iRun Studio

The "Meltdown" Update & Remote Compute Power! Version 0.3 brings major platform upgrades, expanded local memory capabilities, and the ability to link your Apple device to external hardware for heavy-duty inference. iRun Link (Remote Compute Client) Custom Server Connection: Connect iRun directly to your custom Windows or Linux server to offload heavy inference tasks to remote CUDA hardware over your local network. Meet "Meltdown" (New Local AI & Mascot) Custom AI Model: Experience "Meltdown"—our custom-trained local AI model. Downloadable directly in-app or via HuggingFace without inflating the initial app size. Onboarding & Tutorial: Get to know your new mascot with an interactive walkthrough designed to introduce you to advanced tool execution and local workflows. Enhanced Performance & Memory Expanded Device Memory: Unlocked extra memory entitlements to handle larger local models and larger context windows with far greater stability. Lower Heat & Reduced Load: Optimized GPU and CPU thermal management to keep your device running cool and fast during long generation sessions. Core Engine Upgrades: Updated the llama.cpp core build, resolving double mmproj bugs while stabilizing Speculative Decoding and Multi-Token Prediction (MTP). Tool Calling & Sub-Agent Upgrades Smarter Execution: Significantly increased execution accuracy and tool usage reliability across all local models. Fixed Tool Routing: Resolved issues where sub-agents failed to operate in edge cases, alongside fixes for harness execution. Crashes & UI Fixes Fixed a crash when clicking the remote connection button from the chat sidebar. Resolved crash issues when rapidly switching tabs. Refreshed layout elements, updated domains, and overall UI polishing. Enjoying iRun? Leave us a review on the App Store!