v0.6.66b476
HF token in TUI settings
hf_token is a writable setting now. Open /settings → API-Keys, paste in your Hugging Face token, and gated repositories you can pull (Llama, Gemma, anything you've accepted terms on) become reachable from the model catalog without exporting HF_TOKEN in your shell first. (#247)
TUI polish
The bare m footer keybinding that opened the model bar is gone. It collided with normal text entry in the chat input. The model bar still opens via Ctrl+M, the slash command, and the click target on the status bar. (#248)
Release pipeline rebuild
The plumbing under pip install --pre lilbee --extra-index-url https://lilbee.sh/cu125/ was broken: the index page loaded, the wheels behind it 404'd. The wheels are now hosted on each version's GitHub release page and the index links to them directly. The flow is automatic from this release onwards: every tag push uploads the full wheel matrix to the release and the lilbee.sh index regenerates from those URLs the moment the candidate finishes.
Two related changes ride along: the QA matrix is decoupled from the release-candidate run (a flaky TUI test on one platform can't block the wheel publish anymore; run qa-matrix.yml on demand when you want it), and the Homebrew / AUR / Nix / Docker fan-out no longer waits on the ~3h45m Windows nuitka build (only the Linux + macOS-arm64 binaries gate it, since nothing downstream consumes the Windows executable).
Docs
The README install rows now link the four backend wheel indexes (cpu / cu121 / cu124 / cu125) so picking the right one is one click instead of a copy-paste.
What's Changed
- tui: drop the m footer keybinding for the model bar by @tobocop2 in #248
- tui: expose HF token as a writable setting by @tobocop2 in #247
- Host wheels on GitHub releases; let the extra-index actually serve them by @tobocop2 in #249
Full Changelog: v0.6.66b475...v0.6.66b476
