What Are Agentic Tools?

Agentic tools are autonomous systems that run independently, make decisions, and take actions without requiring human approval for each step. They are designed to be ambient—always running, periodically acting, integrated into your workflows.

This differs from coding tools where you drive and the AI assists.

Canonical tool docs live in /tools/. This hub organizes agentic platforms by behavior and risk profile.

Available Platforms

OpenClaw

Self-hosted, local-first autonomous agent platform. Runs on your hardware with extensible skills system.

Moltbook

Agent social network where bots post, vote, and comment. Reddit-like platform for autonomous agents.

Architecture Patterns

PatternDescriptionRisk Level
Local-firstRuns on your hardware, you control dataHigh (your responsibility)
Cloud-nativeVendor-hosted, managed infrastructureMedium (shared responsibility)
Fetch-and-followPeriodically fetches remote instructionsHigh (platform compromise = agent compromise)
Social agentsMulti-agent coordination, reputation systemsHigh (expanded attack surface)

Agents.md

The AGENTS.md file is a standardized way to document AI agent behavior, capabilities, and constraints for your projects. It helps both humans and AI agents understand the context and requirements.

Skills

Skills are modular capabilities that extend agent functionality. OpenClaw uses a skills system for adding new capabilities:

MCP Servers

Model Context Protocol (MCP) servers extend agent capabilities by providing standardized access to external tools and data sources.

Quick Safety Checklist

Before running any agentic tool: