VoiceStudio, Splash and OpenShell: local AI tools overview

This roundup surveys GitHub projects for local AI, app development and agent workflows. The speaker presents VoiceStudio as a fully local alternative to ElevenLabs for voice cloning, transcription, dubbing and audiobook creation. The presenter describes macOS, Windows and Linux support, with a catalog of speech engines that can use CPU or GPU resources. The narration says it needs no accounts, API keys or cloud service.

Splash targets Apple Silicon Macs. The overview describes a Homebrew installation, a server started with one command, and support for GGUF and MLX models. Compatible coding tools can connect to the server, while a built-in web interface provides another way to use it.

For agent infrastructure, OpenShell is described as a sandbox runtime with YAML policies and deployment through Docker or Kubernetes. Those security properties are the speaker's claims. Octop is presented as a self-hosted assistant with separate user workspaces, a dashboard and scheduled automation. The overview describes Worktrunk as managing Git worktrees for parallel agents, and Whiteboard connects software diagrams and review notes to local code.

The remaining projects cover interface design, app SDKs, media conversion and development utilities. The presenter describes the model segment as offering Naive-N0.5-Flash weights in FP8 and GGUF, with FP8 inference requiring capable NVIDIA GPUs. imajev receives a brief introduction as a local image-and-text decision model using MLX or PyTorch.