sybil-solutions/local-studio

Control panel for VLLM, Sglang, llama.cpp, exllamav3

Local Studio – A desktop UI for self‑hosted LLMs

What it is

  • A thin fork of the open‑source T3 Code application that adds first‑class support for running your own large language models locally.
  • Provides a graphical “agent UI” where you can chat with models, run tool calls, view threads, and manage source‑control‑style sessions.
  • Bundles a local controller that discovers any inference server on the machine (vLLM, SGLang, llama.cpp, or any server exposing an OpenAI‑compatible /v1/models endpoint) and re‑exports it under a single OpenAI/Anthropic‑compatible gateway.
  • Enables a fleet of machines linked via Tailscale so a model loaded on any computer in your private network can be used from any other, with one‑click installation of the controller on Linux/macOS hosts.
  • Ships with extra providers – Pi and Oh My Pi – that run alongside popular hosted models (Claude, Cursor, Grok, etc.) and support streaming, tool calls, interrupt, rollback, and token‑usage tracking.

Why it matters

  • Lets developers, researchers, or hobbyists keep their data and models in‑house while still getting a polished desktop experience similar to commercial AI‑assistant tools.
  • The Tailscale‑based fleet makes it easy to pool GPU resources across several machines without writing custom networking code.
  • By re‑exporting local inference servers as standard OpenAI endpoints, existing libraries and tools that expect the OpenAI API can be used unchanged.

Getting started

  1. Download the installer for your OS from the latest GitHub release (DMG for macOS, .exe for Windows, AppImage or .deb for Linux).
  2. Run the installer – macOS builds are signed/notarized; Windows builds currently require a SmartScreen confirmation.
  3. After launch, log in to at least one agent CLI (e.g., pi or omp) so the desktop can discover the local controller.
  4. Open Settings → Local to see any other machines on your Tailscale network. Click Connect if a controller is already running, or Install to have Local Studio automatically install a controller via SSH on a Linux/macOS host.
  5. Choose a provider (Pi, Oh My Pi, or any OpenAI‑compatible server) and start a chat or run a script.

Developer workflow

  • Requires Node 24, pnpm 11, and Bun 1.3.14.
  • Install dependencies with pnpm install.
  • Run the controller alone: pnpm --filter @local-studio/controller start.
  • Launch the desktop in dev mode (hot‑reload) with pnpm dev:desktop --port 18773 --home-dir .t3.
  • For production use, build then start: pnpm build:desktop && T3CODE_HOME="$PWD/.t3" pnpm start:desktop.
  • The repository enforces a 4 000‑line source budget (node scripts/local-check.mjs) to keep the fork lightweight and easy to merge upstream changes.

Release process

  • CI builds signed macOS DMGs, Windows NSIS installers, and Linux AppImages/.deb packages for both x64 and arm64.
  • Controller binaries and update manifests are also published.

License

  • MIT license, same as the upstream T3 Code project. The additional Oh My Pi components have their own license in assets/OMP-LICENSE.txt.

Bottom line: Local Studio is a practical, open‑source desktop front‑end for self‑hosted LLMs, with built‑in multi‑machine fleet management via Tailscale, making it a solid choice for anyone who wants the convenience of a modern AI assistant while keeping the inference stack under their own control.

Related

  • Project
    RoseKhlifa/Image-Studio

    Image Studio is an open‑source, cross‑platform desktop client (Windows/macOS/Linux/Android) built with Go + React (Wails) that lets you generate and edit images via any OpenAI‑compatible API. It supports streaming responses over SSE or WebSocket to avoid proxy timeouts, works with the standard Images API, offers batch img2img, custom aspect ratios, and integrates with a prompt‑import helper. Configuration is manual (base URL, API key, model ID) and the project is AGPL‑v3 licensed.

  • Project
    MindWorkAI/AI-Studio

    MindWork AI Studio is a free cross‑platform desktop app that lets users chat with many LLM providers (cloud or local), use pre‑built assistants, run tool‑enabled RAG, and manage settings centrally for enterprises. It runs on low‑cost hardware, supports voice transcription, and is released under a functional source license that becomes MIT after two years.

  • Project
    lmstudio-ai/lms

    A terminal tool for managing LM Studio – start/stop its API server, list/load/unload local LLMs, create SDK projects, and stream logs.

  • Project
    chadbyte/clay

    Clay Studio is a self‑hosted web workspace that lets humans and locally‑installed AI coding agents (Claude, Codex, Copilot CLI, etc.) collaborate in shared, persistent sessions. It runs a Node daemon on your machine, serves a browser UI, stores project state on disk, and routes requests to the agents’ own CLIs, keeping code and context under your control while enabling real‑time teamwork and background automation.

  • Project
    debpalash/VoiceStudio

    VoiceStudio is an open‑source desktop app (Electron) for voice cloning, voice design, video dubbing, dictation, transcription and audiobook creation in hundreds of languages. It runs locally on macOS, Windows or Linux, supports GPU acceleration, offers a model manager, and exposes a local API/MCP for integration with AI agents.