Skip to content

Releases: tachikomared/tachi-studio

Release list

Tachi Studio 0.1.0

Tachi Studio 0.1.0 Latest
Latest

Choose a tag to compare

View all tags
@tachikomared tachikomared released this 21 Aug 03:40
v0.1.0
f3d8ebc

Rebuilt 2026-08-21. This build fixes a crash on machines without Ollama
installed — opening the chat tab raised "A JavaScript error occurred in the
main process — spawn ollama ENOENT". It also carries a packaging fix: the
previous build's app.asar had swallowed the whole source tree because a
per-platform file filter collapsed the include list.

First public build. Tachi Studio is a local-first desktop AI workstation:
multi-provider chat, coding agents that work on your own repositories, a
node/flow canvas, and local image, video and voice generation — in one window,
on your own machine.

Download

File What it is
Tachi-Studio-Setup-0.1.0.exe The installer. Per-user, so Windows does not ask for admin rights, and you choose the folder.
Tachi-Studio-0.1.0-win.zip Portable. Unzip anywhere, then create an empty folder named tachi-data next to the .exe — the app keeps settings, chats, models and logs inside its own folder and writes nothing to C:\Users\….
latest.yml, *.blockmap Used by the in-app updater. You do not need to download these.
SHA256SUMS.txt Checksums for everything above.

Windows will warn you on first run. The build is not code-signed yet, so
SmartScreen shows "Windows protected your PC". Choose More info → Run
anyway, or verify the download first:

Get-FileHash .\Tachi-Studio-Setup-0.1.0.exe -Algorithm SHA256

and compare it with the line in SHA256SUMS.txt.

What is in it

  • Chat with Anthropic, OpenRouter, Ollama, llama.cpp and a built-in free
    router across roughly seventeen keyless providers — so it works before you
    configure anything. Folders, full-text history search, PDF reading, and
    citations over an attached folder.
  • AUTO model selection — a local, sub-millisecond classifier picks the
    model per message, and learns which ones actually work for which task.
  • Coding agents on a folder you choose, with per-tool permission prompts,
    a path boundary and a network policy. An agent that gets stuck says so.
  • Node canvas — providers, prompts, agents, media and webhooks as blocks;
    save a flow as a file, run it on a schedule, or trigger it by webhook.
  • Local generation — Stable Diffusion images and video, Piper speech,
    Whisper transcription, on your GPU, offline. Engines download on request and
    are verified against the publisher's own SHA256 before they run.
  • Private mode that is enforced in the main process, not by the interface.
  • An OpenAI-compatible server on 127.0.0.1, so Claude Code, Cursor or any
    OpenAI SDK can use this app as their backend. Plus an MCP server, and a
    catalog for installing MCP servers.
  • Eight languages and swappable themes, some of which redraw the window
    frame itself.

Known limits

  • Windows only for now. macOS and Linux targets are configured and the code
    is written for them, but neither has been through a real install and test, so
    no binary is published. Building from source works — reports welcome.
  • Not code-signed. See the SmartScreen note above.
  • Pre-1.0. Expect rough edges, and please
    open an issue when you
    find one.

Building it yourself

git clone https://github.com/tachikomared/tachi-studio.git
cd tachi-studio
pnpm install
pnpm prepare:sidecars
pnpm dev

Full notes on third-party licences — including the components that are not
MIT — are in THIRD_PARTY_NOTICES.md.

Assets 7
Loading