DeepSeek OCR-2 is an OCR ai model that reads in data in a more humanlike manner, resulting in a much more efficient and useful ingestion of information. It's more likely take in your multimodal info and not screw it up.
[via The Decoder] #
skills.sh is an open repository for agentic skills, works accross many different ai models.
#
Kimi k2 is a new model that can apparently design interfaces much much better, thanks to its novel agentic design.
#
Open Code is a claude-code like environment that can host a number of models, open and paid.
#
21st Dev is a repository of UI components. A quick way to ingest design... perhaps.
#
Mobbin is a reference app/site for ui design. the twist: it costs money!
#
n8n is a very powerful automation builder, now with some more ai pixie dust.
#
Google Code Wiki - drop in a github repo and Gemini builds a live interactive guide on how the code works. Autogenerated docs, visual maps and a chat agent that can explain everything to you.
#
Krea AI is a hub for creative visual AI, with models from all over. Their new in-house Realtime Edit looks scary.
#
Tina CMS is an interesting cms option, I'm gonna be playing around with it.
#
Z.ai is a chinese startup which claims to trounce claude code for only 3$ a month. Its AI is reportedly trained on all-chinese Huawei chips
[via The Register] #
Qwen3-tts is a step forward in open source text-to-speech. It can train a voice model from just three seconds of speech.
#
Gemini Canvas is a unified environment to create apps, games, infographics and more. Probably great for presentations.
#
Google Opal creates automations visually but also with ai... somehow.
#
Google Pomelli scans your website and then creates social media posts from that. Sounds not great!
#
Voicebox is an open-source voice synthesis studio.
#