- More reliable builds — the agent smoke-tests your app before handing it off and avoids the dependency pitfalls that caused crash loops. Apps using native packages build correctly now too. - No more long stalls on big requests — large requests now ship a working first version each turn instead of timing out. - Deep Scan for ideas — new toggle for a more thorough analysis when you're exploring an app idea. - Smarter notification prompt — we ask for notification permission during your first build, instead of at a random moment. - Simpler credit top-ups — clean whole-dollar amounts with more options to choose from. - Cleaner first reply — fixed a stray "pong" that occasionally showed up as the agent's first message.

• New "Auto" mode automatically picks the best AI model for each task • Redesigned Add-ons: manage Always-On and memory Boost per app • Smoother deploys — live progress, resume in-flight, one-tap redeploy • Live token counter and a "See it live" shortcut while you build • Links and emails now open straight into the app • Safer account deletion for subscribers, plus reliability fixes and polish

- New: Deploy button — chat header now has a play icon that opens a diff drawer showing every file changed since your last deploy. Deploy when you want to deploy. No more auto-shipping every turn. - New: in-app subscriptions — Builder, Pro, Always-On, Database, and the new App Boost add-on (0.5 vCPU + 1 GB per app) all purchasable from the app. - New: Invite Friends — share your code and you both get $3 in AI credits when they complete their first task. Tap "Invite Friends" in the menu drawer. - New: onboarding wizard — a 2-question setup creates your first app or starts a chat, so you can skip the empty-state. - App templates now ship real backends — saas, dashboard, ecommerce, blog, form, CRUD API, and portfolio templates were rebuilt as multi-file Node scaffolds with working APIs, persistence, and admin views — not toy HTML. - Faster new-app creation — provisioning dropped from ~78s to ~15s. - Faster AI chat — fixed routing bugs that were adding 30–90s of latency on some AI models. - Free Qwen tier — users without their own API key now default to Qwen Flash for fast, no-setup chat. - Voice input fixes — typed text no longer gets overwritten by dictation, mic-permission errors surface inline, recordings cap at 5 minutes. - Emoji picker in both chat composers. - Redesigned status pages with an onboarding-style hero and iOS-aware "Back to Dashboard" buttons that open the iOS app, not Safari.

- Manage your app's environment variables right from the app menu — add, reveal, paste a .env, and redeploy in one tap. - New accounts now start on DeepSeek v3.1, so your starter credit goes much further. - Plus: long agent responses no longer truncate, chat survives backgrounding, voice input fixes, and small polish across the app.

Chat history renders properly on reload, agents stop using the GitHub repo name as their identity, and free-tier loading messaging matches reality. Various deploy and domain fixes under the hood.

- Faster chat — first response in ~5–10s, down from ~30s+ before - Branch labels on Apps and Remote tiles - New filter drawer on Repos tab with search + VibeKit-hosted-only toggle - AI provider picker now includes Free ($0, no setup) alongside Anthropic, OpenAI, DeepSeek - Edit & Resend asks before deleting follow-up replies

Faster, more polished first-deploy experience with a celebratory animation when your app is ready. Smoother chat — quieter background polling, cleaner alerts, better handling of long-running tasks. Fixes for plan limit messaging, cold-start state, and several reliability + billing edge cases.

- Remote chat history no longer disappears when you reopen the app and tap back into a tile - Brief network drops mid-message no longer leave a fake "connection lost" bubble next to your real answer - New workspaces now show a clear "Setting up your agent…" state instead of letting you type before it's ready - Tap "+" on the Remote tab to pair a new machine - Cleaner empty state on a fresh Remote chat (just the boot animation + greeting) - Fixed: new apps were sometimes routing through Claude even for OpenAI-only accounts — now matches your connected provider

- Custom app icons — pick any SF Symbol or keep the auto pixel art - Pick any AI model without silent failures — it tests the model before saving - Faster chats — long sessions stay snappy - Remote tab shows your paired-machine count at the top

- Multi-agent Remote tab — each machine is its own tile with name, custom icon, and live metadata - Independent chat per machine — no more cross-tile transcript bleed - Rename your machines; push notifications use the name you set - Smarter model picker — only valid models, your pick sticks across cold launches - Unified Health drawer with past-hour charts - Face ID required on cold launch - Battery / thermal fix — animation + timer leaks no longer warm the phone

- Remote tab now explains what it does before asking for a link code - Hosted repos are marked in the Repos tab so you can spot them at a glance - AI Providers split into "Bring Your Own" and "Hosted Fallback" — clear which one you're using - Faster Remote reconnect after your laptop wakes from sleep

• New Repos tab — deploy any GitHub repo with one tap • Per-app traffic counters (1d / 7d) on every tile • Refined drawer styling across the app • Remote chat now recovers automatically after your Mac sleeps

• Analytics — see pageviews and unique visitors per app • Push notifications when your agent finishes a task

Ideas: tap the chat menu to have your agent surface what's worth tackling next — security, speed, UX polish, cleanup, docs, and feature suggestions. Tap any idea to hand it straight to your agent. Phone call upgrades: tap the orb mid-call to interrupt the agent and reply right away. Long tasks get periodic check-ins so you're never staring at a silent line.

- Usage dashboard: new Usage row in Settings with avg daily cost inline, drawer shows range toggle, daily cost chart, and recent requests list - Sign in with OpenAI via OAuth (open browser → paste redirect URL) - Bring Your Own Domain drawer with DNS setup in app settings - AI Providers row updates instantly after connecting a key - Stability: retries on transient remote agent WS gaps; app opens back on the tab you left it on - Push notifications: deep-link straight to the right app

Phone Call with Agent: Tap the phone bubble in Remote chat to hop on a real voice call with your agent. Speak naturally, hear it back in a human voice, and the call loops back and forth until you hang up. • Siri-style waveform that reacts to your voice while you talk • Live "Vibing / Cooking / Shipping" status during thinking • Mute just gates the mic — no state reset, no flow break • Auto-picks the most human-sounding voice installed on your device • Smart silence detection (shorter wait on "yes", longer mid-sentence) Plus: • Input bar expands when you type, phone bubble tucks away • Remote chat recovers the full response even if you force-close the app mid-stream • Faster offline detection when your laptop closes • Cleaner bubble heights across the chat

- Fresh app icon - Smoother cycling activity label (ellipsis now disappears between words) - Scroll-to-bottom button matches the input bar style, aligned right - Stability improvements under the hood

Smoother Remote tab. - Agent reconnects fast when your Mac briefly loses connection - Get push notifications when the agent finishes a task - Leave the app mid-run and we pick up where the stream left off - Chat history persists across app restarts - Tap any image in chat to view full-screen - Fixed a tool-list layout bug that caused extra whitespace - Clearer error messages when streams are interrupted

New: Remote tab — link your Mac and chat with your local Claude Code session from your phone. Send messages, attach images, watch tool calls stream in real time. Plus: - Clearer BYOK provider badges (OpenAI, Anthropic, DeepSeek) - Better error messages when the agent is recovering - Polished chat input, attachment picker, and drawer UI - Various bug fixes and performance improvements

- BYOK provider badge now correctly shows your connected key (OpenAI/Anthropic) - Agent tool calls stream in real-time — see what your agent is reading, editing, and running as it works - Cleaner tool call labels (short filenames instead of full paths) - Faster agent responses (~5s, down from ~17s) - Memory tile in app drawer shows clickable chevron for upgrades - Agent management buttons with full-width text wrapping - API key sheets decode all account fields correctly

- Agent management: clear chat, sessions, or memory individually to reset your agent - Fixed duplicate messages appearing in chat - Container health monitoring: view CPU, memory, and disk usage per app - Memory upgrades available directly from the app - Voice input now auto-sends when you finish speaking - Pull-to-refresh on Profile and Settings - Improved message streaming and response formatting - Various UI polish and bug fixes

- Fixed duplicate messages appearing in chat - Improved streaming UI with inline typing indicator - Better activity labels (Thinking, Running tools) - Profile editor with display name support - Balance and Plan rows now tap directly to manage - Separated BYOK and Platform Model in AI Providers - Input bar padding and bubble sizing improvements

- Redesigned AI Providers settings with easy-connect flow for Anthropic and OpenAI - Platform model selection (Anthropic, OpenAI, DeepSeek) with provider icons - Real-time tool call activity display during AI conversations - Share website link from chat - Custom domain support shown in app - Improved onboarding with interactive terminal animation - Redesigned Face ID lock screen - Pull-to-refresh on home screen - Better markdown rendering in chat messages - Plan picker now accurately reflects all features - Various UI polish and bug fixes