Conversation
The composer's reasoning level was read-only, so changing it meant opening the model picker and reconfiguring the model. Make it an inline selector instead. The levels are read from the selected model's own option descriptor, so the client never assumes which levels a provider exposes, and choosing one writes the same selection shape the model picker's configuration screen writes through the composer's existing selection binding. Descriptors that are not a choice list, and values the model does not offer, keep the previous read-only and hidden behavior.
On-device review of the selector asked for two changes. The two prompt-tier "ultra" levels are not wanted in the composer, so the control excludes exactly the `ultracode` and `ultrathink` choice ids and documents that as the single product rule in an otherwise descriptor-driven control. A level excluded here is still displayed while it is the effective one, so the composer keeps telling the truth about a setting made elsewhere, but it is never offered or checked. The composer sits at the bottom of the screen, where an adaptive menu flips its contents and puts the highest level nearest the finger. Pin the menu to the descriptor's declared order so the lowest level reads first.
…ulary Mic button in the composer drives Apple's DictationTranscriber (iOS 26, mic-permission only) with contextual strings extracted from the draft, recent thread messages, and titles, plus a conservative post-recognition corrector for near-miss jargon. Volatile partials preview under the editor; finalized segments append to the draft. DictationTranscriber over SpeechTranscriber because it is the only module honoring contextual vocabulary. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Sequential locale resolution (async calls cannot sit in nil-coalescing), cancellable startup task with checkCancellation after every suspension, full teardown on every engine failure path with weak engine captures to break the results-task cycle, scenePhase and audio-interruption handling, send disabled while dictation is active, exact-first correction ranking with a plain-word fuzzy gate, UUID-fragment filtering, bounded per-message extraction, and a static recording indicator. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ronment model Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ttings The hypothesis now types directly into the draft and refines in place (user edits mid-recording rebase the committed text instead of being overwritten). Recording state is a red mic with a ring that breathes with the live microphone level instead of a stop square. New device- local Voice Dictation settings: hide the button, and choose between Apple's Dictation model (contextual vocabulary) and General speech model (better prose, corrections only). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…, caret follows dictation Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ation The whole input surface (editor, mic, send) gets a soft accent glow that pulses with the live microphone level; the equalizer bars are gone. Send now works mid-dictation: the live hypothesis is already in the draft, so sending cancels recognition and submits what is visible. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
github-actions
Bot
added
size:XXL
vouch:trusted
Thread transfer impact
This comment will update automatically after the next completed run. |
t3dotgg
force-pushed
the
feat/issue110-reasoning-selector
branch
from
September 6, 2026 22:17
2524de5 to
4e361bf
Compare
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Mic button in the composer drives Apple DictationTranscriber (iOS 26, mic permission only, fully on-device) with contextual vocabulary rebuilt each recording from the draft, recent thread messages, and titles. Live hypothesis types directly into the draft and refines in place; a halo glow behind the composer pulses with the live mic level; send works mid-dictation; Settings gains a Voice Dictation section (button toggle, keyboard-while-dictating toggle, Apple Dictation vs General speech model picker). Includes fixes from a cross-vendor review (cancellation races, teardown, backgrounding, correction-algorithm ranking).
Verified: full test suite green on simulator including 14 vocabulary tests; ran on an iPhone 16 Pro.
Built by Claude Fable 5 on Claude Code.
🤖 Generated with Claude Code