Laddr
A Python framework for multi-agent systems where agents communicate, delegate tasks and execute work in parallel, styled as a microservices architecture for AI agents with built-in message queues and observability.
Laddr supports two operating modes: a coordinator-orchestrator mode where a coordinator agent analyzes tasks and makes real-time routing decisions to specialist agents, and a sequential deterministic workflow mode where tasks flow through a fixed pipeline with explicit inputs, outputs and dependencies for each step.
It scales agents independently, running multiple workers per agent with automatic load balancing over Redis Streams, and tolerates individual worker failures without stopping the system. Every agent action is traced automatically to SQLite or PostgreSQL, with optional Langfuse integration, real-time metrics and an interactive dashboard for monitoring agents, traces and batches, plus a playground for testing agents interactively. A laddr init command scaffolds a complete project, and agents can delegate tasks to each other and execute independent batches in parallel with full tracking.