Two reports, one root: the platform's speech engine.
"Loud cracking at the end of every sentence" is the browser/OS TTS popping
between its internal sentence chunks (an idle cancel makes it worse). Hygiene
for that path: never cancel when nothing is pending, normalise the spoken text
(control characters, ellipses, dashes), and a 140ms beat between sections so
speech never runs into the engine's tail.
The real fix is a second voice source that does not use the platform engine:
- lib/providers carries TTS profiles (OpenAI gpt-4o-mini-tts, Groq
playai-tts, any OpenAI-compatible /audio/speech behind the custom provider),
male voices first.
- /api/tts proxies them with the same guards and the key still in headers.
- the picker gains System / Natural, with the provider's voice list.
- lib/podcast-audio plays one clip per section through a hidden <audio>
element, measures the clip's real duration and gives each word a slice of it
proportional to its spoken length - the highlight follows the audio exactly,
and pause/seek/chapters become real audio operations. Clips cache per
session, the next section generates while the current plays, and a voice
change disposes the element and its blobs. Speed applies live (setRate)
without dropping the cache.
podcast-test stubs /api/tts with a real silent WAV and pins the path: the
voice list, the request body, exactly one playing element, and the highlight
riding currentTime. 43 checks. tests/lib.mjs launches Chromium with the
autoplay policy relaxed (a real device has the listener's gesture).
Version 0.7.6, SW v98.
The third segment of the note-powered wizards (Study deck / Homework /
Podcast) takes the same intake - photos, documents, pasted text, the share
target - and turns it into a spoken episode.
- prompts: PODCAST_SYSTEM_PROMPT writes for the ear (plain prose, no markdown,
no visual references), chaptered into 3-8 sections around a length target.
- /api/podcast mirrors /api/generate's plumbing (rate limit, model gate,
provider keys, heartbeats, the 2x deadline, the optional vision pass).
- storage: frezmo.podcasts.v1, zod-validated, included in the backup and the
encrypted sync payload; a new library tab lists episodes with listen /
rename / delete-with-undo, and the wizard switch deep-links in.
- player: the transcript is word-wrapped spans and the word under the voice
rides a lyric-style highlight that keeps itself centred. Two clocks: exact
onboundary events where the browser reports them, a per-word length timer
where it does not (iOS Safari); the first boundary retires the timer.
Pause = cancel + re-speak from the word, which makes resume, chapter jumps
and tap-any-word-to-seek the same operation; the position saves on pause,
page-hide and unmount, so reopening resumes. Speed rebuilds the engine and
keeps playing; the Aero soundtrack pauses itself when the voice starts; a
browser with no voice gets an honest note and a readable transcript.
- the script sits on a glass pane (plain paragraphs over Aero's grass failed
contrast; the scan now covers the player too).
tests/podcast-test.mjs (17th suite) stubs the model and the speech engine and
drives the highlight by hand: 26 checks. All suites green, contrast clean.
Version 0.7.3, SW v95.
Settings -> Providers takes keys for OpenAI, Anthropic (their official
OpenAI-compatibility layer), DeepSeek, Groq, OpenRouter and any custom
OpenAI-compatible endpoint. One adapter (lib/openai-compat) covers
listing, streaming chat, JSON generation and vision, and every model
flow can run on a provider: the chat picker lists each enabled
provider's models (labelled, remembered), and a "Study generation"
choice moves decks, homework, grading and attachment transcription onto
a provider model.
Keys stay in this browser exactly like the Ollama one
(freezmo.providers.v1): never stored server-side, never in a backup or
the sync payload, forwarded only with the request that needs them. The
provider test asserts the key reaches the provider while the routing
header stays on the app side.
New suite tests/providers-test.mjs (13 checks) stubs both protocols:
model listing, chat streaming, study generation and the Settings flow.
Privacy policy updated (en+cs). Contrast 0 findings, axe clean, mobile
126/126, chat 102/102, wizard 95/95, bubbles 34/34, regression 44/44,
i18n 13/13, native 33/33. Version 0.4.7, SW -> freezmo-v72.
- Aero: blue gradients back on the active segment and every primary
(Review due, the card's Study button, the composer's send button); a
frosted backing behind the page heading and the All decks label; a
bottom-weighted rail tint; chat header/composer/welcome panels; calmer
orbs; stronger card veil + a stronger blur bake.
- Dark/Light: a dedicated --track token makes progress bars read (the
first attempt shifted --surface-sunken and the pixel scan caught the
sheet-contrast regression); brighter dark secondary text; darker light
segmented labels and difficulty tints; stronger light card border and
shadow.
- Chat UI: friendly model chip ("Gemma 4 31B") with the raw name in a
tooltip, larger chips, plain centered disclaimer, vertically centered
welcome panel, distinct chat-plus icon for new chat.
- Chat models: every chat-capable model from the server's /api/tags is
selectable (favourites first, search field when long, live-list
validation in the route); default CHAT_MODEL_DEFAULT or gemma4:31b.
- Reduce motion: Settings toggle (default off) that parks the scene,
collapses CSS animations and drives framer-motion's calm mode.
- Folders hint, "New" badge on fresh decks.
Contrast: 0 findings, axe clean, mobile 126/126, chat 102/102, wizard
95/95, regression 44/44, i18n 13/13, native 33/33.
Version 0.4.6, SW -> freezmo-v71.
A user's run failed with the wizard's generic "check your Ollama
settings" card and could not be replicated with the same key - because
the key and model were never the problem: a raw error (a mid-run socket
drop, a wifi/LTE switch, the app backgrounded by the phone OS) leaked
out of the NDJSON reader as a DOMException, and the wizard can only
show the generic card for unknown errors.
The reader now classifies everything: abort -> "cancelled", stall ->
"the model stopped responding", anything else on the wire -> "the
connection dropped mid-run, try again". Pinned with a fetch stub that
kills the stream mid-run in the wizard suite (95 checks).
Contrast 0 findings, mobile 124/124, chat 101/101, regression 44/44,
i18n 13/13, native 33/33. Version 0.4.1, SW -> freezmo-v70.
A user got stuck on the wizard's loading screen: a hung model meant
silence, and silence was indistinguishable from work - the server's own
hard deadline is 2x OLLAMA_TIMEOUT_MS (ten minutes by default), and the
client waited on a dead stream indefinitely.
Both generate routes (deck, homework) now send an NDJSON heartbeat
every 15s, and the client guards the two quiet phases: no response at
all for 45s, and no bytes for 90s mid-stream, both surfacing the
regular error state with a Retry. The caller's abort still maps to the
friendly "cancelled" state.
Verified with a stalled route and Playwright's fake clock in the wizard
suite (93 checks), plus contrast 0 findings, mobile 124/124, chat
101/101, regression 44/44, i18n 13/13, native 33/33.
Version 0.4.0, SW -> freezmo-v69.
- "Check my work" now takes photos and documents of the attempt: documents
are read in the browser into text, photos go to /api/homework as
solutionImages and are transcribed by the vision model before the check
("Reading your solution..."). The finished attempt comes back in
meta.solutionText and is stored as the record's attempt, so "Redo"
re-checks the same work without keeping the files.
- Deleting a deck now drops its due schedules and mistakes from the library's
smart review immediately (the counts resolve against the live deck list);
Undo brings them back, because the learning data itself is never erased.
- The phone override for the glacier blobs was still at 0.1/0.2: it is now a
quarter of the original value there too (0.025/0.05), matching the desktop.
- License: GPL-3.0-only (LICENSE, package.json, README section).
Tests: phase-3 photos-only API + desktop attachment checks (28/28), review
suite deck-delete checks (18/18), i18n/mobile/phase-2 green. README:
decisions 57-58. SW -> freezmo-v45.
Entry points (no sixth tab): Ask in the Library header (desktop button,
phone icon), Ask about this deck in the deck menu, Ask a question on
study/homework results (compact result travels as context), and a Chat
row in the desktop sidebar. Phones get a full-screen route with back +
New chat; the tab bar steps aside.
- Client-side retrieval: chunk scoped decks (~800 tokens), BM25 against
the question + recent turns, top chunks under ~7k tokens, each
labelled with deck + section. Notes come from the deck's stored
sourceText (already kept for Regenerate); decks without it fall back
to generated content and the UI says so.
- Stateless server: /api/chat validates against the CHAT_MODELS
allowlist, trims the history to ~6k tokens, caps context at ~9k and
streams token/thinking/done/error NDJSON. /api/chat/extract reduces
attachments to text through the existing vision step.
- UI: user bubbles + full-width assistant text, streaming with stop,
auto-scroll that pauses with a jump-to-bottom button, safe markdown
(React nodes, no HTML injection; KaTeX dynamically imported only when
a message has math), Thinking row for reasoning, copy/regenerate/
edit-and-resend, model + context pickers (last model remembered),
composer with visualViewport keyboard handling, safe-area, 16px
textarea, Enter-to-send on desktop.
- Usage-limit failures keep the draft and show the shared message with
Settings/Ollama actions; stopping maps the reader abort to a clean
stopped state; offline disables sending with a reason.
- Settings: Keep notes text for chat toggle; deck menu: Delete notes
text. RATE_LIMIT_CHAT + CHAT_MODELS in .env.example and README.
- Tests: chat suite (63 checks, stub Ollama) wired into run-all.
Full set: 12/12 suites, 487 checks. SW -> freezmo-v32.
Also: tests/model-compare.mjs (diagnostic) benchmarks two models on the
same notes (latency, repair rounds, activity types) so the gpt-oss vs
gemma default can be decided on numbers; defaults unchanged.
Check my work (third homework mode)
- One model call with its own system prompt marks the student's own
attempt: verdict, what's right, each issue with why + fix, and the
corrected version that keeps everything the student got right.
Unasked-for extras are never errors. The attempt is typed/pasted
(photos of handwriting are documented as unsupported) and stored in
the record so Redo re-checks the same work.
- Server: mode "check" + solution on POST /api/homework (zod, 400
server.needSolution without one) and the same NDJSON stream; new
checkHomework() pipeline with repair-once validation.
Follow-ups
- New POST /api/homework/refine (plain JSON, RATE_LIMIT_REFINE, zod,
auth before body): shorter / more detail / explain, plain text back,
language preserved.
- Shorter + More detail + Original rewrite the finished answer and the
corrected version in place (FollowUpBar); every guide step has
Explain step N differently; failed calls say so without losing text.
Turn into a study deck
- Any result hands its content (answer, corrected version or assembled
guide) to New deck through a one-shot sessionStorage seed with the
options step pre-filled.
UI
- Phone step 2 gains the third card: tapping it reveals the solution
field + submit; desktop has three cards and the solution field shows
with the mode. Homework history labels and share/card previews cover
the new mode.
Verified: new phase-3 suite 20/20 against a stub Ollama (check payload,
missing-solution 400, refine validation, phone + desktop check flows,
shorter follow-up with restore, guide explain, deck hand-off) plus
regression 74/74, mobile 52/52, declutter 116/116, wizard 80/80,
animation 8/8, native 35/35, phase1 38/38, phase2 43/43, i18n 25/25.
Key parity 734/734. Lint + build clean. SW freezmo-v27.
Homework (/homework — entry points in the drawer and on the create screen)
- Assignment intake reuses the study pipeline: pasted text, photos
(camera/picker, compressed in the browser) and documents (PDF/DOCX/PPTX/
ODT/text) through lib/image + lib/documents, the same vision step and the
same language guard. Raw files never leave the device — only text and JPEG
data URLs reach the server.
- Two large choice cards: “Just make it” returns the finished homework
(answer only, plain text, copy + make it again); “Explain how to make this
homework” builds an interactive guide.
- The guide is one step per screen in the study-feed snap layout:
explanation, three progressive hints (nudge -> hint -> the answer) and an
optional checkpoint rendered by the existing activity components (graded
through /api/grade). The last screen assembles the solution from the step
answers, with a copy button. Checkpoints carry xp 0 so the guide stays free
of XP/streak chrome.
- Results live in freezmo.homework.v1 with a small history (open/delete);
“Make it again” re-runs from the stored assignment text (the vision
transcription when photos were used) — no original files are stored.
- The answer and the guide are written in the assignment's language, never
translated into the UI language.
Server
- POST /api/homework: NDJSON stage events like /api/generate, the same
http-guard, auth checked before parsing, and its own RATE_LIMIT_HOMEWORK
(documented in .env.example + the README table).
- New prompts in lib/prompts.ts and zod schemas in lib/schemas.ts with the
same single repair retry; TEXT_MODEL for content, VISION_MODEL only for
photos. New shared client pieces: useNotesIntake, lib/homework store,
lib/clipboard, and a shared NDJSON reader in lib/client-api.
Also finishing the mobile polish pass from the previous request
- Phone ice scene: the top-left orb moved behind the glass header (it sat
behind the page heading/subtitle). The settings sheet's scroll cue is now a
bare chevron inside a content fade (mask) instead of a chip over the status
banner, and the API key card's note/disclosure/actions are tightened.
- theme-color now follows the in-app theme (SSR metas still fall back to the
OS preference); apple-mobile-web-app-status-bar-style stays “default”
because iOS fixes the status-bar text colour at launch.
Verified: 78/78 homework checks (text/photo/PDF, both modes, EN+CS, dark+light,
360/390px, copy, history persistence, storage privacy, API language, axe), plus
the existing suites re-run (mobile 47/47, pixel contrast 0 below AA, i18n 27/27,
desktop regression 72/72). Lint + build clean; service worker freezmo-v6.
English stays the default; Czech is a full translation. Only the app chrome
is translated - notes and generated decks keep their own language, which is a
core feature, not a setting.
i18n
- lib/i18n-dict.ts: flat en/cs dictionaries (441 keys each), translate() and
translatePlural() with real Czech plural rules (1 sada / 2-4 sady / 5+ sad).
English is the default and the fallback for a missing key.
- lib/i18n.ts: browser store (freezmo.locale.v1), useI18n() hook, and the
x-freezmo-locale header. The hook's t/plural are bound to the locale React
is rendering with, so the first client render matches the server HTML and
then re-renders in the stored language; module-level t is only for event
handlers and non-React helpers.
- <html lang> follows the choice (inline script + setLocale).
UI
- Settings -> Language segmented control (English / Cestina), plus an EN/CS
button on the key gate for first run.
- Every screen translated: sidebar, header, tab bar, library, deck cards,
folder bar, create flow, all seven activities, study chrome, results,
settings + health panel, API-key gate and help, install sheet, offline and
the deck-not-found page.
- Plural-aware counts everywhere (decks, folders, activities, runs, photos,
documents, pages, words, wrong attempts).
Server
- API routes and guards (generate, grade, health, http-guard, rate-limit,
ollama error mapping) answer in the request's language, so error cards and
the health panel never fall back to English mid-sentence.
Fixes
- A hydration mismatch when a non-default locale was stored silently dropped
the .dark class after hydration; render-time t() is now bound to the
hook's locale (UploadScreen.describeExtraction, ResultsScreen.answerFor,
DeckCard's folder fallback).
- Bumped the service worker cache to freezmo-v4 for this deploy.
Tests: i18n 27/27 (gate toggle, persistence, html lang, Czech plurals,
settings switch) and a rebuilt regression suite 72/72 (English default,
library, create, all 7 activity types, results, settings, install sheet,
offline, PWA, security, Czech smoke).
Steps
- iOS now uses the exact flow: Safari users get 4 steps (⋯ → Share/More →
Add to Home Screen → Add); non-Safari (CriOS/FxiOS/EdgiOS/in-app) get a
first 'Open this page in Safari' step and 5 total. The vague 'in some
browsers it's under…' wording is gone.
Chips and icons
- The ⋯ chip renders a single three-dot icon (icon-only pill) instead of
the icon plus an '⋯' character (which showed two sets of dots).
- The More chip uses a chevron-down icon instead of the rounded-square.
- The accent pill uses real iOS blue: #0A84FF dark / #007AFF light, still
the only coloured element.
- Steps and chips share one 16px type scale; chip labels are medium weight.
Typography and layout
- Numbers are small muted circles in a fixed 24px column, aligned with the
first text line via matching line-height.
- Removed trailing periods after chips; steps tightened.
- Bottom padding is calc(24px + env(safe-area-inset-bottom)) and a
secondary 'Maybe later' button sits below the steps.
- max-height 90dvh with inner scrolling; 36×4px low-contrast handle with
real swipe-down-to-close (momentum off, spring back).
- App card shows the hostname only (no port), truncated.
Themes
- Dark (default): layered surfaces #1c1c1e sheet / #2c2c2e card / #3a3a3c
chips with 8% white borders; subtle logo tile instead of solid white;
backdrop is 50% black with 4px blur.
- Light only when the app theme was explicitly set to light, ignoring the
system preference; borders #e0e0e0, chip fills #f2f2f2.
Also fixes a crash the test pressure exposed: a 429 (rate-limited)
/api/health response crashed Settings at health.vision.available.
fetchHealth now rejects non-OK/malformed responses so the dialog shows its
error banner instead of the app error screen.
Install-sheet suite now 64 checks; pwa/final-ux/security/contrast green.
In user-key mode a browser with no stored Ollama key now gets a polished
gate instead of the app shell: one masked input with a show/hide toggle,
a single Continue button, inline validation errors, and a "?" help sheet
with the steps to a free key.
- Validates the draft key through /api/health (real 1-token probe) before
saving; invalid-key and network failures read differently.
- Key still lives only in this browser's localStorage; nothing server-side.
- Gate wraps /, /create and deck study; server-key and local-daemon modes
skip it; removing the key in Settings brings it back.
- Removes the now-redundant dashboard/create API-key banner.
- Monochrome corner glows (dark near-black / light near-white), 24px card,
48px field and button, gentle fade/slide-in, reduced-motion aware.
- Help modal: centred dialog on desktop, bottom sheet on mobile, closes via
X, Escape or backdrop.
- README and browser suites updated (34 gate checks, 15 key-flow checks,
12 shipped-default checks; ui/folders/docs/pwa/security/contrast green).
Next.js 15 app that turns photos, PDFs, DOCX/PPTX and text into a swipeable
feed of AI-generated activities (flashcards, multiple choice, true/false,
fill-in-the-blank, matching, ordering, short answer) using the user's own
Ollama models.
Highlights:
- All AI calls run in server route handlers; the browser never talks to Ollama.
- Bring-your-own-key mode (default): the key lives only in the browser and is
sent per-request; server, local-daemon and shared-key modes also supported.
- Client-side document extraction (pdf.js, mammoth, jszip) with scanned-PDF
page rendering for the vision model; notes stay in their original language.
- ChatGPT-style shell: collapsible sidebar / mobile drawer, neutral token-based
light + dark themes, folder organisation, PWA (manifest, SW, offline page).
- Strict JSON validation with zod + auto-repair, rate limiting, and request
guards (cross-origin, body-size, API-key sanitising).
- setup.sh bootstraps deps, .env.local and a production build.
Gitignore excludes .env*, the generated pdf.js worker and local scratch files.
