Kostiantyn

LLMs × Dev × Analytics × Business.

[Kostiantyn @codalytica]

Put it up here if you want your own: https://t.co/sug0QtdRAS
Bring your own model key (not reselling tokens). Happy to get torn apart on the approach.

[Kostiantyn @codalytica]

I was out, pulled out my phone, dropped this week's sales CSV into a Telegram chat, and asked "which SKUs are killing margin?"
Got the breakdown back in the same thread. No laptop, no terminal, no re-uploading, no re-explaining the schema for the 10th time.
It's just an open-source AI agent (Hermes) — but with its own machine, running 24/7, that I message like a coworker.
I didn't build the agent. I built the boring part: getting it running in minutes instead of an afternoon of DevOps.

[Kostiantyn @codalytica]

At first glance, Grok Build looks really cool.

[Kostiantyn @codalytica]

@bluehatone I can recommend trying https://t.co/afUUHoBf4g
The service lets you create multiple agents, each running in its own isolated environment.

[Kostiantyn @codalytica]

@IBuzovskyi The best part is that Hermes learns from its mistakes. OpenClaw just keeps making the same errors and gets caught in a continuous fix loop.

[Kostiantyn @codalytica]

@kartojal @santimenendez19 As an alternative to a VPS for Hermes, I’d recommend https://t.co/afUUHoBf4g

[Kostiantyn @codalytica]

@kfayazinc You can try CageBox instead of setting up a Hostinger VPS: https://t.co/afUUHoBf4g

[Kostiantyn @codalytica]

What I Built

Cagebox is managed hosting for self-hosted AI agents
(Rust + Firecracker microVMs, ~590ms wake from snapshot,
hardware-isolated — same sandbox AWS Lambda uses).

Each agent gets its own kernel. Idle? Snapshotted
to disk. A Telegram or Discord DM arrives? Awake
before the chat client retries.

– Website: https://t.co/yoJRqOdar7
– Hermes Agent + OpenClaw runtimes
– Bring your own key

#AIagents #Firecracker #buildinpublic

[Kostiantyn @codalytica]

@salaga1497 I’d recommend trying https://t.co/afUUHoBf4g — it lets you run Hermes or OpenClaw from the UI in just a couple of minutes, with no DevOps setup needed.

[Kostiantyn @codalytica]

@MichaelHyatt @Hermes_agentAI @gregisenberg @AlexFinn @NetworkChuck You can try it here: https://t.co/b9zcM6R2cu .
Setup only takes a couple of minutes.

[Kostiantyn @codalytica]

@visitor_xyz @pantarhaia I’d recommend trying CageBox instead of setting up a VPS: https://t.co/afUUHoBf4g

[Kostiantyn @codalytica]

@DavidOndrej1 Tmux is great, but it’s important to manage your own context. Personally, I don’t feel productive when I start working across two or more claude code sessions at the same time.

[Kostiantyn @codalytica]

@Fluyeporlaweb @israfill You can also try https://t.co/afUUHoBf4g instead of a VPS. It should be faster than setting up a server manually.

[Kostiantyn @codalytica]

@imshreekumar You can try it here: https://t.co/afUUHoBf4g

[Kostiantyn @codalytica]

@marclou @arian1_11 Cool. Congratulations. If possible, please share the logic behind the pricing, because looking at the annual revenue, I would expect a higher price.

[Kostiantyn @codalytica]

@suni_code https://t.co/9TIuyf4y4N — workflow automation with a focus on system performance. Built for teams that care about how fast and reliably their processes actually run.

[Kostiantyn @codalytica]

@ardiana008 hono + tanstack start

[Kostiantyn @codalytica]

@theo claude code inside tmux works really well

[Kostiantyn @codalytica]

I use LLMs almost every day. ChatGPT and Claude for learning, Claude Code and Codex for coding, Comet as my main browser. Still figuring out where OpenClaw fits in.
But when I talk to former colleagues and friends who are devs and analysts, many say they barely use LLMs at all.
Same with companies. Some engineering teams could get a real boost just from adopting Cursor or Claude Code, and they still don't.
Is it inertia? No real pain point yet? Or just "everything works fine as is"?

Curious what you think.

[Kostiantyn @codalytica]

@TheTechWorldPod @OmriBuilds +1. I analyzed 3,731 startups with verified revenue data — 52% are at $0 MRR.
Early on, it’s often not traffic; it’s unclear value/offer → no first-$ conversion.
(data thread here: https://t.co/gM3DwJIR6C)