Kafka for Agents

Event broker purpose-built for autonomous AI agents

View on GitHub
Part ofAgentic Wands

Built for Agent Autonomy

Loosely Coupled

Agents don't need to know about each other

Replayable

All state changes are recorded as events, enabling replay

Safe Retries

Idempotency prevents duplicate execution

File-System Based

JSONL storage, debuggable with cat/grep/tail

CLI-First

Everything available via agentika CLI

Deterministic Triggers

OS-level signals only, no LLM in the broker

How It Works

Agentika uses a decoupled, event-driven architecture. Agents communicate asynchronously through a central topic log, enabling scalable and resilient multi-agent systems.

Agent A

Autonomous Node

publish

Topic / Event Log

Central immutable ledger for all agent communications

consume

Agent B

Autonomous Node

stream (SSE)

Dashboard

Real-time Monitoring

Agentika Architecture Illustration

CLI-First Design

Everything you need to manage your agent infrastructure, right from your terminal.

terminal
# Start the broker
agentika serve

# Create a topic
agentika topic create agent-events

# Publish an event
agentika publish agent-events --data '{"action":"process"}'

# Consume events
agentika consume agent-events --group processors

# Tail with filters
agentika tail agent-events --contains "error"
event.json
{
  "id": "evt_01HQZ...",
  "topic": "agent-events",
  "timestamp": "2026-03-23T10:30:00Z",
  "data": {
    "agent": "processor-1",
    "action": "task_complete",
    "result": "success"
  },
  "meta": {
    "correlation_id": "corr_123",
    "reply_to": "responses"
  }
}
Built with Rust
0
tests (passing)
~0K
LOC Rust
0
source files
MIT
License