Release Notes

Dual-provider AI Autopilot — OpenAI Computer Use added

• OpenAI Computer Use — AI Autopilot now supports OpenAI Responses API (GPT-5.5 / GPT-5.4 / computer-use-preview) alongside Anthropic (Sonnet 4.6 / Opus 4.7). GPT-5.5 is fast, Tier 1 friendly, and ~25% faster than Sonnet 4.6 in field tests
• Unified default-model picker — AI Settings > AI Autopilot now has a single dropdown listing all provider × model combinations. Picking a model also sets the default provider
• Safety check dialog — When OpenAI Computer Use requests permission for sensitive actions (irrelevant_domain, sensitive_domain, malicious_instructions), a dialog asks for explicit approval
• LLM model lists refreshed — All providers updated: OpenAI gpt-5.5/5.4/5.2/5.1, Anthropic Sonnet 4.6 / Opus 4.7 / Haiku 4.5, Gemini 3 series (Preview) + 2.5 stable, Groq Llama 4 Scout 17B (multimodal)
• Groq Vision support — Llama 4 Scout enables AI OCR/Click/SmartWait/Validate via Groq's ultra-fast inference
• Provider model dropdowns — Each provider's model is now selected from a curated list with descriptions, instead of free text

Major release: AI Vision (Computer Use) + Apple Intelligence

• AI Click — AI sees the screen and clicks any element by description. 2-pass refinement for pixel-perfect accuracy
• AI Smart Wait — AI monitors the screen and waits for any visual condition
• AI OCR — AI reads any text from the screen using Vision LLM
• AI Validate — AI checks screen state and returns true/false for conditional flows
• AI Autopilot — Autonomous agent loop powered by Anthropic Computer Use API (Claude Sonnet 4.6 / Opus 4.6). Describe a task and AI executes it with native computer control. Requires Anthropic API key
• AI Condition (Boolean) — Use AI to evaluate conditions in if/loop blocks
• Self-Healing — When image/OCR operations fail, AI Vision auto-retries with natural language
• Recording Enhancement — After recording, AI analyzes click positions and suggests converting to AI Click blocks
• Apple Intelligence — On-device AI for the built-in assistant (macOS 26+). No API key needed, free, private, offline
• Separate AI Settings — Independent provider settings for AI Assistant (Apple Intelligence default) and AI Vision Blocks (cloud providers)
• AI Assistant improvements — Provider name shown on responses, console logging
• App Launch fix — Fallback to app name when .app bundle path doesn't exist
• Block insertion fix — AI-generated blocks now auto-connect to the end of the start block chain
• 196 total operations (up from 169)
• 6 AI providers: Apple Intelligence, OpenAI, Anthropic, Gemini, Groq, Custom/Local LLM
• v2.0.2 fixes: NativeAutopilotService crash fix (JSON serialization), Apple Intelligence context window overflow fix (catalog omission), macOS Spotlight workaround in Autopilot system prompt
• v2.0.3: Unified AI block parameter labels to "Prompt". Expanded help documentation with AI block guide, prompt examples, recommended models, execution/debug reference, and keyboard shortcuts
• v2.0.5: Default Gemini model updated to gemini-2.5-flash (free: 250 req/day). Added gemini-2.5-flash-lite (free: 1,000 req/day) as budget option
• v2.0.6: Trackpad support: two-finger swipe now pans the workspace (previously triggered zoom). Pinch to zoom. Cmd+scroll for mouse wheel zoom

The first public release of RocketMouse AI for macOS.

• 169 automation operations across 24 categories
• Visual block editor with 7 block shapes
• Drag & drop, snap connections, undo/redo
• AI Assistant with 5 LLM provider support
• Browser automation via Playwright + Chrome
• Excel read/write with CoreXLSX + AppleScript
• File operations (copy, move, delete, text I/O, ZIP)
• Image recognition & template matching via Vision framework
• OCR text extraction
• Screen capture via ScreenCaptureKit
• Macro recording (mouse + keyboard)
• Debug tools: breakpoints, step execution, DryRun mode
• Variable system with auto-naming
• Custom functions with call stack (depth 100)
• Speed control (0.1x to 5.0x)
• Minimap navigation for large macros
• Block search / palette search
• Natural language template display on blocks
• Full .rmproj compatibility with Windows version
• Localized in English and Japanese (833 keys)
• Universal binary (Apple Silicon + Intel)
• Dark mode UI
• Apple Developer ID signed + notarized