andreibesleaga/kaiban-distributed

#### Implementation of Distributed Actor Model for AI swarms using TypeScript/Node.js, Redis/Kafka, OpenTelemetry.
One of the first open-source projects to combine Enterprise Messaging (Kafka/Redis), Distributed Actor-Model Isolation, AI Multi-Agent Orchestration, and Kanban Visualization into a single JavaScript-ecosystem runtime, for agents and humans.
> Distributed horizontally-scalable Actor-Model Multi-Agent System Runtime, using Kanban style visualization for workflows.
> Run multiple AI agents teams with independently deployed Node.js processes, real-time visibility and multi-agent orchestration via Redis/Kafka pub/sub, A2A, and MCP.
> For running the examples see EXAMPLES.md. For technical documentation, check the files in docs/.
> (For integrating KaibanJS system base, follow documents in docs/KAIBANJSINTEGRATION.md).
> Some short articles can be read on Medium and Substack.
Examples dedicated repository: kaiban-distributed-examples
Website: andreibesleaga.github.io/kaiban-distributed
While most frameworks treat agents as scripts, Kaiban Distributed treats them as Stateful Actors. By using async Node.js, each agent operates in its own space, communicating via a pluggable Messaging Abstraction Layer (Kafka or BullMQ/Redis), horizontally scaling AI workforce.
The systems allows creation of Teams of Agents for various Tasks, Systems Integrations, and Data handling/analytics, with scalable AI workflows, from local usage to customizable Enterprise Grade Systems.
Integrates with existing KaibanJS system and agents, external agentic systems, or any service that can publish via A2A / MCP / Redis / Kafka — connecting them into actor-model team flows or peer-to-peer coordination.
Any Agentic AI Patterns (Loops & Harnesses), Team Archetypes, Workflows, and Human-in-The-Loop (BudgetCheckings) interruptions could be implemented programmatically at any point of the workflow/tasks, given the Actor-Model Isolation, Distributed Node Agents, Custom Orchestrators/Coreography, Messaging Layer - in JS/TS, or via other A2A/MCP comms and agents, or by any other programs and connectors into the messaging layer (visually connected kanban boards and any other display methods connecting directly to the published workflows messages on Redis/Kafka, such as CLI I/O, Enterprise Programs, etc.
Published on npm as kaiban-distributed (canonical) and mirrored on GitHub Packages as @andreibesleaga/kaiban-distributed; the container image is ghcr.io/andreibesleaga/kaiban-distributed.
## Summary
bash
# 1. Clone and install
git clone https://github.com/andreibesleaga/kaiban-distributed
cd kaiban-distributed && npm install
# 2. Configure
cp .env.example .env
# Edit .env — add OPENROUTERAPIKEY or OPENAIAPIKEY + AGENTIDS
# 3. Start the full blog-team demo (Docker Compose, workers, gateway, orchestrator, monitor)
./scripts/blog-team.sh start
# or the Global Research Distributed Team
./scripts/global-research.sh start
# use flags for chaos (20% searcher crash rate) and number of parallel instances
# ./scripts/global-research.sh start --chaos --searchers 6
# → Script prints board URLs when the gateway is ready. Open one in a separate terminal/tab.