docs: document Fluent's speed, accuracy, and token improvements

[inhaq]

This pull request was created by @kiro-agent on behalf of @inhaq 👻

Comment with /kiro fix to address specific feedback or /kiro all to address everything.
_{Learn about Kiro Web}

---

Follow-up to the rebrand (#16). After reviewing every merged PR since the fork (#2-#15), the docs were too narrow and partly inaccurate, so this rewrites the README/CHANGELOG/llms.txt around what actually changed, grouped into three themes.

Speed (key-press to paste)

• Warm-microphone session for instant recording starts (skips AVCaptureSession cold start), with idle cool-down (Speed up dictation pipeline: signposts, fast mode, warm mic, lazy history #12).
• Fast mode drops the screenshot and the vision/context LLM call from plain dictation, so the release path no longer waits on a context request (Speed up dictation pipeline: signposts, fast mode, warm mic, lazy history #12).
• Leaner audio hot path (decode-once buffers, Accelerate/vDSP metering, ~30 Hz level throttling), async mic discovery at launch, lazy history screenshots, cheaper Settings rendering (perf: optimize audio pipeline, level UI, screenshot crop, and history load #2, perf: fix laggy Settings tab switching (drop per-card Material + shadow) #8).
• os_signpost instrumentation to actually measure trigger→paste latency (Speed up dictation pipeline: signposts, fast mode, warm mic, lazy history #12).

Accuracy (the "papers" work)

• Whisper-paper ("Robust Speech Recognition via Large-Scale Weak Supervision") robust per-segment decoding using no_speech_prob / avg_logprob / compression_ratio (Improve quiet-voice accuracy: audio enhancement + Whisper robust decoding #11).
• Audio front-end for quiet voices: 80 Hz high-pass, loudness normalization with safe ceilings, max-quality 16 kHz resampling, with full fallback (Improve quiet-voice accuracy: audio enhancement + Whisper robust decoding #11).
• temperature=0 decoding and custom-vocabulary biasing of the raw transcript (Fix dictation accuracy: stop dropping real speech, bias Whisper with vocab #10).

Tokens (on your terms)

• User-configurable Post-Processing Reasoning setting (Automatic/Off/Low/Medium/High) controlling reasoning_effort — not possible in Free Flow (perf: optimize audio pipeline, level UI, screenshot crop, and history load #2).
• Default cleanup prompt rewritten from ~720 to ~370 words, ~halving fixed prompt tokens (perf: optimize audio pipeline, level UI, screenshot crop, and history load #2).
• Default screenshot 1024→768px when one is used (perf: optimize audio pipeline, level UI, screenshot crop, and history load #2).
• <think>/<thinking> blocks always stripped across the model family (perf: optimize audio pipeline, level UI, screenshot crop, and history load #2).

Corrections to #16

• Earlier README implied reasoning is disabled by default; it is actually user-configurable.
• Privacy note now acknowledges the app contacts GitHub for update checks.
• Replaced the vague "click here" download link with descriptive text.

Tested

• Docs-only change; verified the markdown renders and links resolve. No code or build changes.

Summary by CodeRabbit

• New Features
  • Added more flexible dictation options, including warm microphone sessions, fast dictation mode, optional screenshot capture, custom vocabulary biasing, and configurable reasoning effort levels.
  • Introduced improved audio and transcription behavior, plus optional streamed uploads for smoother processing.
• Bug Fixes
  • Improved output cleanup by removing leftover reasoning tags and refining default prompt behavior.
  • Reduced screenshot size and optimized processing for better performance and lower token usage.
• Documentation
  • Updated the app overview, feature highlights, and privacy notes to reflect the latest release and update-check behavior.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 71e7f23f-d1de-46e2-977c-4e176a6ac6c5

📥 Commits

Reviewing files that changed from the base of the PR and between 1863b1e and 581d113.

📒 Files selected for processing (3)

• CHANGELOG.md
• README.md
• website/llms.txt

---

📝 Walkthrough

Walkthrough

This PR rewrites the release notes, README overview, and website summary to describe the Fluent rebrand and updated speed, accuracy, and token-efficiency messaging. It also adds a README privacy note about GitHub update checks.

Changes

Documentation refresh

Layer / File(s)	Summary
Release notes and overview CHANGELOG.md, README.md, website/llms.txt	Release notes and public overview text were expanded to list rebrand details and the updated feature messaging.
Privacy note README.md	The README privacy text now states that the app contacts GitHub to check for updates.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Possibly related PRs

• inhaq/fluent#16: Continues the same Fluent rebrand and documentation updates across CHANGELOG.md, README.md, and website/llms.txt.

Poem

I’m a bunny with a notebook bright,
I hopped through changelogs in the night.
Fluent sings of speed and care,
with update checks floating in the air. 🐇

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly summarizes the docs-only changes around Fluent's speed, accuracy, and token usage improvements.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch docs/speed-accuracy-tokens

---

_{Comment @coderabbitai help to get the list of available commands.}