What changed, what we tested, and what we learned.
Patch notes report shipped behavior. Guides walk through setup step by step. Research posts show the evidence, method, tradeoffs, and limits behind product decisions.
Give single keys their own hold time, set limits for a third-party API key, choose which stats show, add new keys beside Space and pin voice settings to the keyboard's quick controls.
When a voice model keeps mishearing a name or a term, fix it once in Loquela's Voice Editor and later dictations write it your way. Everything stays on your phone.
Many transcription services keep your audio or transcripts for days or weeks. How to transcribe a recording with Loquela Remote so neither is kept, in your browser or through the API.
Connect Loquela Remote to Claude, ChatGPT or another AI assistant through MCP, then ask it to transcribe a recording and work with the text. The audio is not kept.
Hold the period key for punctuation, see far more in your stats, save a stats picture to your phone, keep pinned clips when you clear, and let a pasted API key set up its provider.
With a third-party API key, Loquela now says where a failed dictation stopped, and tries again by itself when a dictation could not get through and your provider cannot have transcribed it yet.
Swiping down after a period changes the punctuation again, also after a mark you tapped, and number and phone fields get a pad with a period, comma and minus.
Gestures on the letters, a keyboard without its bottom row, an invisible keyboard, a flick down for symbols, stats you can share as a picture and a Settings search that finds near misses.
Stats that stay on your phone, a keyboard that splits by itself on foldables and tablets, Strict Autocorrect, the ClearFlow layout, more Voice Editor settings and a newer Granite Speech 5.0.
Autocorrect catches a Space typed in place of a letter, taps near a key's edge count, dollar amounts keep their space, and there are new choices for app colors, the word Space puts in and clipboard import.
Choose what holding each letter types, and bring your layout, words, dictionaries, settings and themes over from FUTO Keyboard, HeliBoard, FlorisBoard, Unexpected Keyboard or CleverKeys.
Install Loquela, pick a voice model that runs on your phone, check that dictation works with no connection, see where your audio goes, and switch back to your old keyboard whenever you like.
Loquela now works beside a Bluetooth, USB or built-in keyboard, with a slim bar for the microphone and suggestions, a dictation shortcut and corrections on Space. Key vibration can now be felt on phones with simple motors.
See every theme as a keyboard before you choose it. New Loquela Light and Dark looks, a Theme Studio with a live preview, and new choices for keys, icons, fonts, sounds and the voice view.
Three suggestions on the strip, a keyboard that learns your words after two uses, and new ways to teach it: dictation, your own text and a learning file. Next-word suggestions arrive for Italian, Portuguese, Turkish and Bengali.
The first full Google Play release: stronger English typo fixes, mix-up suggestions, a quick-start voice model, complete Thai and Bengali layouts and many fixes.
Hotfix 3 fixes Search and Go in Google and smaller apps without bringing back accidental Discord sends, and adds focused Enter diagnostics for future editor quirks.
Loquela v0.16.0 improves short Parakeet transcriptions, adds free recording controls, quick settings, model switching, and a more capable emoji picker.
Loquela v0.15.0 introduces Batch, Streaming, and Two-pass recognition modes while improving privacy guidance, diagnostics, vocabulary import, and Enter behavior.
Loquela v0.15.1 opens Nemotron-powered Sequential and Both voice modes to beta testers, adds smaller Moonshine v2 models, and fixes local audio transfer failures.
A practical explanation of model downloads, local transcription, the other places speech can go, microphone routing, and what still needs a network connection.
Voice transcription tries to preserve what a person said. AI cleanup interprets and changes it. That product choice affects latency, privacy, and authorship.
Speech models make different tradeoffs in language coverage, context, speed, storage, and working memory. A single hidden default cannot fit every Android phone.