Published , 04:44 Bogota (UTC-5) · 13 sourced items, 13 new since the previous edition · Read the foundations review · RSS
Shopify is moving from React Native back to separate Swift and Kotlin codebases for their native mobile apps. The change is attributed to agents now being able to handle much of the implementation, translation, testing, and review work.
Why it matters: This suggests that AI agents are becoming capable enough to reduce the development overhead of native mobile app development, influencing deployment strategies for AI on phones.
Calif Research released a demo of WeWorm, a zero-click worm spreading through WeChat calls on iOS and Android. The team, working with AI, found the bug and wrote the first remote code execution exploit in about two days, and built the worm in one week.
Why it matters: This demonstrates AI's capability to accelerate security research and exploit development, highlighting potential security implications for phone AI and local systems.
LiteRT-LM v0.17.0 introduces optimized local attention for reduced memory overhead and longer contexts, Metal residency support for Apple Silicon acceleration, and extended Gemma 4 (12B) with multimodal capabilities, multi-token prediction acceleration, and ex
Why it matters: These updates significantly improve performance and capabilities for running models like Gemma 4 on Apple Silicon and other devices, which is key for phone AI and local inference.
llama.cpp releases b10948, which includes tests for macOS Apple Silicon, Linux (x64, arm64, s390x, Vulkan, ROCm, OpenVINO, SYCL), Windows (x64, arm64, OpenCL Adreno, CUDA 12/13, Vulkan, OpenVINO, SYCL, ROCm), and Android arm64. It excludes HY_V4 from WebGPU te
Why it matters: This release indicates broad platform support and ongoing development for llama.cpp, which is crucial for local inference across diverse hardware.
vLLM release v0.29.1rc0 introduces dual-key gumbel-max watermarking for speculative decoding.
Why it matters: This feature could impact how models are served and verified, especially in sandboxed or local environments.
vLLM releases vllm-proto 0.1.0.
Why it matters: This indicates a new protocol version for vLLM, potentially affecting compatibility or new features for local inference setups.
Ollama v0.34.0 allows using Ollama models directly in ChatGPT Desktop, with setup available from the Ollama app on MacOS. It also improves structured output performance on Apple Silicon and adds support for OpenAI-compatible client tool search and response com
Why it matters: This release improves the integration of local Ollama models with existing workflows, especially for macOS users, and enhances performance on Apple Silicon.
Ollama v0.34.0-rc5 adds support for standalone named function outputs in its OpenAI compatibility.
Why it matters: This improves Ollama's compatibility with OpenAI's API, making it easier to use local Ollama models with tools designed for OpenAI.
e2b@2.49.1 patch changes include rejecting invalid Sandbox.create lifecycle options and Sandbox.connect onResume values before requiring an API key. It also adds retries for control-plane HTTP requests after 429 responses, configurable or disableable.
Why it matters: These updates improve the robustness and error handling of the E2B SDK, making sandboxes more reliable for agent development.
E2B infra release 2026.30 removes deprecated access-token authentication, adds sandbox-list sorting and filtering, and introduces sandbox workload identity configuration and feature-gated secrets operations. It also adds dynamic log routing.
Why it matters: These infrastructure updates enhance security, management, and observability for E2B sandboxes, which is important for agent development and deployment.
An E2B blog post describes building an agent workbench on OpenAI's Agents API (beta) and E2B sandboxes, featuring application-managed lifecycle, one sandbox per chat, pause, and fork.
Why it matters: This illustrates how E2B sandboxes can be used to build and manage agent workflows, providing a practical example for local agent development.
e2b@2.49.0 exposes a configurable minimum free-disk target with minFreeDiskMb in JavaScript, min_free_disk_mb in Python, and --min-free-disk-mb in template create.
Why it matters: This allows users to manage disk space more effectively within E2B sandboxes, which is important for resource management in local or sandboxed agent environments.
Alibaba released open weights for Qwen3.8-2.4T-A95B (Qwen3.8-Max), its largest open-weight model, bringing near-frontier capabilities to the open.
Why it matters: This introduces a large open-weight model, which could be relevant for local inference on high-end hardware or for understanding model capabilities.
Sources: arXiv API, Apple Machine Learning Research, NVIDIA, Google Research, Google Developers, Microsoft Research, Hugging Face, MLCommons, MIT News, Nature Machine Intelligence, Communications of the ACM, and official GitHub release feeds (MLX, llama.cpp, Ollama, vLLM, MLC LLM, LiteRT-LM, E2B). Items are filtered by topic rules; summaries are AI-assisted (gemini-2.5-flash) and grounded only in each source's own abstract or post text. Always read the linked source before acting.