VoiceDebrief iOS / versions
Release history
Every release of this repository. A push to dev whose version file names a version that has no tag yet is a release: CI tags it and publishes the site.
| Version | Date | What changed |
|---|---|---|
| v0.1.5 | 13 November 2026 | Vaults. The vault engine can now create a vault, in the form sgit init writes; CI has sgit clone a vault Swift created. A Vaults tab: keep an SG/Send access token in the Keychain, create a vault, add one made elsewhere by its key, and look inside (head, files, history, branches, the key, and every request, decryption and signature check the engine made). After a run, save the debrief, transcript, run and audio into a vault as one signed commit. In test: no vault has been created from a device yet. |
| v0.1.4 | 13 November 2026 | The developer's lane is built into the test build, at the owner's request, so a note can be sent without pasting a link first; a pasted link still takes its place. Sending still happens only when the button is pressed. |
| v0.1.3 | 13 November 2026 | While recording: the words appear as they are spoken, and moving bars show the microphone is being heard. The live text is a preview; the transcript that is kept is still made from the recorded file when the recording stops. And a way to help judge the models: with a lane link pasted on the Diagnostics tab, a button after each run sends that note (audio, transcript and debrief) to the developer over an sgit append lane, encrypted on the device to the developer's key. Nothing is sent without pressing it. |
| v0.1.2 | 13 November 2026 | The debrief prompt, changed after the first real runs on an iPhone and a Mac: the model was copying quotation marks from the prompt's examples, renaming a heading, repeating a one-line note back as its own key point, and turning a passing remark into an action. The prompt now shows the shape with no quotation marks, says the speaker's own tasks are theirs, and says what is not an action. A note of fewer than 25 words is no longer sent to the model: the transcript is the whole note. |
| v0.1.1 | 13 November 2026 | The first feature, as the plan's quality spike: record a voice note or choose an audio file, transcribe it on the device with Apple's SpeechAnalyzer, and write a debrief with Apple's on-device model, reading a long note in parts. A diagnostics screen reports what the device says about both models and exports it. Nothing is saved yet. In test: nobody has judged the results on a device. Also: the TestFlight upload, proven by build 0.1.0 (4), installed on an iPhone and a Mac. |
| v0.1.0 | 13 November 2026 | The repository starts. An iPhone and iPad app that builds and tests, with one screen that says what is and is not built. The SGitVault vault engine copied from PrompTugal__App__iOS, with the door derivation made to work for any site and checked against the published vector. CI: build and test on a Mac, vault interop with sgit, and this site. The plan, published. |