Phone layout: keep the top bar on one line, give the preview the pane

Nudge a rebuild — the Space sat in APP_STARTING with a healthy server

Make it usable on a phone, and stop offering models this Space doesn't serve

README: note the dark theme

Add a dark theme and a switch for it

Back to a link in the foreign embed, and scope the exemption to cross-site

Serve the app in the huggingface.co embed, not a link to it

README: ZeroGPU is the default path, and say where to open the app

Default to this Space's ZeroGPU model, keep the browser one selectable

Drop the COOP/COEP service worker — it fights the headers this Space sends

Send the huggingface.co iframe to standalone instead of a blank page

Give the local model back, and warn when the embed cannot isolate

README: the Space serves the model, not just the files

Send the ZeroGPU startup report ourselves — we never call Blocks.launch

Decorate the Gradio handler itself so ZeroGPU sees a GPU function

Drop ChatInterface(type=...) — gradio 6 has no such argument

Serve the agent's model here on ZeroGPU

Add the @spaces.GPU probe ZeroGPU startup insists on

Bind 7860, not $PORT — the Space's own proxy holds 7861

Serve the shipped yourlocalagent.ai build from this Space

docs: add links to WebGPU version and Twitter

feedback: synchronous dataset upload (ZeroGPU-safe)

like/dislike feedback -> JSONL synced to HF dataset

theme-aware card (CSS vars for dark mode)

English UI + model card + duration=120 + reasoning/code placeholders

Artifacts preview (iframe) + bigger token budget

gradio6: drop show_copy_button

gradio6: drop Chatbot type kwarg

Fix ZeroGPU: preload CUDA runtime libs before llama_cpp import

Soyuz chat + reasoning/raw viewer on ZeroGPU (llama-cpp-python)

initial commit