Repository navigation
v0.3.1
HTML apps now work out of the box on every platform, Windows starts reliably on Intel GPUs, and the README finally shows what byob does.
Batteries included: HTML apps in every download
The full archives (byob-…) now include byob-artifact-host, the Servo renderer that draws the HTML apps the model writes inline beside the chat. Before, that needed a Nix build. The slim archives (byob-slim-…) are still byob on its own; without the renderer, HTML apps open in your browser.
Windows
- Starts on Intel GPUs. byob now draws with DX12 when it's available. Intel's Vulkan driver could close the window at startup.
WGPU_BACKENDstill overrides the choice. - HTML apps render inline, with or without a GPU. The renderer draws through ANGLE (Direct3D 11), falling back to Windows' built-in software rasterizer. That covers VMs, Remote Desktop and machines without a GPU driver. ANGLE's
libEGL.dllandlibGLESv2.dllship next to it, withLICENSE-ANGLE.
Fixes
- HTML apps no longer stall when a library prints to the renderer's output. The channel between byob and the renderer now tolerates stray text: it's skipped, and logged as "artifact host printed: …". The reason a renderer stops is shown on the app card. Older renderers and older byob builds still work with the new ones.
- Each reply shows one card per artifact, on its last change. A sheet edited 24 times in one reply used to leave 24 cards. The version picker still lists every version.
- Settings: the title and Close button stay put while you scroll.
Also
- New README with a recorded session and screenshots, regenerated by
scripts/readme-media.sh. - Each release archive is now its own download, with
.sha256checksums.
Install
Download an archive below. byob-… has everything: the browser UI (byob --serve) and the HTML app renderer (byob-artifact-host; keep it next to byob). byob-slim-… is byob alone. Each has a .sha256 next to it.
Or cargo install byob (crates.io; without the renderer), or nix run github:Scronkfinkle/byob/v0.3.1.
Full Changelog: v0.3.0...v0.3.1