Ultimate AI Connector for WebLLM Changelog

Version 1.1.0 — Yakaburitswa musi wa2026-04-09

Chitsva: Floating chat widget ine admin-bar status indicator — chero mushandisi akapinda muAccount anogona kupa browser-side LLM prompt zvakananga kubva ku front end.
Chitsva: SharedWorker runtime — MLC engine yava kumhanya muSharedWorker kuitira kuti ma tab akawanda agovane GPU session imwe chete pane kukwikwidzana nayo.
Chitsva: apiFetch middleware interceptor — zvikumbiro zveWordPress REST zvinoenderana neAI Client SDK pattern zvinofambiswa pachena ku local WebLLM broker, pasina loopback HTTP round-trip inodiwa.
Chitsva: Widget settings UI muConnector panel yekubatidza kana kudzima chat widget uye kugadzirisa auto-prompt behaviour.
Kugadzirisa: Kumanikidza IndexedDB cache backend kuitira kuti kudhawunirodha kwe model weight kurambe kuripo kunyange paine HuggingFace xet CDN redirects dzinoputsa default Cache API path.
Kugadzirisa: Kusvetuka context_window KV-cache override yemamodeli eembedding (haana decoder uye override yakakonzera runtime error).
Kugadzirisa: Kushambadza cold-start candidate model mu/webllm/v1/models worker tab isati yarodha, kuitira kuti vatengi veSDK vaone model pakarepo.
Kugadzirisa: Kubatanidza WebLLM mu wpai_preferred_text_models filter kuitira kuti AI Experiments feature ifambiswe ku browser engine kana model yagadziriswa.
Zvakavandudzwa: Cache-busting, content normalisation, uye hardware-reference fixes zvakaonekwa panguva ye end-to-end testing.