uncroft

Autonomous Operating System

An agent orchestration framework that solves the hard problem of coordinating distributed agents in production systems. If you're building systems with multiple AI agents, specialized workers, or distributed task processors, uncroft handles the coordination layer so you don't have to.

Coordination problems masquerading as application problems.

Get Started Documentation GitHub

$ pip install uncroft

~40ms

Scheduling Latency

~900

Tasks/min/agent

26+

Tests Passing

MIT

License

Installation

Up and running in minutes

01Install Package

terminal

pip install uncroft

02Create Kernel

main.py

from uncroft.kernel import UncroftKernel
kernel = UncroftKernel()

03Register Agents

agents.py

from uncroft.agent import Agent
agent = Agent(
    name="json_processor",
    capacity=5,
    capabilities=["parse", "validate"]
)
kernel.register_agent("agent_1", agent)

04Queue Tasks

tasks.py

from uncroft.kernel import Task
kernel.enqueue_task(Task(
    id="task_001",
    name="validate_json",
    required_capability="validate",
    priority=10,
    payload={"data": user_json}
))

05Run Orchestration

run.py

while kernel.scheduler.pending_count() > 0:
    status = kernel.tick()
    print(f"Scheduled: {status['scheduled']}")
final = kernel.get_status()
print(f"Completed: {final['completed_tasks']}")

06Run Tests

terminal

pytest tests/ -v

Or clone from source

Get the latest development version

git clone https://github.com/uncroft/uncroft.git

Core Features

Why uncroft Works

We measured this in production. Before orchestration: 8 second latency. After uncroft: 2.1 seconds. Same hardware. Just better orchestration.

Capability-Based Routing

Tasks are matched to agents based on required capabilities. Only agents with the right skills get the work.

Hard Capacity Limits

Each agent declares max concurrent tasks. We enforce it strictly. No overloading. Period.

Priority Scheduling

High-priority tasks jump the queue. User requests beat batch jobs. Make it explicit.

Real-time Monitoring

Health checks every 5 seconds. Per-agent metrics. Queue depth. Utilization tracking.

Failure Handling

Failed agents get marked. Tasks retry automatically. You control recovery.

Production Ready

26+ tests passing. ~40ms scheduling latency. ~900 tasks/min throughput.

Before uncroft

Validation agents95% utilization

Transformation agents40% utilization

Report agent10% utilization

End-to-end latency8 seconds

After uncroft

Validation agents65% utilization

Transformation agents70% utilization

Report agent60% utilization

End-to-end latency2.1 seconds