Apple Intelligence on-device, ?prefix search mode, push when jobs finish, and one-tap Save as Skill from any trajectory - the biggest update since launch. APPLE INTELLIGENCE ON-DEVICE (NEW, iOS 26) Save as Skill, /standup, and /whatif now run entirely on your iPhone or iPad when no Anthropic API key is set. Free, instant, fully private - Apple's FoundationModels framework, nothing leaves the phone. Cascade is API key first (best quality), then Apple Intelligence on-device, then Claude CLI on your Mac as a final fallback. Supported devices: iPhone 15 Pro and newer, M1+ iPads. Toggle in Settings - AI Intelligence (defaulted ON when supported). ? SEARCH MODE Start any message with ? to route it through propose-evaluate-refine - three chains explore different angles, refinement rounds polish the best ones. Tap the result to open the trajectory graph: chains as columns, rounds as rows, best attempt highlighted with a star. Long-press send for "Send with Search" (one-off trigger) or "Speak with Search" (voice). Pin the winner to ground every follow-up message in that conclusion. SAVE TRAJECTORIES AS SKILLS Long-press Select on any research result or background job trajectory, pick 2 to N entries, tap Save as Skill (N). Auto-loads next time a similar question matches its keywords. Works with an API key (3 to 5 seconds) or via Claude CLI on your Mac (30 to 60 seconds). PUSH NOTIFICATIONS WHEN JOBS FINISH Two paths: /notifier setup wires Apple Push Notifications from your own Mac (paid Apple Developer account), or /notifier telegram routes through your own Telegram bot for universal-reach push without a developer account. Either way - real iPhone banners when background jobs complete. NATIVE /research ENGINE /research no longer needs SimpleTES installed on your Mac. The new in-app loop runs propose-evaluate-refine entirely against the Claude API at about a penny per run. Live progress sheet, full trajectory graph, "Send as Message" to paste the winner. iOS 26 NATIVE - Writing Tools (Rewrite, Proofread, Make Friendly) in the chatroom input bar and skill editor - SpeechAnalyzer for /voice and Speak with Search - better accuracy, lower latency POLISH - Slash palette tip-of-the-day surfaces underused commands (/effort, /btw, /think, /duck, /recap, /context, /retry, /pin, /wrapped, /whatif) - ? typing hint chip below the input bar - Skill library "Last fired" badges and a Sort by Last used menu - Cleaner Search section in Background Jobs (auto-prunes ?prefix jobs older than 7 days) - "What's New" sheet on first launch after every update Plus everything from v1.3.0 unchanged: Learn Hub, /frugal, /duck, /whatif, /standup, /wrapped, /screenshot, Code Postcards, Context-Aware AI, Proactive Insights.

The biggest update yet - a unified Learn Hub, ten "wow" tools, and a Claude that finally understands your situation. LEARN HUB One searchable place for everything: 6 interactive tutorial tracks (43 hands-on steps), 36 written tutorials, and 33 examples - all bundled offline. Tap any example's "Try in chatroom" button to drop the command straight into your input bar. Tap Refresh to pull the latest from GitHub. NEW POWER COMMANDS - /whatif [command] - predicts the 3 most likely outcomes before you run a risky shell command, with probability bars and severity colors. - /frugal (or "!" prefix) - strips the AI's preamble for cheaper, faster turns. Skips skills, memory, dream profile, device context - keeps soul + personality + effort. - /duck - switches Claude to Socratic rubber-duck mode. Only asks probing questions, never gives answers. Surprisingly good for unblocking yourself. - /standup - auto-writes your daily standup from yesterday's chat plus git log. Copies to clipboard. - /wrapped - Spotify Wrapped-style 9-page summary of your activity, top commands, AI spend, peak hour, and longest session. - /screenshot and /screenshots - capture chatroom views with auto-tagged topics. Searchable gallery. CONTEXT-AWARE AI (OPT-IN) Claude now adapts to your situation: - Time of day - terse at night, verbose in work hours - Calendar - knows when your next meeting is - Motion - shorter responses when walking or driving - Battery - skips expensive multi-agent commands when low CODE POSTCARDS Long-press any chat message, tap Share as Postcard, and ClawTerminal renders a 1080x1080 image with syntax-highlighted code, gradient background, and branding. One tap to share to Twitter, Discord, Slack, or Messages. PROACTIVE INSIGHTS (OPT-IN) Background pattern detection notices things like "deploys fail 3x more often on Fridays" or "you retry /health 4x per session" and surfaces high-confidence findings as notifications. WELCOME TOUR THAT KEEPS UP The tour now auto-relaunches with each app update so you always see what's new on first launch. POLISH - Export conversation moved to a pinned menu on the room tab bar - always visible no matter how many rooms you have. - Info and Project merged into one sheet with a Project context row inside. - Ambient sound (opt-in) - subtle audio cues for AI thinking, job complete/fail, stream start. - Frugal mode now also tells Claude to skip plan confirmations and "would you like me to" questions, so frugal turns actually feel terse. Plus 60+ existing features unchanged: multi-agent orchestration, dream mode, gestures, mini-terminal, conversation branching, /pr, /git, /health, /monitor, /preview, and more.

v1.2.0 is our biggest update — 50+ new features that make CatClaw the most powerful mobile AI coding tool available. (Details: https://github.com/LiqunChen0606/clawterminal-docs) RACE MULTIPLE AI MODELS /race runs the same prompt on Claude, Codex, and Gemini simultaneously. Compare side-by-side on iPad or swipe between responses on iPhone. New "thinking lenses" give you 6 different perspectives from the same model. GITHUB PR WORKFLOW /pr creates pull requests from your current branch. /pr review runs AI code review with color-coded severity items. Your feedback trains future reviews to focus on what your team cares about. SESSION HANDOFF /handoff mac sends your phone session to a Mac terminal. /handoff picks up any active Mac session on your phone. Start on your commute, finish at your desk. LIVE WEB PREVIEW /preview opens your dev server in-app via SSH tunnel. Auto-detects ports. Multi-port tabs for frontend + backend. Console log capture with "Send errors to Claude" button. Responsive viewport modes. Screenshot + annotate to report visual bugs. VISUAL DASHBOARDS /git shows a visual branch graph. /health scans codebase metrics. /monitor shows live server CPU/memory/disk. /dashboard shows your AI usage analytics. DEVELOPER INTELLIGENCE /spec generates formal requirements. /graph queries codebase architecture. /tribal extracts unwritten project knowledge. /security runs deep scans. /gentest generates tests. /changelog creates release notes. SMARTER AGENTS --routing balanced auto-assigns Opus/Sonnet/Haiku per agent role. --vcs gives each agent its own git branch. Agent reasoning banners show why decisions were made. Failed jobs offer auto-recovery with error context. WORKFLOW AUTOMATION /workflow chains commands into reusable pipelines. /hooks triggers agents on file changes. /notify sets up push notifications for conditions like "when tests finish." UX INNOVATIONS Smart follow-up suggestion chips. Long-press Send to background a job. Split-screen mini terminal. Conversation branching — fork and compare approaches. JSON table cards. @web URL context. /pin file pinning. Cat mascot and haptic feedback.

What's New in v1.1.1 Navigation • iPhone: slide-out drawer sidebar (like ChatGPT/Claude) replaces tab bar • iPad: sidebar navigation fully fixed — tabs switch correctly • Swipe from left edge to open drawer, swipe left to close AI Agent Improvements • Agent Teams: discovery passing between waves improved (10x more context) • Smarter fallback when discovery extraction fails • Adaptive polling: reduces battery usage for long-running jobs Terminal & Keyboard • Orange keyboard dismiss button — easy to spot • iPad: no more overlapping shortcut bars with external keyboard • Chat bubbles: more text width with compact avatar layout Connection Reliability • SSH timeout increased to reduce false disconnects • Exponential reconnect backoff (5s → 15s → 30s) • Orange "Reconnecting..." banner when connection drops • Background heartbeat: notifies you when jobs are still running Commands • /usage — check token usage (alias for /cost) • /help — redesigned with grouped tables • Jobs tab: agent groups auto-collapsed for cleaner view

What's New in v1.1.0 Agent Teams — Mobile Command Center • /team command: wave-based multi-agent orchestration • Visual command center with animated flow graph • Agents share discoveries between waves automatically • Tap "Team" in toolbar to monitor progress Multi-Agent Batch Processing • /batch --agents N: spawn parallel AI workers • Commander decomposes, workers execute, synthesizer merges • --multi flag: auto-assign Claude, Codex, Gemini, or Aider Smart Custom Commands • Named parameters with defaults (/deploy staging main) • Auto-submit as background job or batch • Attach skills and override CLI tools per command New Features • SFTP: create folders and files directly • File picker: attach markdown, code, PDF from Files app • Welcome tour: revisitable from Settings → About Reliability & Fixes • SSH auto-reconnect: retries 3x instead of giving up • Claude CLI login detection with step-by-step fix • First-time setup no longer flips back on connection error • Mac sleep prevention tip in setup wizard