v1.1.7
August 22, 2026FEATURERapid screenshots batch into one turn with one answer.
New
- Take 2, 3, 5 screenshots in quick succession — each user bubble appears immediately in order, and the AI now returns ONE reply that addresses each in sequence ("Screenshot 1 shows…, Screenshot 2 shows…"). Previously every screenshot became its own turn with its own answer, which was noisy for multi-part problems.
- How it works: captures within 800 ms of each other are collected into one batch. First quiet moment fires the batch. Or hit Enter with typed text at any time to send the batch immediately with your own prompt attached.
- Cancel or New Chat now also clears anything sitting in the batch accumulator — you said stop.
Rapid-fire screenshots now show every user bubble in the chat immediately.
Fixes
- When you take screenshot 2, 3, 4, 5 while screenshot 1 is still being answered, all your user bubbles now appear in order right away — not just when the assistant catches up. The answers still stream in one at a time, but you can see everything you've submitted at a glance.
Rapid-fire screenshots now queue instead of being dropped.
Fixes
- Sending a second screenshot while the first response was still streaming used to silently do nothing. Screenshot 2 now queues automatically and fires as its own turn once screenshot 1's answer finishes — same for any subsequent captures.
- Clicking Cancel or starting a New Chat clears anything you had queued.
Screenshots no longer 401 after the app has been open for a while.
Fixes
- Sending a screenshot after the app had been open for ~15 minutes returned "unauthorized" — the access token silently expired and the streaming request path didn't refresh it. It now transparently refreshes and retries, matching every other API call.
Same-day patch for premium-tier users on the new GPT-5.6 models.
Fixes
- Premium tier (own API key) now works on GPT-5.6 Terra and GPT-5.6 Sol — a stale reasoning parameter from the old GPT-5 API was causing every request to fail with a 400 error. Standard tier was unaffected (server hotfixed).
v1.1.2
August 22, 2026FEATUREModel refresh: everyone now runs on the newest generation from both providers.
New models
- Standard tier: GPT-5.6 Terra (OpenAI) and Claude Sonnet 5 (Anthropic).
- Premium tier: GPT-5.6 Sol (OpenAI) and Claude Opus 5 (Anthropic).
- All four support 1M-token context — long transcripts, long code files, big screenshots all fit comfortably.
- Sonnet 5 is priced lower than the outgoing Sonnet 4.6 ($2/$10 vs $3/$15 per million tokens), and Opus 5 leads in coding benchmarks.
Rolling out
- Server-side registry is live now — desktop app picks up the new list automatically on next launch.
- Existing sessions with an old model selected: pick a new one from Settings → AI Model, or restart the app to auto-select the new default.
Auto-scroll, paywall ergonomics, and clearer login errors.
Fixes
- Chat now auto-scrolls smoothly while the response streams in — no more pinning yourself to the bottom of the answer manually.
- When the paywall is up, you can now minimize or hide the app from its own title bar — handy for stepping over to the browser to finish Stripe checkout.
- Wrong password / wrong email at sign-in now shows the actual reason instead of a confusing "Session expired" message.
- Account page subscription date is now consistent with the desktop sidebar wording ("Ends on May 15, 2026").
New models, real-time speaker mode, smoother streaming, and stronger stealth defaults — the biggest update since launch.
New models
- GPT-5 is the new standard OpenAI model — faster, sharper than the previous default.
- GPT-5.4 added in the premium tier for the absolute hardest problems.
- Claude Sonnet 4.6 (standard) and Claude Opus 4.7 (premium) unchanged.
- GPT-5 responses tuned for speed — first token in 1-2 seconds.
Real-time speaker mode
- Voice to text in real time — words appear as they're spoken, no latency.
- When you toggle off, the transcript is sent to AI instantly with no extra round-trip.
Stealth & privacy
- App is now invisible to screen-share at all times, regardless of opacity. Open Settings, change opacity, or use the app freely during a screen-share — your interviewer sees nothing.
- App stays hidden from the Windows taskbar even after restoring or signing in.
- Default opacity is now 60%. The transparency hotkey toggles between 60% and 100%.
- Click-through transparency: when stealth mode is engaged, mouse events pass through to the IDE behind so you can read answers without losing focus.
Better answers
- Every model now produces rich markdown by default — bold inline labels, headings, tables, dividers, blockquotes for "things to mention".
- System Design Mode produces cleaner Mermaid diagrams (architecture flowchart, sequence, and data model ER).
- erDiagram parser hardened — automatically fixes common issues like type-length annotations (varchar(255)) that used to break rendering.
Smoother UI
- Streaming responses now refresh at 60fps instead of token-by-token — text flows smoothly instead of popping line by line.
- Window has minimize / maximize / hide buttons in the title bar.
- Window movement (Ctrl + arrow keys) is clamped to your screen — can no longer drift off-display.
v1.0.0
April 26, 2026LAUNCHPublic launch. Interview Helpers is now live for Windows — the stealth desktop overlay built specifically for tech interviews.
Flagship features
- System Design Mode — press Ctrl+Shift+S, then send a screenshot of an architecture question. The AI answers with rendered Mermaid diagrams (flowchart, sequence, ER) inline. No competitor draws diagrams in the answer.
- Speaker Mode (Pro) — Ctrl+Shift+A toggles real-time transcription of your interviewer's voice through the system speaker. Live captions appear above the input. Toggle off and the AI answers in one tight sentence — perfect for verbal warm-ups and behavioral rounds.
- Screenshot AI — Ctrl+1 captures your active screen and sends it instantly. The AI sees the LeetCode problem, the design doc, the code, and walks through the answer.
Stealth & ergonomics
- Always-on-top transparent overlay, hidden from the Windows taskbar so it doesn't show in screen-shares.
- Ctrl+G toggles transparency AND click-through — read the answer while your clicks pass through to the IDE underneath.
- Ctrl+Shift+U panic-button to instantly hide the overlay if your interviewer asks for full-screen share.
- Ctrl+Arrow keys move the window 50 px in any direction; clamped to the screen so it can't go off-display.
- Custom hotkeys — every action is rebindable in Settings → General → Hotkeys.
AI models
- Standard tier: GPT-5 and Claude Sonnet 4.6.
- Premium tier: GPT-5.4 and Claude Opus 4.7 — for the hardest problems.
- GPT-5 tuned for speed — first token in 1-2 seconds.
- Real-time voice-to-text powers the live speaker transcription with zero perceived latency.
Pricing & trial
- Monthly $39 or annual $375 ($31.25/mo equivalent — save $93/year).
- 10 free messages on signup, no credit card.
- Subscription managed via Stripe. Cancel any time from your account or the Stripe billing portal.
Privacy & security
- Screenshots and audio are not logged by default — sent to the AI provider, then discarded.
- Stealth overlay hidden from the Windows taskbar.
- Per-user daily / weekly / monthly token caps and abuse rate limits to keep your trial-pool fair.