Orloj: Declarative Orchestration Runtime for Multi-Agent AI Systems

What Orloj Does

Orloj is an Apache 2.0 licensed orchestration runtime for multi-agent AI systems. You declare agents, tools, policies, and workflows in declarative YAML manifests, and Orloj handles scheduling, execution, governance, and reliability. The creators built it because running AI agents in production today resembles running containers before Kubernetes: ad-hoc scripts, no governance, no observability, and no standard way to manage agent fleets.

Key Features

• Agents-as-Code: Define agents in YAML with their model, tools, permissions, and execution limits.
• Orchestration Topologies: Compose agents into directed graphs—pipelines, hierarchies, or swarm loops.
• Governance Enforcement: AgentPolicy, AgentRole, and ToolPermission are evaluated inline during execution before every agent turn and tool call. Unauthorized actions fail closed with structured errors and audit trails.
• Policy Controls: Set token budgets per run, whitelist models, block specific tools, and scope policies to individual agent systems.
• Production Reliability: Lease-based task ownership, capped exponential retry with jitter, idempotent replay, and dead-letter handling.
• Scheduling: Supports cron triggers and webhook-driven task creation.
• Tool Isolation: Configure per tool based on risk level—direct, sandboxed, container, or WASM execution.
• Native MCP Support: Register MCP servers (stdio or HTTP), auto-discover tools, and apply governance policies.

Architecture

The system uses a server/worker split:

• orlojd server: Hosts the API, resource store (in-memory for dev, Postgres for production), and task scheduler.
• orlojworker instances: Claim and execute tasks, route model requests through a gateway (OpenAI, Anthropic, Ollama, etc.), and run tools in configurable isolation.

For local development, run everything in a single process with: orlojd --embedded-worker --storage-backend=memory.

Getting Started

Download orlojd (server) and orlojctl (CLI) from GitHub Releases, extract them, and run:

# Start the server with an embedded worker
./orlojd --storage-backend=memory --task-execution-mode=sequential --embedded-worker

Open http://127.0.0.1:8080/ for the web console, then apply a starter blueprint:

# Apply a starter blueprint (pipeline: planner -> research -> writer)
./orlojctl apply -f examples/blueprints/pipeline/
# Check the result
./orlojctl get task bp-pipeline-task

Three starter blueprints are included: pipeline, hierarchical, and swarm-loop. The team is building templates for operational workflows like incident response triage, compliance evidence collection, CVE investigation pipelines, and secret rotation auditing.

Current Status

This is v0.1.0 with active development. The roadmap includes hosted cloud, compliance packaging, and more. The full runtime is open source today.