Qualixar OS

The Universal OS for AI Agents

One platform. Every model. Every framework. Every transport.

Qualixar OS is the operating system that runs AI agents — yours and everyone else's. It doesn't replace your agent framework. It powers it. Import agents from OpenClaw, NemoClaw, DeerFlow, CrewAI, LangGraph, or build native. Run them all through one dashboard, one config, one runtime.

Install in 30 Seconds

npx qualixar-os

That's it. Dashboard opens. Start chatting with AI, or design a multi-agent team.

# Or install globally
npm install -g qualixar-os

# Start the dashboard
qos serve --dashboard --port 3001

# Open your browser
open http://localhost:3001/dashboard/

Why an OS?

Agent frameworks give you building blocks. Qualixar OS gives you the complete runtime:

| What You Need | Framework Approach | Qualixar OS | |---|---|---| | Route to the right model | Write custom logic | Built-in (9+ models, cost/quality/latency routing) | | Run multi-agent teams | Build from scratch | 12 topologies, auto-designed by Forge AI | | Quality assurance | Hope for the best | Judge pipeline with consensus protocols | | Persistent memory | Add a vector DB | SLM-Lite cognitive memory (4 layers, local) | | Monitor costs | Check your bill later | Real-time budget tracking per agent | | Manage tools | Hardcode per agent | Marketplace with categories, one-click install | | Dashboard | Build your own | 24 tabs, production-ready | | Security | DIY | 4-layer sandbox, RBAC, credential vault |

Universal Compatibility — The Claw Bridge

Qualixar OS doesn't compete with agent frameworks. It runs them.

# Import agents from any framework
qos claw import ./my-openclaw-agent.yaml       # OpenClaw
qos claw import ./nemoclaw-config.json          # NemoClaw (NVIDIA)
qos claw import ./deerflow-workflow.json         # DeerFlow (ByteDance)
qos claw import ./gitagent-spec.yaml             # GitAgent (Microsoft)

# Or use native Qualixar agents — designed by Forge AI
qos task "Build me a code review pipeline"
# → Forge auto-designs a team of 4 agents with the right tools

| Framework | Import | Run | Manage from Dashboard | |-----------|--------|-----|----------------------| | OpenClaw | Yes | Yes | Yes | | NemoClaw (NVIDIA) | Yes | Yes | Yes | | DeerFlow (ByteDance) | Yes | Yes | Yes | | GitAgent (Microsoft) | Yes | Yes | Yes | | LangChain / LangGraph | Via adapter | Yes | Yes | | CrewAI | Via adapter | Yes | Yes | | AutoGen | Via adapter | Yes | Yes | | MCP Servers | Native | Yes | Yes | | A2A Protocol | Native | Yes | Yes | | Custom Agents | skill.json manifest | Yes | Yes |

The Dashboard — Everything in Your Browser

24 interactive tabs. No CLI required for daily use.

| Tab | What It Does | |-----|-------------| | Chat | Talk to AI with streaming, file upload, model selection | | Forge | AI auto-designs specialized agent teams from your prompt | | Builder | Visual workflow editor — drag, connect, run | | Marketplace | Browse, search, install skills and tools with one click | | Agents | Monitor running agents, view stats, kill/restart | | Swarms | Watch multi-agent execution in real-time | | Judges | Quality verdicts, consensus scores, approval/rejection | | Cost | Per-model, per-agent, per-task cost breakdown | | Memory | Persistent agent memory — search, inspect, manage | | Pipelines | Multi-step task orchestration with checkpoints | | Tools | Tool palette with 6 categories, drag onto agents | | Traces | OpenTelemetry spans for every agent action | | Settings | Providers, models, budget, security, connectors | | ...and 11 more | Flows, Connectors, Logs, Gate, Datasets, Vectors, Blueprints, Brain, Audit, Lab, Marketplace |

Forge — AI Designs Your Agent Teams

Tell Forge what you need. It designs the team.

qos task "Review my GitHub PRs for security issues"

Forge responds:

• Creates 3 agents: Security Analyst, Code Reviewer, Report Writer
• Assigns tools: github_pr_read, code_search, file_write
• Picks topology: Sequential (analyze → review → report)
• Sets budget: $0.05 estimated cost
• Runs the team. Returns results.

No manual configuration. No YAML files. One sentence, full team.

Marketplace — One-Click Tool Installation

Browse the global skill registry from your dashboard. Install with one click. Forge auto-discovers new tools.

Dashboard → Marketplace → Search "GitHub" → Install → Done
Your agents now have GitHub tools. Forge uses them automatically.

6 Tool Categories:

| Category | Tools | |----------|-------| | Web & Data | Web search, crawl, scrape, RSS, API connectors | | Code & Dev | GitHub, file I/O, shell, linter, test runner | | Communication | Slack, email, Discord, webhook | | Knowledge | Vector search, document reader, database, RAG | | Creative | Image gen, video gen, TTS, diagrams | | Enterprise | CRM, project management, analytics, cloud |

18 built-in skills ship with the product. Community skills available via the global registry.

12 Execution Topologies

| Topology | Pattern | Use Case | |----------|---------|----------| | Sequential | A → B → C | Pipelines, step-by-step tasks | | Parallel | A ‖ B ‖ C → merge | Independent analysis, speed | | Hierarchical | Boss → Workers | Task decomposition | | DAG | Directed graph | Complex dependencies | | Debate | Pro vs Con → Judge | Quality decisions | | Mesh | All-to-all | Collaboration | | Star | Hub ↔ Spokes | Central coordinator | | Grid | Matrix communication | Structured teams | | Forest | Multiple trees | Parallel hierarchies | | Circular | Round-robin | Iterative refinement | | Mixture-of-Agents | Ensemble → Synthesize | Best-of-N outputs | | Maker | Build → Test → Ship | Engineering workflows |

Configuration

# ~/.qualixar-os/config.yaml
mode: companion  # or 'power'

providers:
  azure:
    type: azure-openai
    endpoint: https://your-resource.cognitiveservices.azure.com/
    api_key_env: AZURE_AI_API_KEY

models:
  primary: claude-sonnet-4-6
  fallback: gpt-4.1-mini

# ~/.qualixar-os/.env
AZURE_AI_API_KEY=your-api-key-here

CLI

qos serve [--port 3001] [--dashboard]   # Start server + dashboard
qos task <prompt>                        # Run a task (Forge designs team)
qos chat                                 # Interactive companion mode
qos forge <prompt>                       # Direct Forge access
qos marketplace search <query>           # Search skills
qos marketplace install <skill>          # Install a skill
qos claw import <path>                   # Import from any framework
qos config show                          # Show configuration

API

27+ REST endpoints. Full OpenAPI documentation at /api/docs.

| Method | Path | Description | |--------|------|-------------| | POST | /api/tasks | Submit a task (Forge auto-designs team) | | GET | /api/tasks | List tasks with status | | GET | /api/agents | Active agents | | GET | /api/cost | Cost breakdown | | GET | /api/tools | Tool catalog with categories | | GET | /api/skill-store/browse | Browse marketplace | | POST | /api/chat/conversations/:id/messages | Chat (streams via WebSocket) | | GET | /.well-known/agent-card | A2A discovery |

Tech Stack

| Layer | Technology | |-------|-----------| | Runtime | Node.js 22+ (ESM) | | Language | TypeScript 5.7 | | HTTP | Hono | | Database | better-sqlite3 (39 tables) | | Dashboard | React 19 + Vite | | AI SDKs | Anthropic, OpenAI, Azure, Ollama | | Testing | Vitest (2,472 tests, 169 files) | | Protocols | MCP, A2A, HTTP, WebSocket, CLI |

Development

git clone https://github.com/qualixar/qualixar-os.git
cd qualixar-os
npm install
npm run build
npm test              # 2,472 tests
npm run typecheck     # 0 errors

License

MIT

Built By

Qualixar — AI Agent Reliability Engineering.

5 published papers. 3 live tools. One mission: make AI agents trustworthy.