v0.8.2

google-gemini/gemini-cliv0.8.2Jun 1, 2026by github-actions[bot]

AI Summary

A feature-rich patch release adding HIP (AMD ROCm) backend support, pause/resume functionality for downloads, and fixes for RAG limits and CSP compliance.

Key Highlights

Added HIP (AMD ROCm) backend support
Implemented pause and resume for model downloads
Fixed RAG attachment size limits
Fixed CSP compliance by stopping Tauri nonce-ing style-src
Added API key copy and reveal buttons

New Features

HIP (AMD ROCm) backend support
Pause and resume model downloads
API key management UI improvements

Full Release Notes

## What's Changed
 - fix(rag): honor configured attachment size limit in document parser by @qnixsynapse in https://github.com/janhq/jan/pull/8248
  - fix(mlx): register mlx capability so its commands pass ACL on macOS by @qnixsynapse
  - fix(threads): persist assistant parameters so sampler settings survive restart by @qnixsynapse
  - fix(csp): stop Tauri nonce-ing style-src so unsafe-inline works by @qnixsynapse
  - test(extensions): run llamacpp-extension tests in the global suite by @qnixsynapse
  - fix(llamacpp): raise default ctx-size to 8192 by @qnixsynapse
  - fix(ui): move toast accent tint to CSS class to satisfy CSP by @qnixsynapse
  - fix(ui): restore toast styling on Windows; tint toast with accent color by @qnixsynapse
  - fix(ui): make notification toast visible and theme-aware by @qnixsynapse
  - fix(i18n): add Italian to language switcher, fix Catalan locale code by @qnixsynapse
  - fix(providers): surface friendly message on transport-level request failures by @qnixsynapse
  - feat(providers): add copy and reveal buttons to API key inputs by @qnixsynapse
  - fix(updater): bind requestIdleCallback to window to avoid Illegal invocation by @qnixsynapse
  - perf(llamacpp): defer backend dep verification to browser idle by @qnixsynapse
  - perf(updater): defer backend update check to browser idle by @qnixsynapse
  - build(mlx): build the MLX plugin and extension on macOS only by @qnixsynapse
  - perf(updater): defer initial app update check until after first paint by @qnixsynapse
  - fix(llamacpp): start router before deferring backend update check by @qnixsynapse
  - fix(server): list MLX models in Local API Server default model picker by @qnixsynapse
  - fix(chat): make attachment chip text legible in user bubble by @qnixsynapse
  - feat(settings): add toggle for automatic app update check by @qnixsynapse
  - fix(assistant): make the thread's agent authoritative so in-chat switching works by @qnixsynapse
  - fix(startup): increase logo–caption gap and center loader caption by @qnixsynapse
  - feat(downloads): pause and resume model downloads from the hub panel by @qnixsynapse
  - fix(chat): route prompt-too-big errors to the Increase Context Size banner by @qnixsynapse
  - fix(llamacpp): recover crashed models and surface mid-request crashes by @qnixsynapse
  - feat(startup): branded loader with phased status; defer MCP past first paint by @qnixsynapse
  - fix(linux): keep tao's draggable Wayland titlebar; follow GNOME button-layout by @qnixsynapse
  - feat(llamacpp): add HIP (AMD ROCm) backend support by @qnixsynapse
  - fix(llamacpp): unblock UI from router startup; watchdog against blank window by @qnixsynapse
  - style(message): use primary color for user message bubble by @qnixsynapse
  - chore(release): bump app and plugin versions to 0.8.2 by @qnixsynapse



**Full Changelog**: https://github.com/janhq/jan/compare/v0.8.1...v0.8.2