuron 👉 buff.ly/ck1HIOu #PHP #AIAgents

Attach a delivery channel to a Neuron agent, and streaming responses are forwarded in real time—no need to keep the front end synchronized with an HTTP request. Full video: youtube.com/watch?v=57L07JpF… #Realtime #DevTools

Your user hits a tunnel mid-answer. The middle of the reply is gone. IQ-Stream replays exactly what was missed, then goes live again. 0 events lost on reconnect (our benchmark). Included in Neuron IQ and Premium, coming soon 👉 stream.neuron-iq.dev #PHP #AIAgents

Adding an AI agent to a PHP app used to mean leaving PHP. Neuron changed that. Coming soon, Neuron IQ: the license with advanced extensions, priority fixes and direct support from the team that created Neuron 👉 neuron-iq.dev #PHP #AIAgents

See what Neuron IQ and Neuron IQ Premium include 👇 neuron-iq.dev/ Agents already in production? Tell us where your project stands via "Talk to us" on the site.

See what Neuron IQ and Neuron IQ Premium include 👇 neuron-iq.dev/ Agents already in production? Tell us where your project stands via "Talk to us" on the site.

See what Neuron IQ and Neuron IQ Premium include 👇 neuron-iq.dev/ Agents already in production? Tell us where your project stands via "Talk to us" on the site.