Full Summaries for Long Meetings — and a Retry When One Fails
Longer recordings now summarize the whole conversation instead of just the opening minutes, and a transcription that fails is no longer a dead end — one click retries it.
Long meetings summarize in full — recordings past about ten minutes now transcribe and summarize the entire session, not just the first stretch, so your summary and notes cover the whole conversation
A failed transcription is no longer a silently blank note — the recording shows a Retry button so you can re-run it in one click
One rough patch of audio won't sink the whole recording — a bad stretch is skipped and the rest still transcribes
One Question, One Answer — Sharper Live Interviews
A tune-up of the live answer engine driven by a real interview session: no more duplicate answers, no more answering half a question, better recognition of technical terms, and meeting notes that keep every code snippet.
No duplicate answers — when the same question settles twice or gets restated, Cloak answers it once and stays quiet the second time
No more answering fragments — vague check-ins like “Sorry, could you?” or “How can we do this?” wait for the full question instead of triggering a stale answer
Follow-ups aren't lost — a new question asked right on the heels of an answered one is picked out and answered on its own
Technical terms transcribe correctly — live transcription is tuned toward interview vocabulary, and you can add your own terms for a role or company
⌘⇧A knows when it's looking at your meeting window — instead of answering from the call controls, it reads your actual screen
Answers remember the ground rules — constraints and definitions stated earlier in the session ("only use the data we gave you") now carry into every later answer instead of being forgotten
Design questions get a design answer — an open "how would you approach this" no longer gets forced into a rigid code-drill template just because the conversation involved code
Cloak answers the interviewer, not you — when it can hear both sides, your own replies no longer trigger answers, and your notes list real people instead of a jumble of "Speaker 1, Speaker 2, Speaker 3"
Meeting notes keep your code — post-session notes no longer drop code blocks and tables from the Q&A section
Long answers open with a line you can say out loud while you scan the rest
Follow-up questions remember what was already suggested — a screen capture now keeps its own first answer on hand, plus what you said afterward, so a later "have you run this?" is checked against reality instead of guessed fresh
Two-way listening is on by default — Cloak separates you from the other person out of the box, no setup needed
Friendlier short-session notes and sharper coding answers
Short or empty recordings now get a clear welcome note instead of a broken-looking summary, coding answers follow a progressive teaching walkthrough in Auto mode, and queued Code Hint screenshots keep moving.
Short recordings get a friendly onboarding note instead of “Summary unavailable” or a hallucinated one-word summary
Longer recordings stay on the normal summary path, even if an AI summary pass fails
Auto-mode coding answers use the progressive NeetCode-style walkthrough (understand → brute force → insight → optimal → code)
Queued screenshots continue into the Code Hint vision path instead of getting stuck on the capture placeholder
Windows installers build cleanly again for download builds
Scroll, Queue, Answer — Plus Interview Mode for Screen Problems
Long LeetCode-style problems no longer need one cramped screenshot. Queue each scrolled section, answer once, and get a spoken teaching walkthrough when Interview mode is on.
Screenshot queue — press ⌘⇧H on each scrolled section of a problem, then ⌘⇧↵ to answer from the whole thing as one question
Interview mode in overlay settings — turn it on and screen captures use a progressive teaching walkthrough (understand → brute force → insight → optimal → walkthrough → code) instead of a generic summary
Clearer code in answers — syntax highlighting in the overlay makes functions, strings, and keywords easier to scan mid-interview
Smarter auto-answer pickup — directive-style asks (“design a class that…”) are detected more reliably so hands-free mode fires on the full spec, not a fragment
Gateway answers work on download builds — connecting your Cloak license now actually routes AI through the gateway in the desktop app you install from the site, not just in dev builds
The Desktop App Opens, and Permissions Actually Prompt
A stability release for macOS: the app launches reliably, the microphone, screen, and automation prompts appear where you'd expect, and image-heavy notes render again.
Reliable launch — fixed startup failures (including a crash and a code-signing issue) that could stop the app from opening
Permissions prompt properly — microphone, screen, and automation access now prompt so recording and one-key capture work
One-key capture finds your app again — ⌘⇧A reliably detects the app you're looking at
Images work again — restored the image engine some notes and previews rely on
Clearer offline-transcription guidance — Intel Macs get a plain explanation and a nudge to a cloud option instead of a confusing error
On-Device Transcription That Just Installs, and One-Tap Permission Repair
A reliability release: offline transcription now sets up cleanly on a fresh install, permissions are one tap to fix when macOS gets confused, and Cloak tells you the moment it stops hearing your meeting.
Offline transcription installs on the first try — the on-device speech models (Whisper, Moonshine) no longer fail to load after downloading, so fully offline mode works out of the box
One-tap Repair Permissions — if microphone, screen, or automation access gets stuck, a single button resets it so you can re-grant and get back to capturing audio
Know the instant audio drops — a clear banner appears in the overlay if system audio stops mid-meeting, instead of leaving you with a silently frozen transcript
No more first-launch hiccup — capturing the active app with ⌘⇧A works right after install instead of erroring on the very first try
Sharper meeting follow-ups — post-meeting notes now come with ready-to-send follow-up drafts tailored to what was actually discussed
Stop babysitting the mode chips. Cloak now reads each question and answers in the shape that fits it — you only pick a mode when you want to override the choice.
Auto mode, on by default — Cloak chooses the answer style for each question instead of locking onto the last chip you tapped, so a coding question and a comp conversation no longer come back in the same shape
One tap to take over — tap any mode chip to pin it for the moment, and tap Auto to hand the wheel back
Screen grabs read cleanly — capturing your screen is no longer forced through a mode you picked earlier, so the answer fits what's actually on screen
Sharper Screen Answers, a Live Indicator, and Offline Mode
Answers from your screen now come back in full interview shape, you can see at a glance when Cloak is listening, permissions stop sending you into System Settings, and you can run everything fully offline.
Screen answers shaped for the round — grab a coding or SQL problem straight off your screen and get the full structured answer (approach, code, dry run, complexity, follow-ups) instead of a plain reply
Attach screenshots to a question — the paperclip grabs one or more screenshots and asks about them, and sending no longer hangs
A live recording indicator — animated bars by the stop button show at a glance when Cloak is listening
Grant access without the detour — the microphone and system-audio prompts now appear right where you click instead of dropping you in System Settings
Run fully offline — opt into on-device transcription and a local AI model with a one-tap setup that downloads in the background, lets you keep working, and skips anything you already have
Your Live Interview Copilot: Capture, Solve, Visualize
A big upgrade for technical interviews. Grab the problem off your screen with one key, get an answer shaped for the round you're actually in — data, coding, or system design — and step through the tricky parts with an animated walkthrough.
One-key capture — press ⌘⇧A and Cloak reads the whole problem on screen (including what's below the fold), figures out whether it's a data, coding, or system-design question, and answers it
Data-interview mode — pandas and SQL answers that show your workflow (assumptions, data profiling, plan, validation, and a plain-English business takeaway), not just a bare query
Cleaner coding answers — a short menu of approaches with the best one expanded (idea, gotchas, code, complexity); type 1, 2, or all to open the rest; every meaningful line is commented so you can explain it out loud, and the code runs end-to-end
Animated walkthroughs — hit Visualize to step through arrays, pointers, hash maps, and stacks with play and speed controls
System-design answers that hit the rubric — capacity estimates, an algorithm comparison table, data model, architecture, scaling and consistency, and honest tradeoffs, with an architecture diagram drawn for you
Pick your language once — set your preferred coding language (Python by default) and Cloak sticks to it even when the site shows another
Easier to read — rendered math and tables, collapsible chips and ask box to free up space, and a dismissable microphone prompt
A Cleaner Overlay — and Bring Your Own AI Endpoint
The invisible overlay gets a tighter, better-behaved redesign, and you can now point Cloak at any OpenAI-compatible endpoint or LiteLLM proxy with a single key.
Redesigned overlay pill — a compact branded pill sits centered above the panel; click the logo to jump straight to your workspace
Minimize that actually works — the center chip collapses and expands the panel both ways, with the right label and chevron every time
Menus land where you look — the model and settings menus now open anchored to their own buttons instead of in a far corner
The panel hugs its content — it sizes to what's actually in it instead of leaving a tall empty box, and the bottom bar no longer gets clipped
Clearer mode chips — every mode now carries its own icon
Connect any AI endpoint — add a LiteLLM or OpenAI-compatible gateway in Settings with one base URL and key, and every model behind it becomes available; test the connection right there
Prep Pack: Prep From the Company's Own Interview Guide
Attach the hiring pack, interview guide, or take-home a company actually sent you, and Cloak preps you on their real process — the exact rounds and values — instead of a generic guess from the job post.
Attach anything authoritative — drop in a PDF or Word doc, or paste the text of a hiring pack, interview guide, recruiter email, or take-home brief onto any target job
Prep built around the real interview loop — Cloak reads the actual rounds from the document and organizes your practice questions stage by stage, so you prepare for the interviews you'll really face
The document wins — the interview process and company values stated in your pack override anything Cloak inferred from the web
Grounds your live copilot too — the same context feeds your answers during the real interview, not just your prep
Attach or paste in seconds, and swap it out anytime
Demo Coach: Invisible Teleprompter for Live Product Demos
Turn the handoff your Codex session writes into an on-screen teleprompter and live Q&A copilot — visible only to you, invisible to every screen-share and recording tool.
One-click handoff — paste the copyable prompt from Settings into Codex, and it writes a structured HANDOFF.md grounded on the real diff: demo script, talking points, likely questions, and gotchas
Import to workspace — drop the file into the Demo Coach panel and Cloak creates a tagged, editable doc in your workspace
Present in overlay — clicking the button opens the invisible overlay with the full HANDOFF rendered as a formatted teleprompter card
Live Q&A from the handoff — the HANDOFF content loads into its own AI grounding slot, so Cloak answers audience questions directly from the doc without touching your résumé or career context
Demo chip — a dedicated mode in the overlay reframes every answer as a tight spoken response drawn from the script and talking points
Step-through hotkeys — jump between sections of the demo script with ⌘⇧. (next) and ⌘⇧, (previous) while you talk
Reliability fix — resolved a live session bug where a slow gateway response could lock Cloak out of AI answers for up to 30 seconds; answers now retry correctly every time
Cloak now reads more of the live screen when it matters, keeps coding-assessment context tighter, and shows a cleaner model list backed by routes we can verify end to end.
Screen-aware answers — capture the full screen or a selected region and let Cloak use the visible problem, code, error, or diagram as context
Better coding assessment extraction — Cloak preserves visible prompts, examples, constraints, starter code, and errors before shaping the answer
Vision modes for live interviews — choose vision-first, vision-only, or private vision behavior depending on how much screen context you want included
Sharper model routing — the model picker now hides unverified routes and keeps advertised gateway models tied to the upstream provider that actually answered
Bring your own Codex — answer through OpenAI's Codex by signing in with your ChatGPT subscription or pasting an API key, no extra setup
More reliable live testing — Revolut and GetYourGuide-style smoke prompts now exercise transcript, answer, deadline, and model-routing behavior together
Cloak now notices more of the moments where you need help mid-conversation and turns them into visible, one-tap actions instead of making you hunt through modes.
Dynamic action cards — Cloak surfaces recap, brainstorm, clarify, and negotiation coaching suggestions with the transcript signal that triggered them
One-key accept — press Tab or click the card to run the suggested action, or dismiss it when it is not useful
Visible timing signals — Cloak shows when an answer is waiting, falling back, or streaming, and turns phrases like "you have 30 minutes" into a live timer
Memory across restarts — recent live Q&A context now survives relaunches, so follow-up questions can refer back to prior answers
Richer answer control — Assist, Shorten, bullets, detailed, and code-only behavior are now first-class in the live assistant
Career prep now works in stages instead of dumping every research finding at once. Cloak grounds the workspace in the interview loop it finds from the job description and company research, then highlights the one stage and practice question that matter next.
Stage-first prep — see the current interview stage, how far through the loop you are, and what comes later
Focused practice — Cloak highlights one question at a time and tucks extra same-stage questions away until you need them
Cleaner company research — full dossiers, prep artifacts, salary notes, culture signals, and sources now sit behind expandable sections instead of crowding the page
Better question routing — architecture prompts like event streaming and provider fallback land in system design, while generic labels such as technical or hard no longer appear as answer signals
Local Skills, Faster Local Speech, and Smarter Fallbacks
Cloak can now run local SKILL.md instructions from the overlay, transcribe locally with lower latency, and recover more gracefully when an AI provider is slow, stuck, or out of quota.
Run local skills from the overlay — invoke built-in or folder-based SKILL.md instructions with /skill or $skill
Faster private transcription — local speech-to-text now chunks audio as you speak, with separate model choices for mic and system audio
Cleaner local transcripts — Cloak trims tiny trailing fragments like "Tha" before they trigger an answer
Smarter AI fallback — Cloak detects rate limits, quota exhaustion, auth failures, stalls, and empty streams, then falls through before anything is shown
Clearer provider logs — answer logs now show the actual provider and resolved model, so a Gemini answer no longer looks like it used the requested fast Groq model
Press ⌘K and type anything to search your meeting notes, docs, and prep material with AI. Cloak finds the most relevant pages, reads them, and answers with source links you can click to open. Plus: meetings now group by Today and Yesterday, your next few events show up in one glanceable card, and you can pull local AI models without leaving Cloak.
Ask your workspace with ⌘K — type a question and Cloak finds the relevant docs, reads them, and answers with clickable source citations
Meeting sessions grouped by Today, Yesterday, and date — so your history is easy to scan at a glance
Up to three upcoming events in one banner — see what's next with a Join link and a time chip (Now / in N min / Today 3:00 PM)
Pull local AI models right from Cloak's settings — no terminal needed, watch the progress bar fill as the model downloads
Live transcription used to go quiet — and quietly lose words — when the speech service hiccupped mid-meeting. Now Cloak rides out those blips and keeps your transcript whole.
No more missing minutes — if the speech connection drops for a moment, Cloak buffers your audio and replays it on reconnect instead of losing it
Rides out connection blips — live transcription reconnects smoothly after a hiccup rather than going silent for the rest of the call
Knows your limits up front — Cloak checks your transcription budget before a session starts and tells you, instead of cutting out partway through
Meeting Notes With Summary, Transcript & Usage Tabs
Recorded sessions now open with Summary, Transcript, and Usage tabs — jump straight to the section you need. Getting Started walks you through Cloak's shortcuts, modes, and setup.
Summary / Transcript / Usage tabs — meeting notes get header tabs that jump you straight to each section
Rich Getting Started guide — shortcuts, mode chips, API setup, workspace basics, and demo transcript + Q&A sections
Template gallery — browse and import local templates from the sidebar without AFFiNE branding
Cleaner live transcript — overlay shows system audio only; cross-stream echo is deduplicated in meeting notes
Code Hint can now read a region you draw instead of your whole screen — frame the exact problem, diagram, or snippet you want help with and leave the rest out.
Drag to crop — on desktop, trigger Code Hint and drag a box around the part of your screen you want read
Works in the browser too — share a tab, hit Crop, and select the region right on the captured frame
Sharper answers — sending only the part you care about cuts the noise, so the response stays on the question you actually asked
Cloak now answers the question being asked right now — on its own, the moment it's asked — instead of circling back to whatever came up earlier.
Hands-free answers — as soon as a question lands, Cloak answers it automatically, no key press needed
On-topic, every time — answers track the current question, so a behavioral question gets a behavioral answer and a coding question gets code
Clearer overlay mic — your own voice comes through reliably from the overlay for the two-sided transcript
Custom context that actually lands — what you put in Settings → Custom context now grounds every answer, so you can paste in your own prep material and Cloak goes deep on it
Recordings keep their smart name — the auto-generated meeting title now sticks when you open the note instead of reverting to a generic one
Transcription that recovers itself — if the speech connection drops mid-meeting it now reconnects automatically instead of going quiet for the rest of the call
Answers no longer collapse — what streams in is what stays; answers used to shrink to a short version the moment they finished, and now you keep the full response
Auto or manual answers — choose hands-free auto-answers or manual mode, where Cloak answers only when you tap a mode or hit a shortcut. Auto now waits for you to finish the question before answering, so it no longer jumps in on half a sentence
Whole-question answers — when a question gets split across a pause (“What is X? And why does it matter?”), Cloak now answers the complete question instead of just the trailing part
Answers shaped to the question — Cloak now recognizes what kind of question it is (design, coding, behavioral, or a plain explanation) and structures the response to match, on any topic, instead of always returning a flat list
Cloak now separates who said what in the live transcript and lets you choose the speech-to-text engine behind it — including managed Google, with no key of your own to set up.
Two-sided transcript — your voice and the other person are labeled You / Them as they speak, so it's clear who said what (best on a headset)
Managed Google speech-to-text — pick Google as your engine and it runs through your Cloak plan, no separate API key required
Cleaner separation — echo handling keeps your mic and the call audio from bleeding into each other's lines
A few small moves that make the live overlay faster to drive: copy any code answer in one tap, launch a mode by typing a slash, and attach a screenshot to your next question.
Copy button on every code block — grab a snippet straight from an answer, no selecting
Type "/" in the ask box to pull up your answer modes and run one without reaching for the mouse
Attach screenshots — capture what's on screen, queue a few, and send them with your next question for a vision answer
Streaming code holds its shape instead of jumping around as the answer fills in
Cloak now gives new desktop users a short, keyless trial and makes plan limits easier to understand across web and desktop, with upgrade prompts that stay clear without getting in your way.
Start a desktop trial in one click — no API key needed just to see Cloak answer in the overlay
Quota banners now show when free sessions are running low or used up, so you know what changed before a session starts
Premium prompts explain what upgrading unlocks, including longer access, better limits, and account-backed AI
Support nudges are dismissible and restrained, matching Cloak's quieter product style
The gateway now backs the trial flow with signed tokens and server-side metering, so the app and API stay in sync
Coding Rounds: Reason Out Loud, Then Walk the Code
Cloak now answers coding questions the way a strong candidate thinks — building from the obvious brute force up to the optimal solution — and can walk you through the code step by step, out loud, with the data structures drawn as they change.
Progressive solutions — answers climb from brute force to optimal, each with its complexity, so you can narrate how you reasoned to the answer instead of just reciting it
Common pitfalls — every answer ends with the mistakes and edge cases that trip people up on that exact problem
Live walkthrough — one shortcut turns the solution into a step-by-step trace; each step is a plain sentence you can say while pointing at the code
See it change — the array, its pointers, and the hash map are drawn beside each step and update as you move through it
Step at your own pace — move forward and back through the trace while you explain
Run speech-to-text entirely on your machine — no API key, nothing leaving your device. Pick a model, download it once, and Cloak transcribes locally, now faster and with a new lightweight option.
Moonshine models — a fast, compact choice for English or multilingual speech, alongside the Whisper models
Fully offline and private — download once, then everything runs on-device with no key and no upload
Quicker and more reliable — local transcription is faster and keeps up with what it captures
Code Hint Debugs With You — and Answers Sound Human
Code Hint now goes a step further: run its suggestion, and if you hit an error, one shortcut re-reads your screen and fixes it. Spoken answers also got cleaner — sharper, on-topic, and free of AI-template filler.
Debug follow-up — after a Code Hint, press its debug shortcut and Cloak reads the error or failing output on your screen and gives the root cause plus the corrected code
On-topic answers — coding questions stay about the code instead of pulling in your résumé or the job description, so the response is tighter
Sounds like you, not a bot — spoken answers drop corporate filler and clichés and stay short enough to actually say out loud
Cloak now senses what kind of question you're facing and shapes the answer to match — code-first for a coding round, a tight STAR story for a behavioral one, anchor-smart tactics the moment talk turns to comp. No mode-switching required.
Coding questions lead with the code and spell out time and space complexity
System-design questions come back structured — requirements, design, trade-offs, named tech
Behavioral questions get a confident, first-person STAR answer you can say out loud
Salary and offer talk gets negotiation-savvy phrasing that never anchors against you
Always on, with no added wait — and a small badge shows the read, so you're never surprised
Stuck on code or an error on screen? Press the Code Hint shortcut and Cloak looks at what's actually in front of you and answers — the fix as a ready-to-paste snippet, plus a quick why.
One shortcut captures your screen and reads the code or error on it
Get the fix as a fenced code block, with a couple of lines on what changed and why
If it's an error or stack trace, Cloak names the root cause first
Runs through your Cloak plan — nothing extra to set up
Turn a page into a social post without leaving Cloak. Draft from your notes, choose your connected accounts, and schedule it — or post right away. Experimental, on desktop.
Compose from any page — Cloak drops the page's text into an editable composer, ready to post
Post across platforms — pick your connected X, LinkedIn, Instagram and more in a single go
Schedule or post now — with a clear confirmation of exactly which accounts and when, before anything goes out
Real status, not guesswork — per-platform results come back in-app, so a failed post never hides behind a false "posted"
Opt-in and experimental — connect it with your own scheduling key in settings; it stays off until you do
Your answers now ride a resilient, multi-provider path. If the model behind an answer is rate-limited, region-blocked, or down, Cloak fails over to a healthy one mid-request — so you get a real answer instead of silence.
Automatic cross-provider failover — if one AI provider stumbles, Cloak instantly retries on another, behind the scenes
Load spread across multiple keys per provider, so a single rate limit doesn't stall your answers
It just works — the handoff happens mid-request, with no empty replies and nothing for you to configure
Your live transcript and meeting notes now read the way you'd write them. Cloak rewrites spoken numbers, currency, and units into clean written form on the fly — so "five point two million dollars" becomes $5.2M and "thirty kilometers per hour" becomes 30kph, right in the overlay and your saved notes.
Money reads like money — "a hundred and forty thousand dollars" becomes $140k, and "$5.2 million" tightens to $5.2M
Units and percentages clean up too — 30kph, 60mph, 15%
Conservative by design — it only rewrites what it's sure about, so plain speech, times, and years are left exactly as spoken
Works everywhere your transcript appears: the live overlay strip, the answer prompt, and your post-meeting notes
Bring your Cloak account to the Mac and Windows apps. Sign in with Google or a magic link to unlock account features and carry your plan, entitlements, and AI access across web and desktop — one account everywhere.
Sign in on the desktop app with Google or a one-tap magic link
One Cloak account spans web and desktop — your plan and entitlements follow you
Account-backed AI runs through the Cloak gateway, so your model access and quota stay in sync
Still local-first — signing in is optional and stays out of your way until you want it
Bring an OpenRouter key and Cloak now squeezes more reliability, speed, and privacy out of it — answers fall back to a backup model the moment one stumbles, your context is cached between turns, and you can route for throughput, cost, or zero data retention.
Automatic model fallback — list your preferred models and Cloak instantly retries the next one if a provider fails, so an answer rarely drops
Faster, cheaper follow-ups — your résumé, company dossier, and mode context are cached between turns instead of re-sent every time
Built-in web grounding — opt into live web search on any model, complementing Exa
Tune the trade-off — route for fastest throughput (ideal for the live copilot) or the lowest price
Privacy routing — restrict answers to zero-data-retention providers only
Cost visibility — exact per-answer cost across whichever model handled the request
Bring Your Own Everything: Models, Endpoints & Integrations
Plug in any model or service. Four new AI providers, a custom-endpoint mode for any REST API, multi-key failover, calendar + issue-tracker integrations, a live negotiation coach, and faster fully-offline transcription.
Four new AI providers — xAI Grok, Mistral, Perplexity, and Cohere — alongside the existing lineup
Custom endpoint — point Cloak at any AI or speech-to-text API with your own request template ({{TEXT}}, {{AUDIO}}, …)
Multi-key API pools — add several keys per provider for automatic round-robin and failover when one is rate-limited
Task export — push a meeting's action items straight to Jira, Linear, Asana, or ClickUp
Outlook / Microsoft 365 calendar — connect alongside Google for upcoming-meeting context
Negotiate mode — a live coach for salary and offer conversations, grounded in your company research
Faster on-device transcription with new Moonshine models (GPU-accelerated) and a fully-local, offline embeddings option for private document search
Pair your phone as a wireless control panel and AI display — on the same Wi-Fi or anywhere through the cloud relay. Plus a new stealth toolkit: disguise the app as a system tool and let clicks pass straight through the overlay.
Phone Mirror — switch modes, type questions, and read live AI answers from your phone (no audio leaves your Mac)
Two pairing modes: LAN over QR on the same Wi-Fi, or Anywhere via the cloud relay for any network
Process Disguise — masquerade as Terminal, System Settings, or Activity Monitor in the Dock and process list while undetectable
Mouse passthrough (⌘⇧B) — keep the overlay visible while clicks pass through to the app beneath it
Reliability & security hardening across audio capture and the gateway
Try Cloak instantly with a no-signup trial, see who said what with speaker identification, and turn any answer into a diagram — with one-click updates so you're always current.
10-minute free trial — no account, no API key, abuse-protected
Speaker identification (diarization) — transcripts labelled by who's talking
AI diagram generation — flowcharts and architecture diagrams rendered inline in answers
One-click in-app updates — download and install without leaving the app
Profile Intelligence v2 — sharper, evidence-checked grounding with live interview deadlines
Full career engine now gate-able via Experimental Features. Bring your own Exa key for real-time company dossiers, JD gap analysis, and a tailored résumé export.
Career Workspace behind enable_cloak_career flag (Settings → Experimental)
Exa Dossier: company intel, culture, salary bands via your Exa API key
JD gap analysis — paste any job description, surfaces what to highlight
One-click .docx résumé tailored to the active job description
Seven expert personas that hard-override the AI identity, not just tweak a prompt. Switch mid-call between Technical Interview, Sales, Recruiting, Lecture, and more.
The AI settings panel now shows provider cards with SAVED/RECOMMENDED/ACTIVE badges. Gateway mode (cloakapp-api) is distinguished from direct BYOK keys.
Provider cards in Audio + AI settings with status badges
Gateway vs BYOK distinction — gateway fronts all providers with quota management
Multi-provider fallback order surfaced in UI
Groq, Gemini, OpenAI, Anthropic, DeepSeek, Ollama all configurable
Answers now render rich markdown — code blocks, inline code, bold, lists. Seven answer-mode chips (Answer, Clarify, Recap, Code Hint…) let you steer the AI without stopping capture.
Full markdown rendering with syntax-highlighted code blocks
System audio capture lands. Google streaming STT gives a rolling live transcript. AI answers from up to 6 providers with automatic fallback on failure.