Features

A reference for the features Loach ships with today. We're continuously extending this list — including new RAG and agentic capabilities. Pick a category below, or open any feature for the full description.

Chat experience

• Provider selectionSwitch between Ollama local models and any OpenAI-compatible endpoint (OpenAI, llama.cpp, LM Studio, vLLM, Groq, OpenRouter…) directly from the chat header — no restart required.
• Local model managementPull, copy, customize and delete local models from inside the app — no terminal required. The Models library tab is a full UI over the Ollama HTTP API.
• Per-chat parametersSet temperature, top_k, top_p, min_p, max_tokens, context length and per-chat system prompts. Values are layered over Modelfile defaults and per-model defaults.
• Personas and tonesPick a role (Code Reviewer, Translator, ELI5…) and a delivery style (Direct, Detailed, Playful, Skeptical, Socratic…). They layer on top of your custom instructions.
• Custom instructionsSet instructions globally, per-Space or per-chat — with template variables for your name, the date, time, weekday and timezone. Narrower scope wins.
• Generation statsAn inline tok/s · total tokens · elapsed line under every model turn so you always know how the model performed.
• Markdown renderingFull GitHub-flavoured Markdown with tables, task lists, footnotes and LaTeX support — rendered live as the response streams. Thinking traces fold into their own collapsible block.
• Code blocks & canvasCode blocks are syntax-highlighted with line numbers and offer copy, export and open in canvas options. The code canvas opens any code block in a wider, theme-aware view that's easier to read and review.
• Concurrency & queueOne generation at a time across all chats, with a FIFO queue and a Respond now action for jumping the line. Cancelled streams preserve their partial output.
• Private ChatA dark-only, ephemeral chat surface. Nothing is written to disk, MCP tools are blocked, and only local Ollama models are offered — the transcript wipes the moment the overlay closes.

Organization

• SpacesGroup chats around a project, with shared instructions, reference files, memory and pinned model defaults. Every chat inside the Space inherits the context.
• SnippetsSave reusable prompts with an optional pinned model and click Run to start a fresh chat pre-filled and ready to send.
• Chat archiveMove chats out of the sidebar without deleting them. Restore or delete them from a dedicated archive view.
• SearchA global Ctrl + K palette across chats, Spaces and Snippets, plus a browser-style in-chat finder with phrase highlighting.
• Import and export contextExport chat context to JSON or Markdown, and paste exported data — or any text — back into any chat's context.

Input & output

• AttachmentsDrag and drop images, code, PDFs and DOCX files into the composer. Loach parses each one and inlines its content into the request.
• Voice dictationA mic button in the composer turns speech into text in real time. Useful for long prompts and hands-free use.
• Web fetchDrop a URL into a message. Loach fetches the page, sanitizes it, and inlines the content into the request — no separate browser extension needed.
• Temporal awarenessOptionally inject current date, time, weekday and timezone into the system prompt, so models can answer "what day is it today?" correctly.

Advanced

• MCP supportRegister Model Context Protocol servers (Streamable HTTP), test the handshake and inspect the tools they provide. Once registered, tools become available to the model during a chat — bringing structured external capabilities into your conversation.
• Default model selectorPick a model new chats open with — per provider. No need to re-select on each fresh conversation.
• Model preloadingOptionally warm your default local model into VRAM at launch, so the first message streams faster.
• Low VRAM modeGlobal or per-chat toggle that sends Ollama's low_vram flag with every request — useful on lower-spec devices.

Security & data

• App lockOptional PIN, password or PIN + password gate at launch. Credentials are hashed with a strong KDF and stored in the OS credential manager, never on disk in plaintext.
• Data managementMake a backup of your content to a JSON file, restore data, or permanently delete everything with a few clicks.

Appearance

• ThemesTwo themes — Aurora (glassy, gradient) and Solid (flat) — both available in Dark and Light variants. Three font sizes scale every text size across the app.

Updates

• OTA updatesGet new features, bug fixes, performance improvements and security patches directly from the app — no need to manually download a new installer.

Reference

• Onboarding wizardThe six-step first-launch wizard: welcome, name, provider, prompt, features and start. Re-runs after a factory reset.
• Keyboard shortcutsGlobal search, in-chat find, composer send and newline behaviour, palette navigation — at a glance.