Kacper Włodarczyk

AI Engineer @VstormCommunity | Creator of pydantic-deepagents | PydanticAI & LangChain contributor | Building AI agents in public 🛠️

[Kacper Włodarczyk @Kacper95682155]

MIT, macOS, Node 18+. It is early (v0.1), so I want to hear what categories you would add.

What is the biggest thing silently eating your disk right now?

Repo: https://t.co/Mc7hCSKcjS

[Kacper Włodarczyk @Kacper95682155]

A server can ask you for input: it answers input_required, for example with a form, and the client resends the call with your answer.

Protocol sessions are gone too. For state between calls, a server hands out an ID, like basket_id, and the model passes it back as an argument.

[Kacper Włodarczyk @Kacper95682155]

MCP dropped the initialize handshake in revision 2026-07-28. 🤝

Each request carries its protocol version and client capabilities in _meta. To see what a server supports, call server/discover. Versions are dates, and a server that doesn't speak yours lists the ones it does.

[Kacper Włodarczyk @Kacper95682155]

AgenticOS supports reusable skills: Markdown procedures you can attach to an agent. ✏️

For a support-triage skill, describe when to use it and what to check. The model can load the instructions when that task comes up.

Built at @VstormAgenticAI .

[Kacper Włodarczyk @Kacper95682155]

OpenAI added GPT-6 Sol and Luna below Astra this week, at half the price of GPT-5.6 Sol and Luna: Sol is $2 per million input tokens, Luna $0.10.

That and four more things for agent builders, in under two minutes 👇

[Kacper Włodarczyk @Kacper95682155]

@MicroLaunchHQ It's FREE and Open Source! https://t.co/kBdCSLHbFM

[Kacper Włodarczyk @Kacper95682155]

AgenticOS, our AI agent platform at Vstorm, can run tasks on a schedule. ⏰

For a daily issue review, configure the agent's model and repository access, then set a routine's prompt and time. Activate it and use Run now to check the output.

[Kacper Włodarczyk @Kacper95682155]

We've added message navigation to AgenticOS, the AI agent platform we build at Vstorm. 🧭

In desktop chat, hover over a marker to preview a message. Click to jump back to it.

[Kacper Włodarczyk @Kacper95682155]

I just published Claude Code Failed My Writing Test. Codex Did Better. https://t.co/Aiyp9ePEEK

[Kacper Włodarczyk @Kacper95682155]

@spacedevin @Joi2James https://t.co/B6Pl50uZvl

[Kacper Włodarczyk @Kacper95682155]

@TVanWelsenes https://t.co/B6Pl50uZvl

[Kacper Włodarczyk @Kacper95682155]

@Joi2James https://t.co/B6Pl50uZvl

[ClaudeDevs @ClaudeDevs]

Cloud sessions are officially available and out of research preview! They let you keep Claude Code working, even when your laptop is closed.

Existing subscribers get a one-time credit to try them: $100 on Pro, $250 on Max.

[Kacper Włodarczyk @Kacper95682155]

You can't animate anything inside an img tag. The document in there is sealed off from your stylesheets, so no CSS of yours ever reaches it.

Which is why AgenticOS draws its generated avatars as inline SVG instead. About a dozen nodes a face, and they blink. 🫧

[Kacper Włodarczyk @Kacper95682155]

I just published Learn agent components through Amigo’s Arcade https://t.co/XEItIKNZFk

[Kacper Włodarczyk @Kacper95682155]

Building an AI agent, but as a platformer. 🕹️

AgenticOS's arcade turns tools, skills and context into things you collect, then lists the collection in an agent spec.

It's a teaching game. Collecting a tool there doesn't connect a real one.

[Vstorm I Agentic AI @VstormAgenticAI]

Vstorm will be at the PRINTING AI Pavilion for all three days of PRINTING United Expo 2026 as its development partner. 🖨️

Our Co-founder & CTO, Bartosz Gonczarek, will be there throughout the event, starting on Day 1 with a fireside chat on production-grade Agentic AI and AI sovereignty.

The conversation will go beyond what happens in the AI interface: How do you move from everyday AI tools to agents embedded in core business processes? And once they’re there, how do you maintain control over the technology, infrastructure, and data?

We’ll also be sharing what we’ve learned after deploying AI agents in the print industry.

If you’re attending PRINTING United Expo, visit the PRINTING AI Pavilion and come talk to Bartek and the team.

📅 September 23–25, 2026
📍 Las Vegas

[Kacper Włodarczyk @Kacper95682155]

���� GitHub Repo: https://t.co/kBdCSLGDQe

[Kacper Włodarczyk @Kacper95682155]

A browser action as a choice from a list, not a string the model writes.

AgenticOS's browser_choice shipped in v0.0.479. Jev picks the operation and the target, reached through Pydantic AI's typesafe extra.

Asked it for two Wikipedia pages and it ran two browsers at once. 🧭

[Kacper Włodarczyk @Kacper95682155]

These are not the same sentence:

"It searches only your documents."
"It answers only from your documents."

The first is an access boundary. The second is a promise about what a model writes, and nobody configures it.

In AgenticOS the answer comes with its passage. Open it. 🔍

[Kacper Włodarczyk @Kacper95682155]

Discuss a workflow for your team with Vstorm:
https://t.co/9hTVhU8UlN

[Kacper Włodarczyk @Kacper95682155]

A useful first AI agent: help one team find answers in its own procedures.

In AgenticOS, connect the documents and try the questions people actually ask. Include one the docs don't answer. Check the sources before widening access.