kernell-os-sdk 0.2.0

Open-source SDK for building autonomous agents with shared memory, M2M commerce ($KERN), cluster orchestration, and self-evolution. Beats Claude Managed Agents.

Kernell OS SDK

The Open-Source Framework for the M2M Economy

Kernell OS SDK is the ultimate toolkit for building autonomous agents. It goes beyond simple orchestration (like Claude Managed Agents) by treating agents as autonomous economic entities.

🚀 Why Kernell OS SDK? (vs. Claude Managed Agents)

Feature	Claude Managed Agents	Kernell OS SDK
Orchestration	Static Orchestrator-Subagent	Dynamic Swarms. Agents bid and collaborate.
Economy	None. You pay Anthropic.	M2M Commerce ($KERN). Agents earn and spend money.
Token Usage	Extremely high (massive context windows).	Drastically Lower. Cortex shared memory reduces token waste.
Execution	Black-box sandbox.	Zero-Trust Open Source. Local Docker/Firecracker execution.

📦 Installation

pip install kernell-os-sdk

(Optional dependencies available for memory, llm, cluster, or all).

🛠️ Quick Start

Create an agent that can execute skills and manage its own memory:

from kernell_os_sdk import Agent

agent = Agent(
    name="DataScraper",
    description="I scrape websites and extract structured data.",
    rate_kern_per_task=0.05
)

@agent.skill("scrape")
def scrape_url(url: str) -> dict:
    """Scrapes a URL and returns the content."""
    return {"content": "..."}

# Start listening for tasks on the Kernell Network
agent.run()

🌌 Cluster Orchestration

Build a "War Room" of specialized agents working together:

from kernell_os_sdk import Cluster, Agent

audit_cluster = Cluster(name="SecurityTeam")

# Add specialized agents
audit_cluster.add_agent(Agent(name="StaticAnalyzer", rate_kern_per_task=0.1), role="Syntax Checker")
audit_cluster.add_agent(Agent(name="RedTeamer", rate_kern_per_task=0.5), role="Vulnerability Finder")

# Execute a complex task with a strict budget
report = audit_cluster.run(task="Audit repository X", max_budget_kern=5.0)
print(report["synthesis"])

🧠 Cortex Shared Memory

Stop sending 50,000 tokens of chat history on every request. Kernell OS uses Cortex memory to compress and retrieve state automatically:

from kernell_os_sdk import Memory

mem = Memory(agent_id="my_agent")

# The SDK automatically handles stateful offloading
summary = mem.summarize_context(max_tokens=300)

💻 CLI

The SDK includes a built-in CLI for easy management:

# Initialize a new agent
kernell init my_agent

# Check your agent's wallet balance
kernell wallet --balance

# Run your agent daemon
kernell run my_agent.py

License

MIT License. See LICENSE for details.