Juno
A production-grade, multimodal AI platform and native client ecosystem.
Streaming Chat · Deep Research · Durable Memory · Canvas Artifacts · Model Context Protocol (MCP) · Realtime Voice Relay · Cloud & Device Code Agents · Native macOS & iOS Apps
Overview
What is Juno?
Juno is a modern, full-stack, multimodal AI platform designed for serious everyday work. It combines a high-performance web experience with dedicated native desktop (macOS) and mobile (iOS) applications, backed by a unified synchronization engine and strict security architecture.
Why does Juno exist?
Most open-source AI chat frontends are either lightweight web wrappers around single provider APIs or monolithic prototypes lacking production security, multi-device synchronization, or deep agentic workflows. Proprietary platforms, conversely, vendor-lock users into closed ecosystems with opaque data practices.
Juno fills this gap by delivering:
- Model & Provider Interoperability: Zero vendor lock-in across OpenAI, Anthropic, Google Gemini, xAI Grok, Mistral, Meta Llama, DeepSeek, Perplexity, Cohere, and local/custom endpoints.
- First-Class Native Experience: Native Swift applications for macOS and iOS that share contracts, tokens, and offline/online state synchronization with the web core.
- True Agentic Capabilities: Unified tool execution, Model Context Protocol (MCP) connectors, autonomous "Code" agent sessions, and persistent agentic teammates with goals and memory.
- Defense-in-Depth Security: Nonce-based CSP, cryptographic encryption-at-rest for user conversations, strict ownership query enforcement, and action-bound approval receipts.
Key Capabilities
- Unified Multi-Provider Generation: Seamlessly stream turns, visible thinking/reasoning blocks, and tool executions across 14+ model providers with automatic failover and reasoning effort ladders.
- Deep Research & Canvas Artifacts: Interactive document workspaces, live code previews (React, Tailwind, HTML, SVG, Mermaid), version history, and full-screen reading modes.
- Durable Memory & Personalization: Long-term memory distillation with encrypted summary profiles and user-controlled forgetting.
- Model Context Protocol (MCP) & Connectors: Dynamic discovery, OAuth registration, and tool calling across local and remote MCP servers (Linear, GitHub, Notion, Slack, Google Workspace, Apple ecosystem).
- Multimodal Realtime Voice: Low-latency audio streaming relay (
relay/) supporting duplex speech-to-speech, read-aloud synthesis, and continuous dictation.
- Code & Work Sessions: Sandboxed cloud execution environments (
runner/) and device-bridged agents capable of inspecting file systems, running bash commands, and checking execution outcomes.
- Native Ecosystem: Full-featured native clients for macOS (Menu Bar, Global Hotkey, Spotlight-like quick access) and iOS (Dynamic Island, Live Activities, Lock Screen widgets).
Architecture at a Glance
┌────────────────────────────────────────────────────────────────────────┐
│ Clients │
│ Web (Next.js 15) │ macOS (Swift/AppKit) │ iOS (Swift/SwiftUI) │
└───────────┬──────────────────────┬──────────────────────┬──────────────┘
│ │ │
▼ ▼ ▼
┌────────────────────────────────────────────────────────────────────────┐
│ Juno Backend Core │
│ • Next.js App Router (Modular API & Route Handlers) │
│ • NextAuth v5 + TOTP 2FA + Immediate Session Revocation │
│ • Ownership Query Guard (Mandatory User-Scoping) │
│ • Encrypted Transcript & Credential Storage (AES-256-GCM) │
└───────────┬──────────────────────┬──────────────────────┬──────────────┘
│ │ │
┌───────┴────────┐ ┌───────┴────────┐ ┌───────┴────────┐
▼ ▼ ▼ ▼ ▼ ▼
┌──────────────┐ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐
│ PostgreSQL │ │ S3 Storage │ │ Voice Relay │ │ Cloud Runner │
│ (Prisma │ │ (Attachments,│ │ (WebSocket │ │ (Agent Core, │
│ Migrations) │ │ Artifacts) │ │ Service) │ │ Sandboxed) │
└──────────────┘ └──────────────┘ └──────────────┘ └──────────────┘
Security Model
Juno implements defense-in-depth controls audited for production workloads:
This is the beginning of the README. Read the rest at the original. Read the full README ↗
