github omer-faruq/assistant.koplugin v1.17

4 hours ago

📦 Android users: this release repairs the accelerated text renderer that was silently broken on Android — AI answers and dictionary results will render noticeably faster after updating.

Highlights

Do more from the screen: tune your AI providers, test them, and update the plugin — without opening a configuration file.

What's New

Provider Setup, Now Fully On-Screen

  • Reasoning Option — dial back (or off) a model's "thinking" in one tap. The dialog shows only the choices your selected provider understands, applies immediately, and never touches your config file.
  • Test Connection — probe a new provider right from the add/edit form; tells you instantly if the address, key, or model name is wrong, with the provider's own explanation (your key is never shown).
  • Cleaner provider management — Add moved to the presets menu, Delete now lives inside Edit with a confirmation, simplified button row.
  • Better model picker — title names the provider, "Manual" renamed to Custom, explicit OK to confirm.

OTA Updates You Can Trust

  • The update dialog now pre-fills the right target for you: stable installs upgrade to the latest stable release by default, development installs stay on the bleeding edge — unless you explicitly type a different branch or tag to switch.
  • New releases are checked quietly in the background (at most once per 48 hours) with a toast when one is out.
  • Failed downloads are reported persistently instead of flashing by, Android installs are far more robust, and only the files the plugin actually needs are installed.

Improvements & Fixes

  • Your Ask dialog choices stick — Web Search and Copy-to-Clipboard now keep their state after a restart instead of resetting.
  • Consistent dialog buttons — confirm buttons are now uniformly OK across all dialogs instead of a mix of Save/OK.
  • Errors in plain language — see what the provider actually said ("invalid API key", "model not found") instead of a raw data blob.
  • Quicker recovery from rate limits — retries now wait at most 5 seconds (was up to a minute) and show the reason.
  • Follow-up suggestions per prompt — each prompt decides whether the model offers follow-up questions, instead of one global switch.
  • Simpler config template — fresh setups are built around four clearly-labelled service types; existing configs unaffected.
  • Cleaner debug logs — verbose request data trimmed, so logs attached to bug reports stay readable.

Translations

All languages refreshed with the new on-screen settings.

Don't miss a new assistant.koplugin release

NewReleases is sending notifications on new releases.