Loach
A native, local-first AI workspace for desktops. Run local LLMs with Ollama, or connect any OpenAI-compatible endpoint side-by-side - with a focused UX, native apps for Windows, Linux and macOS, and powerful features out of the box.
What is Loach?
Loach is an all-in-one desktop AI workspace built around a single idea: talking to an LLM should feel effortless. Simple from the first click and ready to grow with you as your needs do. It talks to a local Ollama server and accepts any OpenAI-compatible endpoint as a provider - including llama.cpp, LM Studio, vLLM and LiteLLM.
With local models - Qwen, Gemma, DeepSeek, GPT-OSS, Mistral and others - your data stays on your machine. There is no telemetry, no required network access, no paid subscriptions and no usage limits. An optional OpenAI-compatible API key lives in your operating system's credential manager.
Behind a calm, beautifully crafted UI sits a rich feature set - ready when you need it, out of the way when you don't.
Highlights
Two providers, side-by-side
Switch between a local Ollama server and any OpenAI-compatible endpoint right from the chat header - and start Ollama from Loach when it isn't running.
Work in folders
Add a folder to a chat and the model can list, find, read, search and edit its files - sandboxed, with no shell, and asking before it changes anything.
Global memories
Opt in and Loach notes lasting facts about you and brings them into every chat. Facts that change get updated, with an Undo on every change.
MCP servers
Connect Model Context Protocol servers over HTTP or run them as local programs (npx, uvx, …), and approve each tool call before it runs.
Spaces
Group chats around a project, with shared instructions, reference sources - uploads or Library items - and their own memory.
Library
Save texts, files and links once and attach them to any chat or Space. Small sources go to the model in full; big ones are searched by keyword or by meaning.
Snippets
Reusable prompts with an optional pinned model and fill-in variables - click Run and a fresh chat is pre-filled.
Built-in tools
Local tools the model can call - calculator, dates, JSON, text diff, PDF generation and more - plus opt-in web fetch.
Code canvas
Open any code block in a wide, resizable view that updates live as the model streams. HTML previews as a page in an offline sandbox, with scripts off until you run them.
Search
One Ctrl + K palette across chats, message text, Spaces, Snippets and the Library, plus a find-in-chat bar.
Local-first & private
Chats, Spaces, the Library and settings live in one local SQLite file. Your API key and app-lock hash go to the OS credential manager.
Private Chat
A dark-only, ephemeral overlay for conversations that leave no trace. Wiped the moment you close it or Loach locks.
Where to next?
- Installation - pre-built packages for Windows, Linux & macOS, or build from source.
- Quick start - your first chat with a local model in two minutes.
- Core concepts - providers, chats, Spaces, the Library, Snippets, parameters.
- Features - the full feature reference.
- FAQ - the questions people ask first.
- Troubleshooting - common problems and how to fix them.