Joyus AI - Open-Source Multi-Tenant AI Agent Platform

Joyus AI is an open-source platform for deploying AI agents with skills-based mediation, content intelligence, and multi-tenant governance. Founded by Zivtech.

Overview

Most AI deployments are undifferentiated: the same model, the same defaults, the same outputs for every user and every organization. Joyus AI inverts that by making organizational knowledge a first-class platform primitive.

Core ideas:

• Skills as encoded knowledge - organizational standards, voice guidelines, domain rules, and workflow constraints are packaged as skills that constrain and guide AI outputs
• Content intelligence - writing profiles built from real corpora enable attribution, fidelity monitoring, and voice-consistent generation
• Open core, private skills - the platform is open source; client- and org-specific skills live in private repos and are loaded at runtime
• MCP-native - all agent capabilities are exposed via the Model Context Protocol, making them composable with Claude and other MCP-aware tools

Architecture

graph TD
    A[Claude / AI Client] -->|MCP| B[MCP Server<br/>Express]
    A -->|MCP| C[State Package]
    B --> D[Skills · Profiles · Session State]
    C --> D

Loading

• The MCP server is the primary interface between AI clients and platform capabilities.
• The state package maintains session continuity across Claude sessions and compactions.
• The content intelligence system (writing profiles, fidelity verification, drift monitoring) is maintained in a separate private package.
• Skills are modular prompt fragments loaded at runtime; org-specific skills live outside this repo.
• Additional capabilities — including structured knowledge capture, interactive research tooling, and artifact lifecycle management — are under active development in private repositories.

Packages

joyus-ai-mcp-server/ - Remote MCP Server

Express-based MCP server hosting platform tools and operator-defined skills. Connects to external services (issue trackers, version control, messaging) and exposes them as MCP tools.

• TypeScript / Node.js / Express
• Drizzle ORM + PostgreSQL for persistent state
• Deployable via Docker Compose on any cloud VM

joyus-ai-state/ - Session State

Maintains working state across Claude sessions. Captures git context, open files, decisions, and test status so Claude can restore context at session start without manual re-orientation.

MCP tools exposed:

• get_context - restore working state at session start
• save_state - snapshot after significant actions
• verify_action - pre-commit guardrails
• check_canonical - route to authoritative document copies
• share_state - share context with teammates

web-chat/ - Chat UI

Minimal browser-based chat interface for local development and demonstration. Not intended for production use.

Getting Started

Prerequisites: Node.js 20+, Docker (optional)

MCP Server

cd joyus-ai-mcp-server
npm install
cp .env.example .env   # configure database and service credentials
npm run build
npm start

State Package

cd joyus-ai-state
npm install
npm run build
# Add joyus-ai-mcp (MCP server binary) to your Claude Desktop / Code MCP config

Full Stack (Docker)

Production deployment configuration is maintained in a separate private repository. For local development, see the per-package instructions above.

Specs and Development

This project uses Spec Kitty for spec-driven development. Feature specifications live in kitty-specs/.

Current status snapshot (source: python scripts/pride-status.py on 2026-02-23):

Spec	Description	Status
001	MCP Server AWS Deployment	Complete
002	Session Context Management	Complete
003	Platform Architecture Overview	Spec-Only
004	Workflow Enforcement	Complete
005	Content Intelligence (Profile Engine)	Complete (Phases A–C, WP01–WP14)
006	Content Infrastructure	Complete (WP01–WP12)
007	Org-Scale Agentic Governance	Planning

Project-level architecture decisions, implementation plan, and constitution are in spec/.

Branch Protection

The default branch (main) is protected by a GitHub ruleset that requires pull request reviews from code owners, passing status checks, and prevents force pushes and branch deletion. The ruleset configuration is stored in .github/ruleset-default.json.

To apply the ruleset to a new fork or repository:

gh api repos/OWNER/REPO/rulesets \
  --method POST \
  --input .github/ruleset-default.json

Code ownership rules are defined in .github/CODEOWNERS.

Contributing

See CONTRIBUTING.md for development setup, branch conventions, and contribution guidelines.

Please read our Code of Conduct before participating.

License

Apache License 2.0 - see LICENSE for details.