L Loach Docs
Ctrl K

Quick start

Loach has no built-in models. The very first thing to do is choose a provider: a local Ollama server, or any OpenAI-compatible HTTP endpoint.

First launch

The first time you open Loach, a short setup wizard does most of this page for you: it finds your Ollama server (and can start it), recommends and pulls a model sized for your machine, or verifies an OpenAI-compatible endpoint, then asks for your defaults, tools and custom instructions. Every step except the provider can be skipped, and anything you set there can be changed later in Settings. The sections below cover the same ground by hand.

Using with Ollama (local, recommended)

Install Ollama and start the server, then pull at least one model:

ollama serve
ollama pull gemma4:e4b      # or any other tag

Loach 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 customize existing ones from inside the app - no terminal required.

You don't have to run ollama serve yourself either. When nothing is answering, the model picker offers a Start Ollama button, and the Auto-launch Ollama switch in Settings → Providers starts it every time Loach opens. Both only work when Ollama is installed on this computer and the base URL points at it. See Providers.

Tip

Pick a model that fits your GPU. If you have no models yet, the onboarding wizard recommends one sized against your VRAM (or system RAM without a discrete GPU) and labels every catalog entry with a verdict - Runs well, Runs OK, Runs slowly, Too big for this machine or Not enough disk space. On a typical 8 GB card it suggests qwen3.5:4b; gemma4:e4b also fits, with less headroom. For lower-spec devices, see Low VRAM mode.

Using with an OpenAI-compatible endpoint

Open Settings → Providers, paste your API key (stored safely in your OS credential manager), and point the base URL to a compatible endpoint. Local servers such as LM Studio, vLLM or llama.cpp usually need no key at all. The menu next to the base URL field fills in the common ones:

ProviderBase URL
OpenAIhttps://api.openai.com/v1 (default)
llama.cpp llama-serverhttp://localhost:8080/v1
LM Studiohttp://localhost:1234/v1
vLLMhttp://localhost:8000/v1
LiteLLMhttp://localhost:4000
Otherper 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.

Your first chat

  1. Open Loach. The sidebar lists your chats; the main area is the composer.
  2. Open the model dropdown in the chat header. It lists the models of both providers, grouped under Ollama and API.
  3. Pick a model.
  4. Type a prompt and press Enter.

That's it. Tokens stream in, generation stats appear under each model turn, and you can copy code blocks or open them in the canvas with a single click. Type / at the start of the composer for the command palette - /fork, /regenerate, /model and more - and keep an eye on the context bar just below the message box as the conversation grows.

Next steps

Previous ← Installation