Two complementary open-source platforms — runtime actor-model orchestration and visual workflow design — for engineers shipping production AI systems.
What we build
Wactorz
Runtime actor-model orchestration
Describe agents in natural language — they spawn at runtime over MQTT, persist across restarts, and scale from laptop to edge. No code changes. No restarts.
Waldiez
Visual multi-agent workflow builder
Design, orchestrate, and execute multi-agent workflows through an intuitive drag-and-drop interface. Export to Python or Jupyter notebooks with one click.
Wactorz
Actor-model concurrency · MQTT · Runtime spawning · Live dashboard
Each agent runs in its own async loop with an isolated mailbox. No shared state, no race conditions — true actor isolation at every layer of the stack. Crash one, the rest keep running.
One LLM call classifies intent. The right agent spawns on demand — no predefined types, no restarts, no downtime.
All inter-agent messaging flows through MQTT pub/sub — loose coupling, real-time telemetry, and edge-native from day one.
MainActor, MonitorAgent, DynamicAgent, PlannerAgent, HomeAssistant agents, InstallerAgent, CatalogAgent — plus any agent you describe and spawn yourself.
Agent health, message flows, and cost meters updated in real time. Spawn, stop, and inspect any agent from the browser. Your entire agent system at a glance.
Agent state writes to disk on every tick. Agents survive crashes and restore full context on restart — zero data loss.
Docker, systemd, or native Python runner. InstallerAgent deploys remote nodes via SSH, bridging them into the central MQTT graph.
REST, WebSocket, Discord, WhatsApp, Telegram, or CLI. Streaming responses with rolling history summarization.
Per-agent and aggregate cost tracking across all providers. See spend in real time before it becomes a surprise.
Waldiez
Drag-and-drop multi-agent workflow builder — powered by AG2
Compose multi-agent flows with drag-and-drop. Connect agents, tools, and data sources without writing orchestration code.
Real-time collaboration between autonomous agents using the AG2 framework — battle-tested in production.
Export to Python scripts or Jupyter notebooks with one click. Import, fork, and iterate at speed.
JupyterLab Extension, VS Code Extension, and Waldiez Studio — meet your workflow wherever you already work.
🌐 Waldiez Hub
Upload, discover, and fork multi-agent workflows. Build a public portfolio. Let your flows speak for themselves.
Real-world use cases
Control Home Assistant entities through natural language. Agents map your home, issue commands, and react to sensor events in real time via MQTT.
Build reactive "if X then Y" pipelines over MQTT. Trigger agents when sensors fire, thresholds breach, or schedules hit — no polling.
DynamicAgent writes and runs Python at runtime from LLM output — sandboxed, introspectable, with live result streaming back to chat.
PlannerAgent decomposes complex goals into subtask pipelines, spawning specialist agents for each step automatically.
InstallerAgent deploys agents on remote nodes via SSH. A self-contained edge runner bridges them into the central MQTT graph — no wactorz package required on the edge.
MonitorAgent watches all actors for crashes and resource spikes — automatic crash detection and recovery built in.
Interact via Discord, WhatsApp, or Telegram. Agents respond in-thread with streaming output and spawn sub-agents on demand.
Design multi-agent flows in Waldiez's visual builder, share on the Hub, and execute or export to Python/Jupyter.
Integrations
Switch providers per-agent or per-workflow. No vendor lock-in, ever.
