Skip to content
All apps

OpenClaw blueprint

About this app

Explore a personal AI agent as a blueprint: channels, gateway, memory, skills, tools, models, browser and sandbox. Six scenarios trace a message through the system.

The idea

OpenClaw is a personal AI agent that runs as a service on your own machine and is reached through chat apps. A gateway receives messages from channels such as WhatsApp, Telegram, Slack or Discord and assigns them to sessions. The agent works in a loop of reading, deciding and acting: it pulls context from a memory of Markdown files, loads skills (instructions as SKILL.md) when needed, calls tools such as a browser or shell, and queries an interchangeable language model. Risky commands run in a sandbox and need human approval.

What the app shows

The architecture as a technical drawing: channels on the left; in the middle the gateway with router, cron and heartbeat, memory, skills, agent loop and tools; on the right the model, nodes (paired devices), browser and sandbox. A numbered parts list explains each component, and a “Detail A” loupe shows the inside of the part currently at work. Six scenes send a dot through the plan, for example a question via WhatsApp, a morning briefing triggered by cron, browser research or a coding task that needs approval.

Explore it

Pick a scene and play it, or go through it with “Step”; the sequence panel describes each step in words. Notice that the housing always stays the same and only the path of the task changes. The scene “New skill from ClawHub” is especially telling: at first only the skill’s name and description sit in the system prompt, and the full text is loaded only when needed. The mouse wheel zooms, dragging pans, and clicking a component opens its loupe.

Frequently asked questions

How is a personal AI agent like OpenClaw structured?
At the centre is a gateway that runs as a service on the machine and receives messages from chat channels. A router assigns them to sessions, and the agent loop reads context from memory, decides with the help of a language model and acts through tools. Skills provide instructions for specific tasks, and a sandbox limits which commands run without asking.
What is an agent loop?
An agent loop is the working cycle of an AI agent: it reads the current context, lets a language model decide the next step and carries it out with a tool. The result flows back into the context, and the loop continues until the task is done. This lets an agent handle several steps in a row, such as reading pages, clicking and building a table.
What are skills in an AI agent?
In OpenClaw a skill is an instruction file in Markdown (SKILL.md) with a description and requirements, not built-in code. At first only the name and description appear in the system prompt. When the agent sees that a skill fits, it loads the full text, which keeps the context lean.
What does the OpenClaw blueprint app show?
It shows the OpenClaw architecture as an interactive drawing with a parts list and a detail loupe. Six scenes follow a task step by step through channels, gateway, memory, skills, tools and model. You can play, step through, zoom and click on any component.

Subject: AI literacy

More from AI literacy

A VisuApp by heyprof: interactive, free in your browser, no sign-up. What is a VisuApp? · All apps