just played with @hyperfreams @suno and @claude-code

spin up a techno party in your browser 👽

⭐⭐⭐ Codescribe · VOICE WORKSPACE · macOS

Speak into your workflow.

Talk directly to your agents — in their apps, CLIs, and headless harnesses. Your words enter the conversation where the work happens.

Choose who you’re speaking to, give them direction, and keep the conversation moving.

Codescribe’s Demux Bus (cs-bus) streams and routes messages through each agent’s native communication layer. One clear, human-friendly interface for the whole fleet.

Choose an agent. Say what’s next.

Dzisiaj walczymy na HackYeah2026 !!

Our solution, the Paw Patrol's goal to remember where it seen your things - indoor, and be sure everything is all right - outdoor.

It looks for everyday items like lost keys, hidden mobile phones, and left water bottles. Once it goes out, it check for broken fences and other possible anomalies.

You can have a discussion with the Dog over a chat interface of asking questions like "Where is my cup?" and you can see its detection stream in an AR glass.

Reins

Any VLM. Your hands on the reins.

The novelty here is not the model. It is the harness around it.

Reins optimizes the harness for VLM-driven robot control. Making it model-agnostic means its value compounds passively: every improvement from Anthropic, OpenAI, or anyone else lands in Reins for free. We don't have to win the model race. We just have to be the best place to plug the winner in.

Training robots in simulation is cheap. Moving them into the real world usually breaks them. This is why robot deployments are slow and expensive.

We're solving it. With ordinary cameras and a phone.

We scan your space with an iPhone LiDAR, build a calibrated digital twin, and train or control the robots there. In the real room, it moves exactly as it did in simulation. Adding any new robot takes minutes.

Wojtek is an autonomous robotic dog that uses a Vision-Language Model (VLM) to interpret natural-language commands and navigate its environment. It provides real-time visual feedback through a standard monitor, while Snap Spectacles extend the system for outdoor terrain monitoring by displaying key robot feedback directly in the operator’s field of view. In the future, the Spectacles interface could be expanded to support hand-gesture control.

TrashDrop: pinch the air, the robot follows 🦾

Snap Spectacles control a reBot B601 arm with hand gestures, with the overhead camera in your glasses. No controller or calibration. Plus a two-arm SO-101 trash sorter with a MuJoCo twin.

Built at Alien Bazaar 2026.

Today I'm launching fabryka.ai -
INDEPENDENT ARTIFICIAL INTELLIGENCE RESEARCH LABORATORY · WARSAW, POLAND

1. We serve inference of open weight models
2. We train models from scratch
3. We teach people how to train models

Join our discord https://discord.gg/aJ4dAmXaV

XRAY is a mobile sensing robot that sees what you can't and streams it live to AR glasses and to other robots, so an operator (or a cheap robot) can perceive through walls, around corners, and beyond the human spectrum.

Bin There, Sort That

is an automated trash sorting system using computer vision to classify waste, then a robot arm picks items with homography - based positioning.

I've build a a hacker bloc website and we are thinking about doing a startup focused hackerspace in our house 🤘