Jin

A native macOS app for working with multiple LLM providers from one conversation workspace. Conversation-first, tool-aware, and built for macOS instead of Electron.

Why Jin

• Conversation-first — Built around chat flow and fast iteration, not workspace clutter
• Provider-aware controls — Exposes model-specific settings like reasoning, search, caching, PDF handling, service tier, code execution, and media options instead of hiding them behind generic presets
• Native macOS workflow — SwiftUI app with drag-and-drop, split views, custom keyboard shortcuts, recovery tools, and Sparkle updates
• One app for chat + tools — Chat, MCP, built-in search, artifacts, voice, coding workflows, and media generation live in the same workspace

Features

• Multi-provider chat — Work across OpenAI, Anthropic, Gemini, Vertex AI, xAI, gateway providers, and more from one app
• Parallel multi-model chat — Add up to 3 models to one conversation, compare responses side-by-side, and keep per-model context independent
• Multimodal threads — Mix text, images, files, audio, PDFs, and generated media in one conversation
• Markdown, code, and LaTeX rendering — Syntax-highlighted code blocks, markdown rendering, inline/block math, and copy-friendly output
• Reasoning and advanced model controls — Per-chat controls for reasoning budget, web search, prompt caching, PDF mode, OpenAI service tier, and other provider-specific options
• Search and grounding — Provider-native web search plus built-in search plugins, source cards, citation timeline, and Google Maps grounding for Gemini / Vertex AI
• MCP tool calling — Connect external tools and data through MCP servers over stdio or HTTP, with persistent or ephemeral server lifecycles
• Agent and coding workflows — Codex App Server support with per-chat working directory, sandbox mode, and personality controls, plus optional local Agent Mode for shell/file/search tools through the bundled RTK helper
• Provider-native code execution — Run supported model-side code execution flows with visible activity timeline, logs, generated images, and downloadable files
• Artifacts workspace — Inline HTML, React, and ECharts artifacts with split-pane preview, version history, source export, and reusable artifact IDs
• Image and video generation — Image generation/editing across OpenAI (GPT Image / DALL·E) and xAI; video generation across xAI, Google Veo (Gemini & Vertex), and OpenRouter SeedDance, including source-image and source-URL edit workflows where available
• PDF and OCR handling — Native PDF upload where supported, with Mistral OCR / MinerU / DeepSeek / OpenRouter / Firecrawl as cloud OCR fallbacks plus local macOS extraction
• Voice workflows — Speech-to-text and text-to-speech via cloud providers or on-device WhisperKit / TTSKit models, with a floating mini-player for chat playback
• Assistants and defaults — Named assistants with custom prompts, language preference, model defaults, temperature/output settings, and optional history truncation
• Quoted reply blocks — Pin a snippet from any message into the composer as a structured quote so the next turn keeps the reference highlighted in the thread
• macOS polish — Configurable shortcuts, drag-and-drop attachments, storage breakdown with chat diagnostics, recovery pack export/import, and in-app Sparkle update controls

Screenshots

Supported Providers

Configure providers in Settings > Providers. Jin supports direct providers, gateways, and coding-focused runtimes:

OpenAI · OpenAI (WebSocket) · Anthropic · Claude Managed Agents · Gemini (AI Studio) · Vertex AI · xAI · Perplexity · Groq · Cohere · Mistral · DeepInfra · Together AI · Fireworks · SambaNova · Cerebras · DeepSeek · Zhipu Coding Plan · MiniMax · MiniMax Coding Plan · Xiaomi MiMo Token Plan · Zyphra · MorphLLM · OpenCode Go · GitHub Copilot · OpenRouter · OpenAI Compatible · Cloudflare AI Gateway · Vercel AI Gateway · Codex App Server (Beta)

• Most providers use an API key.
• Vertex AI uses a service account JSON.
• Claude Managed Agents uses an Anthropic API key and routes through Anthropic's managed-agent runtime (custom tools, persistent sessions, prompt caching).
• Codex App Server can use an API key, ChatGPT account login, or local Codex auth from $CODEX_HOME / ~/.codex. Jin can also launch a localhost codex app-server for you from provider settings. Recommended runtime: codex 0.107.0+.
• Xiaomi MiMo Token Plan ships in two flavors — Anthropic-compatible and OpenAI-compatible — so you can use whichever upstream API surface a given MiMo model expects.
• Gateway providers such as OpenRouter, Cloudflare AI Gateway, Vercel AI Gateway, and OpenAI Compatible can route upstream models while still benefiting from Jin's model metadata when the exact upstream model ID is known.

Model Catalog

• Jin ships with curated seed models for major providers so you can start chatting immediately after adding credentials.
• Use Fetch Models to pull fresh provider model lists when the provider supports catalog fetching.
• You can also add model IDs manually, including gateway-prefixed IDs like openai/... and anthropic/....
• Known models use exact-ID capability metadata for context window, reasoning behavior, vision, web search, PDF handling, code execution, image/video generation, and other features.
• Unknown model IDs still work, but Jin falls back to conservative defaults until metadata is available.

Plugins

All plugins are optional and configured in Settings > Plugins.

Plugin	Services
Web Search	Exa, Brave Search, Jina Search, Firecrawl, Tavily, Perplexity Search
Text-to-Speech	OpenAI, OpenRouter, Groq, ElevenLabs, Xiaomi MiMo, TTSKit (on-device)
Speech-to-Text	OpenAI, OpenRouter, Groq, Mistral, ElevenLabs, WhisperKit (on-device)
PDF OCR	Mistral OCR, MinerU, DeepSeek (via DeepInfra), OpenRouter OCR, Firecrawl OCR — used as fallback when the active model can't natively ingest PDFs
Chat Naming	Automatic conversation naming with a selected model
Cloudflare R2 Upload	Upload local videos to R2 and use public URLs in video workflows
Agent Mode	Local shell/file/search tools via the bundled RTK helper and local file operations

MCP (Model Context Protocol)

Connect MCP servers under Settings > MCP Servers.

• Supports persistent and ephemeral server lifecycles
• Supports stdio and HTTP transports, including streaming HTTP setups
• Supports server presets plus mcpServers JSON import
• Keeps per-server tool enablement and per-chat MCP selection separate
• Supports HTTP authentication with bearer token or custom headers

Installation

Download

Download the latest release from the Releases page.

• Release assets typically include Jin.zip, and release automation may also publish Jin.dmg
• Jin uses Sparkle for in-app updates after installation
• Current packaged builds are Apple Silicon-only
• If your release is zipped, unzip it first

If macOS blocks the app

If macOS shows a warning like "is damaged and can't be opened" or "Apple could not verify":

1. Move Jin.app to /Applications.
2. Right-click Jin.app and choose Open once.
3. If it is still blocked, open System Settings > Privacy & Security.
4. Click Open Anyway for Jin, then confirm Open.
5. If needed, clear quarantine and retry:

xattr -dr com.apple.quarantine /Applications/Jin.app

Requirements

• Apple Silicon Mac for packaged release builds
• macOS 14 (Sonoma) or later

Getting Started

1. Launch Jin.
2. Open Settings > Providers and add a provider credential. Most providers use API keys, Vertex AI uses a service account JSON, and Codex App Server can be configured without an API key if you use ChatGPT or local Codex auth.
3. Start a new conversation and pick one model, or add up to 3 models to the same chat.
4. Optional: enable plugins in Settings > Plugins for search, OCR, voice, cloud upload, or local Agent Mode.
5. Optional: add MCP servers in Settings > MCP Servers for tool calling.
6. Optional: customize General settings for appearance, keyboard shortcuts, chat defaults, updates, and data / recovery tools.

Build from Source

git clone https://github.com/hrayleung/Jin.git
cd Jin
swift build
swift test
swift run Jin                 # Run from the command line (Debug)
open Package.swift           # Open in Xcode
bash Packaging/package.sh    # Build arm64 .app bundle and create dist/Jin.zip
bash Packaging/package.sh dmg

Packaging/package.sh also bundles the RTK helper used by Agent Mode into the final app bundle.

Requires Swift 5.9+ / Xcode 15+.

Updates & Release

Jin uses Sparkle for in-app updates.

• Settings > General > Updates exposes automatic update checks and optional pre-release channel opt-in
• Update feed and signing key are configured in Packaging/Info.plist (SUFeedURL, SUPublicEDKey)
• The appcast lives at docs/appcast.xml
• CI packaging, signing, notarization, and release publishing live in .github/workflows/ci-cd.yml

Contributing

Contributions are welcome. Unless explicitly stated otherwise, contributions are accepted under the same Apache License 2.0.

License

Apache License 2.0 — permissive open-source licensing with an express patent grant. See the full license text for details.

Acknowledgments

• MCP Swift SDK — Model Context Protocol client library
• Sparkle — In-app update framework for macOS
• Alamofire — Networking primitives
• WhisperKit — On-device speech-to-text
• Kingfisher — Image loading and caching
• swift-collections — Ordered/persistent collection types
• Lobe Icons — Provider icon assets