A LITTLE MORE CLARITY
The little release notes
New name: Minth. Developed by James van Oosterwijk. Official website: minthai.pages.dev. Install version 1.0.88 on each PC; older installations do not update automatically. Existing chats, settings, encrypted credentials, voices and models keep their legacy storage paths; no data migration or new download is needed for the rename. The green symbol and white interface remain. Explicit YouTube commands now accept “open christmas hits on youtbe”, “you tube” and “you-tube”; this is bounded command matching, not improved speech recognition.
New: YouTube and volume together
“Play christmas hits on youtube and put my volume to 100” now performs one validated YouTube/volume pair, in either order (English/Dutch). Ask first previews both; Automatic runs both without model inference. Both clauses validate before either runs. Invalid/extra/unsupported clauses do not partially execute. 100% can be loud; playback is not guaranteed. The exact-command packaged test mocked both actions, so actual volume was not raised and real media was not played. Qwen2.5 and all saved settings remain.
Minth 1.0.88 · local AI public Windows edition.
Retained from 1.0.85
Restores Qwen2.5 1.5B Instruct Q4_K_M after slow responses with the larger model. Optional automatic DuckDuckGo web excerpts remain: enable once in Tools & voice, then current public questions search automatically. Saved consent stays unchanged; fresh profiles default off. Excerpts are not full-page verification. No cloud AI, paid API or automatic retry.
Updating each laptop
Install Minth-Setup-1.0.88-x64.exe on each PC; older installations do not update automatically. Existing verified Qwen2.5 files are reused. If absent, approve the approximately 1.14 GB model/runtime download. Qwen3.5 files are retained but not loaded. Voice downloads stay separate. 8 GB RAM recommended and at least 2 GB free disk for local AI. Chats, voices, settings, encrypted credentials and saved approval policies are preserved. No personal profiles, keys or downloaded models are bundled.
Performance and tests
A real rollback greeting took about 1.5 seconds after a 14-second preload on one PC. These are individual samples, not guarantees for another laptop. Dutch and factual quality remain poor; not ChatGPT-level. Packaged CPU inference, cancellation/recovery, web-excerpt consent/UI, ten-action previews, drawings and mock media/volume routing passed. Real Google lookup was blocked by verification; prior YouTube lookup found an ID, not confirmed playback. Windows volume helper read/same-value write passed.
Retained safeguards
Ask first is the new-profile desktop default; optional Automatic retains validation and no-overwrite rules. Ten launches, simple drawings, explicit Google/YouTube commands and output volume 0–100%. No arbitrary clicking, typing, deletion or shell commands. Only the current public query goes to DuckDuckGo; no chat history, attachments or custom instructions. Opened sites use normal browser privacy rules. Cloud AI/speech remain disabled.
No security or accuracy guarantee, automatic updates or support SLA. Unsigned: verify the checksum; never disable protections. Back up and close Minth before updating.
Verify this installer
Filename: Minth-Setup-1.0.88-x64.exe
Size: 120280819 bytes
SHA-256e9e94742d0cb6e26aa8a051e282ef169b77f42774372f6ee41354aebe53a276a
Third-party software
Keep bundled Electron/Chromium/dependency notices. Qwen2.5 uses Apache 2.0; llama.cpp uses MIT. Separately downloaded Whisper/Piper/voice models retain upstream terms. No Dii model included. Setup and safety guidance →