A native, local-first AI workspace for desktops
Run local LLMs with Ollama or connect any OpenAI-compatible API endpoint side-by-side, with a focused UX, local-first design, native apps for Windows, Linux and macOS, and simple yet powerful features out of the box.
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, including Qwen, Gemma, DeepSeek, GPT-OSS and Mistral, all your data stays safe and private. There is no telemetry, no required network access, no paid subscriptions or usage limits. Optional API keys live in your OS 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.
- Provider selection - switch between Ollama local models and any OpenAI-compatible endpoint (OpenAI API, llama.cpp, LM Studio etc.) from the chat header.
- Local model management - pull, copy, customize and delete local models from inside the app.
- Default model selector - pick a model new chats open with, per provider - no need to re-select on each fresh conversation.
- Model preloading - optionally warm your default local model into VRAM at launch so the first message streams faster.
- Per-chat parameters - set temperature, top_k, top_p, min_p, max tokens, context length, per-chat system prompts and more. Layered over Modelfile and per-model defaults.
- Low VRAM mode - global or per-chat toggle that sends Ollama's
low_vramflag to every request. Useful on lower-spec devices.
- Spaces - group chats around a project, with shared instructions, reference files and memory.
- Snippets - save reusable prompts with an optional pinned model and click
Runto start a fresh chat pre-filled and ready to send. - Custom snippet variables - parameterize snippets with static globals and prompt-on-use placeholders that fill in when you run them.
- Fork chats - branch any conversation into a new copy that links back to its source.
- Private chats - an ephemeral chat that writes nothing to disk and wipes its transcript the moment you close it; open it from the ghost icon in the title bar.
- Chat archive - move chats out of the sidebar without deleting them; restore or delete them from dedicated archive view.
- Search - search across chats, spaces and snippets, plus a browser-style in-chat finder with phrase highlighting.
- Slash commands - type
/in the composer for a command palette:/fork,/regenerate,/compact,/private,/model,/persona,/snippet,/rememberand more. - Personas and Tones - pick a role (Code Reviewer, Translator, ELI5...) and delivery style (Formal, Casual, Direct, Detailed...).
- Context management - a live bar under the composer shows how full the context window is, with one-click compaction that summarizes older turns to free space.
- Import / export context - export chat context to JSON or Markdown, optionally summarized to compact it, and paste exported data - or any text - back to any chat's context.
- Tools - let models call local tools including calculate, date/time, count, hash, UUID, base64, JSON, unit convert, text diff, sort, IP math and PDF generation.
- PDF generation - the built-in pdf tool turns a model's structured spec (headings, lists, tables, page breaks) into a real PDF attached to the reply.
- MCP support - register Model Context Protocol servers (Streamable HTTP), test the handshake and inspect the tools they provide.
- Web fetch - add URLs to messages and they will be fetched, sanitized and inlined to context.
- Temporal awareness - inject current date, time, weekday and timezone into the system prompt so models can answer to "what day is it today?".
- Code canvas - open any code block in a wider view that's resizable and updates live as the model streams, with copy, export and Open in VS Code actions.
- Attachment previews - click an attachment to open the right viewer: image lightbox, multi-page PDF preview, code canvas, or a file-info card with Save.
- Data management - make backups of your content to JSON file, restore data or permanently delete it with a few clicks.
- App lock - optional PIN, password or PIN + password gate at launch; credentials are hashed and stored in OS credential manager.
- Themes - glassy, gradient Aurora or flat Solid, both available in Dark and Light variants.
- OTA updates - get new features, bug fixes, performance improvements and security patches directly from the app.
...and more!
We are working on extending the list above, including new RAG and agentic features.
Loach can be installed from a pre-built package (Windows, Linux, macOS) or built from source.
With each stable release we publish pre-built .exe, .deb, .rpm, .AppImage or .dmg packages ready to be downloaded and installed according to your operating system.
👉 Download a pre-built package from the latest stable release
Note
For local models make sure Ollama is up and running (ollama serve). If you don't have any models pulled yet, Loach will offer to install one during onboarding.
The macOS build is Apple Silicon (M-series CPUs) only and is currently not notarized in the Apple Developer Program. On first launch macOS will block the app with a "Loach is damaged and can't be opened" or "Apple cannot verify..." warning. Bypass it once and the app runs normally:
- Right-click
Loach.appin Applications → Open → click Open in the prompt, or - Open Terminal and run:
then launch the app normally.
xattr -cr /Applications/Loach.app
Auto-updates are delivered through Loach's own signed updater (independent of Apple), so you only need to do this once on install. If a future update is blocked the same way, the same workaround applies.
- Node.js 20.19+ (or 22.12+) and npm — Vite 7 won't run on older 20.x point releases.
- Rust stable toolchain via
rustup - Platform build tooling — install once via the official Tauri prerequisites guide: https://tauri.app/start/prerequisites/
- Windows: Microsoft Visual Studio Build Tools, WebView2 runtime (preinstalled on Windows 11)
- Linux:
webkit2gtk-4.1,libayatana-appindicator3-dev,librsvg2-dev,build-essential,libssl-dev,pkg-config,libsecret-1-dev - macOS: Xcode Command Line Tools (
xcode-select --install)
git clone https://github.com/ztcs-software/loach.git
cd loach
npm installnpm run tauri devThe Vite dev server runs on http://localhost:1420 and the Tauri shell embeds it.
npm run tauri buildOutputs land in src-tauri/target/release/bundle/:
- Windows:
.exe(NSIS) - Linux:
.deb,.rpmand.AppImage - macOS:
.dmgand.app(Apple Silicon)
Loach doesn't provide built-in models, therefore it requires a provider choice - Ollama or OpenAI-compatible endpoint (OpenAI API, llama.cpp, LM Studio etc.)
ollama serve
ollama pull gemma4:e4b # or any other tagLoach probes http://localhost:11434 on launch. Pulled models appear in the chat header dropdown automatically; the Models library tab also lets you pull new tags and customise existing ones from inside the app.
Open Settings → Providers, paste your API key (stored safely in your OS credential manager), and point the base URL to any compatible endpoint:
| Provider | Base URL |
|---|---|
| OpenAI | https://api.openai.com/v1 (default) |
| vLLM | http://localhost:8000/v1 |
| LM Studio | http://localhost:1234/v1 |
| LiteLLM | http://localhost:4000 |
llama.cpp llama-server |
http://localhost:8080/v1 |
| Groq, OpenRouter, Together, Fireworks, Mistral | per their docs |
Only one OpenAI-compatible endpoint is active at a time — switch the base URL in Settings whenever you want to point to a different provider.
| Layer | Choice |
|---|---|
| Desktop shell | Tauri 2.x (Rust) |
| Frontend | React 18 + Vite 7 + TypeScript |
| Styling | Tailwind CSS + shadcn/ui (Radix primitives) + tailwindcss-animate + @tailwindcss/typography |
| Icons | lucide-react |
| State | Zustand (in-memory; backed by SQLite for persistence where appropriate) |
| Storage | SQLite via rusqlite (bundled feature — no system dependency) |
| Secrets | keyring crate → Windows Credential Manager / Linux Secret Service / macOS Keychain |
| Argon2id | argon2 + rand_core (for app-lock hashing) |
| HTTP | reqwest with streaming + rustls-tls |
| Markdown | react-markdown + remark-gfm + rehype-highlight (highlight.js) |
| Document parsing | pdfjs-dist (PDF) + mammoth (DOCX) |
| PDF generation | printpdf 0.9 with a bundled Liberation Sans subset (Unicode-capable output) |
| System tray | Tauri 2 built-in (tray-icon feature) |
| Bundle targets | .exe (NSIS) on Windows; .deb / .rpm / .AppImage on Linux; .dmg on macOS (Apple Silicon) |
| What | Where |
|---|---|
| Chats, messages, spaces, snippets, snippet variables, MCP servers, app settings | SQLite at <app-data-dir>/loach.db |
| OpenAI API key | OS credential manager (Windows Credential Manager / Linux Secret Service / macOS Keychain) |
| App-lock hash + hint | OS credential manager — same store, separate entry |
| Attached files (images, text) | Inlined into the message at send time; no separate file store |
| App data dir on Windows | %APPDATA%\dev.loach.app\ |
| App data dir on Linux | ~/.local/share/dev.loach.app/ |
| App data dir on macOS | ~/Library/Application Support/dev.loach.app/ |
Loach launches and works completely offline as long as you stick to local providers. The Models library, Spaces, Snippets, search, parameter sidebar, app lock, and chat history are all available without network access. Only chat generations against remote endpoints (OpenAI, Groq, OpenRouter, …) require internet connection.
MIT © ZTCS
