Skip to content
BranchesTags

Repository files navigation

Moraya — The Elegantly Minimal Markdown/Typst AI Agent for the Local AI Era

Discover Moraya, the epitome of elegant simplicity in an open-source Markdown/Typst AI agent that bridges cutting-edge AI power with unwavering user sovereignty—empowering you to create, collaborate, and innovate in total privacy.

🎉 Announcing Moraya Web — Now Live

Moraya is now in your browser: web.moraya.app — no install, open a tab and start writing.

Moraya Web carries the same elegant WYSIWYG Markdown/Typst editing experience to the web, and extends it into an AI-native knowledge workspace:

  • Same editor core — instant WYSIWYG, three editing modes, outline, find & replace, powered by the shared @moraya/core engine for pixel-perfect rendering parity with the desktop app.
  • Your creation, your keys, your AI — Bring Your Own Cloud (BYOC) storage (Aliyun OSS, AWS S3, Cloudflare R2, Tencent COS, Backblaze B2) and end-to-end encryption with client-side keys; your content stays yours.
  • AI everywhere — multi-provider AI chat, AI agents over your documents, and voice-first capture with real-time transcription and AI structuring.
  • Works offline — installable as a PWA with autosave and an offline buffer, so your writing never waits for the network.

Prefer a fully local, ~10MB native app? The desktop version below remains free and open-source — same Moraya, your choice of where it runs.

Moraya is a free, open-source, ultra-lightweight (~10MB) editor crafted with Rust and Tauri v2 — WYSIWYG for Markdown, source + live preview for Typst — with local AI and MCP built in as a privacy-first "Personal Assistant" that never leaves your device. Derived from "mora" (Latin, "a moment") and "ya" (Chinese, "elegance").

User Manual / Wiki

Why Moraya? Key Advantages

  • Ultra-Lightweight & Native Performance — ~10MB installer, instant launch, tiny memory footprint.
  • True Instant WYSIWYG — Type # and see a heading instantly (ProseMirror-powered).
  • Two Authoring Formats, One App — Markdown WYSIWYG plus native Typst (.typ) authoring with a live compiled preview, both exportable to a genuinely typeset PDF.
  • Most Powerful Local AI Integration — Multi-provider streaming chat (Claude, OpenAI, Gemini, DeepSeek, Ollama, custom endpoints), 71+ AI templates across 10 categories, AI image generation, and smart writing commands.
  • Leading MCP Ecosystem — Dynamic MCP container, one-click Marketplace (Official, LobeHub, Smithery), autonomous local AI services, tool calling, and custom agent workflows — all fully self-hosted.
  • Complete Modern Workflow — Visual/Source/Split modes, publishing tools, SEO assistant, AI images, and automatic RSS feeds.
  • Security by Design — API keys stored in OS Keychain, all API calls proxied through Rust backend, CSP enforcement, path traversal protection. Everything can run offline with local models; your data never leaves your machine.

Features

Editor

  • Three Editor Modes — Visual (WYSIWYG), Source (raw Markdown), Split (synced side-by-side with block-level scroll anchoring). Toggle with Cmd+/ or Ctrl+/.
  • Full Markdown Support — CommonMark + GFM: tables with floating toolbar, task lists, strikethrough, emoji, definition lists.
  • Math Rendering — Inline and block LaTeX via KaTeX.
  • Code Blocks — Syntax highlighting, language selector dropdown (25+ languages), one-click copy, hover toolbar.
  • Mermaid Diagrams — 9 diagram types (flowchart, sequence, gantt, state, class, ER, pie, mindmap, journey) with edit/preview dual mode, lazy-loaded rendering (~1.2MB loaded only on first use), and automatic theme adaptation.
  • Image Tools — Floating toolbar for resizing, right-click context menu, drag-and-drop.
  • Sidebar File Explorer — Directory memory across sessions, real-time file refresh, list/tree dual views, right-click context menu (new, rename, delete), and full-text file search across the open folder.
  • Find & Replace — Full-text search and replace within documents.

Typst Support

Typst is a compiler, not a WYSIWYG format, so .typ documents get their own authoring surface — deliberately distinct from the Markdown editor above.

  • Native .typ Authoring — Opening or creating a .typ file switches to a dedicated Typst editor: raw source on one side, a live compiled preview on the other (Overleaf-style). Visual/Source/Split modes apply here too — Visual shows the rendered pages, Source shows raw .typ text.
  • Typst-Compiled Export — Export a .typ document as a native PDF, per-page PNG, or semantic HTML — all produced by the real Typst compiler, not a DOM screenshot.
  • Typeset PDF from Markdown, too — Any Markdown document can also be exported through the Typst engine (via the cmarker package) for a genuinely typeset PDF, alongside the existing native PDF export path.
  • Same Menus, Same Shortcuts — Headings, bold/italic/strikethrough, inline code, links, images, lists, code/math blocks, quotes, rules and tables share ONE menu item and ONE shortcut across both formats; each emits the right markup for the document you are in (= Heading / *bold* / #quote(block: true)[…] in Typst). Actions with no counterpart in the other format (Task List, cloud audio/video) are greyed out rather than hidden, so the menu never shifts shape.
  • Zero-Cost When Unused — The typst compiler isn't bundled into the ~10MB binary. It downloads on first use from the official typst/typst release and is cached locally — Markdown-only users pay nothing for it.

AI-Powered Writing

  • Multi-Provider Support — Claude, OpenAI, Gemini, DeepSeek, Grok, Mistral, GLM, MiniMax, Doubao, Ollama, and any OpenAI-compatible endpoint. Multi-model configuration with active/inactive switching.
  • 71+ AI Templates — 10 categories (Writing, Translation, Student, Kids, Marketing, Professional, Personal, Chinese Games, English Games, Quiz) with 5 flow types (auto, input, selection, parameterized, interactive).
  • Streaming Chat Panel — Real-time AI responses with insert/replace/copy actions.
  • Vision / Multimodal Input — Paste, drag-and-drop, or pick images to include in AI conversations. Auto-compression for oversized images; thumbnail preview with lightbox viewer. Compatible with Claude, OpenAI, Gemini, and Ollama vision models.
  • AI + MCP Tool Integration — LLM can call MCP tools with auto-retry loop, enabling autonomous AI workflows.
  • AI Image Generation — 5 modes (article, design, storyboard, product, moodboard) × 10 styles each, with 7 aspect ratios and 3 resolution levels. Supports OpenAI DALL-E, Grok, Gemini Imagen, Qwen, Doubao, and custom providers.

AI Voice Transcription

  • Real-Time Speech-to-Text — Stream microphone audio to Deepgram, Gladia, AssemblyAI, or Azure Speech Services with sub-second transcription latency.
  • Speaker Diarization — Automatically distinguish and label multiple speakers per session using pitch-based gender detection, with support for custom naming.
  • Voiceprint Archive — Cross-session speaker recognition via stored voice profiles; sample audio is captured automatically during recording and capped at 30 seconds per profile.
  • Transcription Panel — Color-coded per-speaker segments, one-click AI meeting summary generation, and Markdown export directly into the editor.
  • Voice Settings — Per-provider key management via OS Keychain, test-connection verification, and voice profile CRUD with playback preview.

MCP Ecosystem

  • Three Transports — stdio, SSE, and HTTP for maximum compatibility.
  • Marketplace — Browse and one-click install MCP servers from 3 data sources (Official Registry, LobeHub, Smithery).
  • Dynamic MCP Container — AI can create MCP services on-the-fly with a lightweight Node.js runtime. 4 internal tools: create, save, list, and remove services.
  • Built-in Presets — Filesystem, Fetch, Git, Memory one-click setup.
  • Claude Desktop JSON Import — Paste mcpServers JSON config to auto-add servers.
  • Knowledge Base — Multi-knowledge-base management with quick-switch dropdown and per-KB AI behavior rules via MORAYA.md (automatically injected into AI context). Sync KB content with MCP servers for context-aware AI.

Publishing Workflow

  • Multi-Target Publishing — Publish to GitHub repos and custom APIs with front matter and file naming templates.
  • SEO Assistant — AI-generated titles, excerpts, tags, slug, and meta descriptions.
  • Image Hosting — Auto-upload to SM.MS, Imgur, GitHub, Qiniu Kodo, Aliyun OSS, Tencent COS, AWS S3, Google GCS, or custom providers. HMAC request signing for object storage handled in Rust backend.
  • RSS Feed — Auto-update RSS 2.0 feed on publish (zero-dependency XML generation).

Plugin System

  • Decentralized Registry — GitHub-based open registry; no central server required. Community plugins hosted and distributed as standard GitHub repositories.
  • Plugin API v1 — Hook into editor commands, AI chat, AI image generation, and voice transcription workflows via a versioned JavaScript API.
  • Marketplace — Browse, install, and update plugins with one-click install, real-time GitHub release data, and zero-configuration setup.
  • Supply Chain Security — SHA256 version pinning and per-plugin permission model; plugins declare required capabilities upfront.

Security

  • OS Keychain Storage — API keys stored in native secure storage (macOS Keychain, Windows Credential Manager, Linux Secret Service). The OS may prompt for your system password when Moraya first accesses the Keychain — this is the operating system verifying your identity before granting access to securely stored credentials, not Moraya itself requesting a password.
  • Rust AI Proxy — All external API calls routed through Rust backend; keys never exposed in WebView.
  • CSP Enforcement — script-src 'self', connect-src locked to IPC and localhost.
  • MCP Hardening — Command validation, startup confirmation dialogs, environment variable filtering, zombie process prevention, buffer limits.
  • Path Traversal Protection — All file operations validate and canonicalize paths.
  • HTML Export Sanitization — DOMParser-based XSS prevention on export.

Privacy

  • Bring Your Own Key (BYOK) — You provide your own API keys. Keys are stored exclusively in your OS's native secure storage (macOS Keychain / Windows Credential Manager / Linux Secret Service), encrypted at rest, and never transmitted to any Moraya server.
  • No Intermediary Servers — AI prompts and content are sent directly from your device to the provider's API. Moraya does not operate any relay or proxy servers — the data path is simply: Your Device → Provider API. Authentication is injected on-device by the local Rust backend before any request leaves your machine.
  • Full Privacy Policy — Available in-app via Help → Privacy Policy, or at privacy-policy.md.

Platform & UI

  • Cross-Platform — macOS, Windows, Linux, and iPadOS via Tauri v2. iPad builds distributed via TestFlight with Tab bar multi-file editing, floating touch toolbar, and Magic Keyboard shortcut support.
  • Frameless Window — Custom title bar with native macOS traffic lights.
  • Multi-Window — Multiple editor windows with macOS Dock right-click menu support.
  • Auto-Update — Silent daily update checks with one-click install.
  • Native Menus & Shortcuts — Full platform-native menus (File, Edit, Paragraph, Format, View, Help).
  • Themes — Light, Dark, and system-sync modes.
  • Internationalization — English & Simplified Chinese with auto-detection.
  • Export — HTML, native PDF (vector, selectable text), and LaTeX built-in for Markdown; PDF/PNG/HTML via the Typst compiler for .typ documents. DOCX/EPUB planned.

Architecture Overview

┌────────────────────────────────────────────────────────┐
│              Tauri WebView (Frontend)                  │
│        Svelte 5 + ProseMirror + TypeScript             │
│                                                        │
│  ┌───────────┐ ┌───────┐ ┌──────────┐ ┌───────────┐    │
│  │  Editor   │ │  AI   │ │ Settings │ │  Voice /  │    │
│  │ProseMirror│ │ Panel │ │  Panel   │ │  Publish  │    │
│  │ + Source  │ │       │ │ (9 tabs) │ │  Plugin   │    │
│  └─────┬─────┘ └──┬────┘ └────┬─────┘ └────┬──────┘    │
│        │          │           │            │           │
│  ┌─────┴──────────┴───────────┴────────────┴───────┐   │
│  │              Services & Stores                  │   │
│  │  (file, AI, MCP, voice, publish, plugin, i18n)  │   │
│  └───────────────────┬─────────────────────────────┘   │
│                      │ Tauri IPC (invoke)              │
└──────────────────────┼─────────────────────────────────┘
                       │
┌──────────────────────┼─────────────────────────────────┐
│              Rust Backend (Tauri)                      │
│                                                        │
│  ┌─────────┐ ┌──────────┐ ┌──────────┐ ┌───────────┐   │
│  │ File I/O│ │ AI Proxy │ │ MCP Proc │ │  Speech   │   │
│  │Commands │ │ HTTP/SSE │ │ Manager  │ │  Proxy    │   │
│  └─────────┘ └──────────┘ └──────────┘ └───────────┘   │
│  ┌─────────┐ ┌──────────┐ ┌──────────┐ ┌───────────┐   │
│  │Keychain │ │  Object  │ │  Plugin  │ │   Menu    │   │
│  │(keyring)│ │ Storage  │ │ Manager  │ │           │   │
│  └─────────┘ └──────────┘ └──────────┘ └───────────┘   │
└────────────────────────────────────────────────────────┘

Tech Stack

Layer Technology Version
Runtime Tauri v2 >=2.9,<2.10
Backend Rust 2021 edition
Frontend Svelte 5 + SvelteKit (SPA via adapter-static) ^5.0.0 / ^2.9.0
Editor Engine ProseMirror (direct) + prosemirror-markdown latest
Math Rendering KaTeX ^0.16.28
Diagrams Mermaid (lazy-loaded) ^11.x
Typst Engine typst CLI (on-demand download, not bundled) 0.15.1
Build Tool Vite ^6.0.3
Package Manager pnpm 10.x
Language TypeScript (strict mode) ~5.6.2

Install

macOS (Homebrew)

brew tap zouwei/moraya
brew install --cask moraya

Upgrade: brew upgrade --cask moraya · Uninstall: brew uninstall --cask moraya

All Platforms

Download the latest release from GitHub Releases.

Platform File Install
macOS (Apple Silicon) Moraya_x.x.x_mac_aarch64.dmg Signed & notarized — drag to Applications, double-click to launch
macOS (Intel) Moraya_x.x.x_mac_x64.dmg Signed & notarized — drag to Applications, double-click to launch
Windows Moraya_x.x.x_win_x64_en-US.msi Run the MSI installer
Linux (Debian) moraya_x.x.x_linux_amd64.deb sudo dpkg -i moraya_*.deb
Linux (AppImage) Moraya_x.x.x_linux_amd64.AppImage chmod +x then run

Getting Started

Prerequisites

Development

# Install dependencies
pnpm install

# Start dev server with hot reload
pnpm tauri dev

# Frontend only (no Tauri shell)
pnpm dev

Build

# Full production build (frontend + Rust + bundle)
pnpm tauri build

# Type checking
pnpm check

# Rust only
cd src-tauri && cargo check

Keyboard Shortcuts

Action macOS Windows/Linux
New Cmd+N Ctrl+N
New Window Cmd+Shift+N Ctrl+Shift+N
Open Cmd+O Ctrl+O
Save Cmd+S Ctrl+S
Save As Cmd+Shift+S Ctrl+Shift+S
Settings Cmd+, Ctrl+,
Find Cmd+F Ctrl+F
Replace Cmd+H Ctrl+H
Toggle Visual/Source Cmd+/ Ctrl+/
Toggle Split Mode Cmd+Shift+/ Ctrl+Shift+/
Toggle Sidebar Cmd+\ Ctrl+\
Toggle AI Panel Cmd+Shift+I Ctrl+Shift+I
Toggle Outline Cmd+Shift+O Ctrl+Shift+O
Quick Open Cmd+P Ctrl+P
Command Palette Cmd+Shift+P Ctrl+Shift+P
Export HTML Cmd+Shift+E Ctrl+Shift+E
Heading 1–6 Cmd+1–6 Ctrl+1–6
Bold Cmd+B Ctrl+B
Italic Cmd+I Ctrl+I
Strikethrough Cmd+Shift+X Ctrl+Shift+X
Inline Code Cmd+E Ctrl+E
Link Cmd+K Ctrl+K
Insert Image Cmd+Shift+G Ctrl+Shift+G
Code Block Cmd+Shift+K Ctrl+Shift+K
Quote Cmd+Shift+Q Ctrl+Shift+Q
Ordered List Cmd+Option+O Ctrl+Alt+O
Bullet List Cmd+Option+U Ctrl+Alt+U
Task List Cmd+Option+X Ctrl+Alt+X
Zoom In/Out/Reset Cmd+=/-/0 Ctrl+=/-/0
AI Send Message Cmd+Enter Ctrl+Enter

Markdown & Typst: editing shortcuts are shared by both document formats — the same binding emits the right markup for whichever format you are in. Settings ▸ Shortcuts groups every binding into Markdown / Typst shared, Markdown only and Typst only, and all of them are customizable.

AI Chat Input: Enter inserts a newline; Cmd+Enter / Ctrl+Enter sends the message. This avoids conflicts with CJK IME composition.

AI Configuration

Open Settings (Cmd+, / Ctrl+,) and select the AI and Voice tab. Configuration is split into three independent sections.

Chat Providers

Provider API Key Default Models
Anthropic Claude Yes claude-opus-4-6, claude-sonnet-4-6, claude-haiku-4-5-20251001
OpenAI Yes gpt-5.2, gpt-5.2-pro, gpt-5, gpt-5-mini, o4-mini, gpt-4o, gpt-4o-mini, o3, o3-mini
Google Gemini Yes gemini-3.1-pro-preview, gemini-3-flash-preview, gemini-2.5-flash, gemini-2.5-flash-lite
DeepSeek Yes deepseek-chat, deepseek-reasoner
Grok (xAI) Yes grok-4, grok-4-1-fast-reasoning, grok-4-1-fast-non-reasoning, grok-code-fast-1, grok-3
Mistral AI Yes mistral-large-latest, mistral-small-latest, magistral-medium-latest, codestral-latest, devstral-latest
GLM (Zhipu AI) Yes glm-5, glm-4-plus, glm-4-air, glm-4-flash, glm-z1-flash, glm-z1-air
MiniMax Yes MiniMax-M2.5, MiniMax-M2.5-highspeed, MiniMax-Text-01
Doubao (ByteDance) Yes doubao-seed-2-0-pro, doubao-seed-2-0-lite, doubao-seed-2-0-mini, doubao-seed-2-0-code
Ollama (Local) No llama3.3, llama3.2, qwen2.5, qwen2.5-coder, phi4, gemma3, deepseek-r1, mistral, codellama
Custom API Optional Any OpenAI-compatible endpoint

Image Generation Providers

Provider API Key Models
OpenAI Yes dall-e-3, dall-e-2, gpt-image-1
Grok (xAI) Yes aurora
Google Gemini Yes imagen-3.0-generate-002, imagen-3.0-fast-generate-001
Qwen (Alibaba) Yes wanx2.1-t2i-turbo, wanx2.1-t2i-plus, wanx-v1
Doubao (ByteDance) Yes doubao-seedream-5-0-260128, doubao-seedream-3-0-t2i-250415
Custom API Optional Any OpenAI-compatible image endpoint

Voice (Speech-to-Text) Providers

Provider API Key Models
Deepgram Yes nova-3, nova-2, nova, enhanced, base
Gladia Yes solaria-1, fast, accurate
AssemblyAI Yes universal-streaming
Azure Speech Yes latest (region required)
AWS Transcribe Yes (AWS credentials) general, medical, call-center (region required)
Custom Optional Custom WebSocket endpoint

Built-in Custom WebSocket protocol adapters:

Endpoint Pattern Protocol Audio Transport Notes
*.dashscope*.aliyuncs.com/api-ws/v1/inference DashScope FunASR Binary PCM Supports same-protocol regional nodes (e.g. Beijing/Singapore), auto run-task / finish-task
asr.cloud.tencent.com/asr/v2/ Tencent Cloud ASR v2 Binary PCM Sends {"type":"end"} on stop, parses result.voice_text_str
iat-api*.xf-yun.com/v2/iat iFLYTEK IAT v2 JSON base64 frames First/middle/last frame protocol (status 0/1/2); set APPID in model or URL app_id
api.openai.com/v1/realtime OpenAI Realtime input_audio_buffer.append Auto session.update, commits on stop, parses transcription delta/completed
ai-gateway.vei.volces.com/v1/realtime Volcengine Realtime (VEI Gateway) input_audio_buffer.append Auto transcription_session.update, parses delta/result/completed

All API keys are stored exclusively in your OS Keychain — never in plaintext. Click Test Connection in each section to verify before use.

About

An extremely elegant open-source Markdown/Typst AI Agent, Bridging AI capabilities and user sovereignty.

moraya.app

Resources

Readme
Apache-2.0 license
Activity

Stars

337 stars

Watchers

1 watching

Forks

31 forks
Report repository

Releases

Packages

Contributors

Languages