I've been through openclaw, hermes, omp, amp, opencode, claude. They don't do what i want: code and automation. At least not as straight forward as i wish it would be. This is Poimandres PINE → npm install -g Poimandres/pine pine init pine

@pmndrs 👀 @bot investigate this and write up a report for me on why I'd care and what I'd personally use it for

it's react for ai basically. claude, for instance, is just input chat. "do this do that". and maybe a scheduler. but pine agents react to any outside stimuli. from watching folders, commands, websites, ftp endpoints, zigbee, webhooks, to any other event source. meaning you can hook them up and let them react to and operate your physical devices in the house, or digital assets. for instance, i bought a scanner for bills. i drop a paper in, press OK, pdf lands in /documents/scans, an agent watches that and extracts the payment info, prepares the banking. you can code all this in claude. but these are agents you can talk to, that have memory and understanding of events.

+ Opencode like CLI, everything's interactive, with support for images, videos, git, diff, ... + Automation like openclaw, without the configs, you just tell agents what you want. they never mess with configs, all goes through tool calls. + Driven by PI ----------------- = PINE

Real world example: main > bought a bambu, here's the protocol *drops pdf*, make it an event source main > create agent "printer" ⭐️ *pin it* printer > watch my bambu, if moisture is above 20%, dry. if it reaches 15% you can stop.

Another: main > here's my ring cam, add to events main > create new agent "kevinmccallister" ⭐️ *pin it* kevinmccallister > you watch the house now, use ring from events. should anyone try to break in while we're out, ring the alarm

Chatting abroad: main > watch telegram, so i can chat with you when im out pine > *opens token dialog* (agents never see tokens)

It ships a few ready-made event sources: commands, folders & file watchers, ftp, http, mqtt, telegram and webhooks. But you can easily make your own. It's a typescript service with a schema. Once in all agents can subscribe against it. Other inputs: MCPs, skills, scheduler

@stolinski @kenwheeler @theo don't think it crosses into t3, it's more for self host automation scenarios and small code tasks.

@pmndrs Coding and automation have different rhythms: one needs tight feedback while editing, the other needs durable state between runs. Making that transition predictable matters more than simply adding another set of tools.

@pmndrs Try @Empryo_ai youd be surprised

i saw the site but if it does do reactive agentic engineering it doesn't mention it. you are a neural network with eyes and ears and feelers that feed impressions into a network. i don't know why every single harness i come across just skips that. they're all just chat directional input.

@pmndrs pine has a way simpler workflow than all those tools you listed, no question.

@pmndrs such a nice demo

@pmndrs @Teknium does Hermes do this already?

@pmndrs there's an engineer shipping an open computer-use runtime and a human-agent issue board who'd actually wire pine into his own tooling. couldn't get a dm through so the intro's here