Atomic Chat Files

🐦 Ornith 1.5 is here!

• Ornith 1.5 9B — vision-capable dense model with a thinking toggle, 262K context, and hybrid attention that makes long context unusually cheap (GGUF)
• Ornith 1.5 35B A3B — 35B MoE with only 3B active, so it runs fast on a single 24GB card (GGUF)

🚀 New Features

• DeepSeek Harness is in integration — a new open-source coding agent you can set up right from the app
• Onboarding now picks models that actually fit your hardware and recommends lighter ones when needed

🔧 Improvements & Fixes

• Cancelling a download no longer fires off a crash report
• Starting a download while a recommendation is fading out doesn't lose the download anymore
• Fixed a crash on older Safari that broke the sidebar and the drawer
• Files without an extension keep their original name when downloaded
• Provider logos are readable on dark backgrounds again
• Japanese is now fully translated across the whole app (#239)
• A batch of build and test fixes so Windows and web builds stop breaking
• Failed downloads clean up after themselves — your existing models stay untouched
• You can now delete installed local models from the app
• Multi-part GGUF models download all their shards, and embedding models are finally detected as embedding models

🙏 Contributors

Thanks to @Vect0rM, @Ayush7614, @dtrknt15 for their contributions to this release!