Skip to content

WandaSystems/WANDA

WANDA

Autonomous AI Operating Partner

A modular AI system that doesn't just assist — it operates. Built from first principles. Early development.

Status TypeScript License


What is WANDA?

WANDA is an autonomous AI operating partner — a system designed to coordinate multiple AI agents, manage persistent memory, handle voice interaction, and execute real operational workflows.

It's not a chatbot wrapper. It's infrastructure for AI that works independently.


Architecture

┌─────────────────────────────────────────────────┐
│                    WANDA Hub                     │
│         Provider Routing · Auth · State          │
├──────────┬──────────┬──────────┬────────────────┤
│  Memory  │  Voice   │  Tools   │   Channels     │
│ (mnemos) │(Vox-Voice)│  (MCP)  │ (Multi-platform)│
├──────────┴──────────┴──────────┴────────────────┤
│              Agent Workspace Layer               │
│        22 Specialized Agent Archetypes           │
├─────────────────────────────────────────────────┤
│             Hook System (ADR-0005)               │
│    Provider-agnostic middleware · 13 events       │
└─────────────────────────────────────────────────┘

Monorepo Structure

apps/
├── wanda-hub          # Central orchestration service
├── wanda-bot          # Agent runtime
└── wanda-webchat      # Web interface

packages/
├── core               # Shared types, config, utilities
├── memory             # Long-horizon memory (mnemos)
├── voice              # Speech interaction (Vox-Voice)
├── providers          # LLM routing (Gemini, Anthropic, OpenAI, Ollama)
├── channels           # Multi-platform messaging adapters
├── tools              # MCP tool execution
├── secrets            # Credential management (Key Vault)
├── workspace-mcp      # Workspace MCP server
├── sandbox            # Isolated execution environment
└── shared             # Cross-package utilities

Design Principles

Graph Memory, Not Vector Search

Precision retrieval through a bi-temporal knowledge graph. No approximate nearest neighbor guesswork.

Voice-First

WebRTC + Whisper + ElevenLabs at the OS layer. Voice is a primary channel, not an afterthought.

Full Observability

Every agent decision is traceable. The MCC dashboard provides real-time transparency.

Modular Subsystems

Each component ships independently:

  • mnemos — Memory infrastructure (beta)
  • Vox-Voice — Speech and voice stack
  • Key Vault — Secret management

Quick Start

git clone https://github.com/WandaSystems/WANDA.git
cd WANDA
pnpm install
pnpm run typecheck
pnpm run dev

Requirements: Node.js 20+, pnpm


Status

Early development. The Gen-2 routing backbone (LLM Gateway, AgentBus, OAuth) is operational. Memory, voice, and workspace layers are being integrated from their independent modules.

This project is not yet ready for production use. Contributions and feedback are welcome.


Documentation

Area Link
Project Overview docs/00_overview
Architecture docs/02_architecture
Decisions (ADRs) docs/03_decisions
Gen-2 Build Logs docs/06_gen2_build_logs

Team

Built at WandaSystems.

License

MIT — see LICENSE.

About

Workspace-aware AI orchestration system for agent workflows, voice interfaces, and operational automation.

Resources

License

Code of conduct

Contributing

Security policy

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors