How It Works
Powered by SOUL.md
OpenClaw is an open-source AI agent framework that gives your agent identity, memory, and skills. This page explains how OpenClaw works in practice, why it feels different from a chatbot, and how builders in Sofia use it to turn intent into working systems.
SOUL.md — Who Your Agent Is
The first file your agent reads. Not a system prompt — a soul. You write who you want your agent to be. Direct or gentle. Opinionated or neutral. OpenClaw's setup wizard walks you through your first one. The file evolves. As you and your agent figure each other out, the soul changes with you. No two agents are alike.
# My Agent You are thoughtful and direct. You remember what matters to me. You challenge my assumptions when I need it. Personality: warm but honest Style: concise, no fluff
Memory — How It Remembers
Your agent maintains its own memory: daily notes (raw logs of what happened), long-term memory (curated essence, distilled over time), and active context (what you're working on right now). Each session, the agent reads its memory and picks up where you left off. Like a human reviewing their journal.
## Daily Notes 09:14 User started new project 09:22 Scaffolded React app 11:45 Deployed to staging ## Insights User prefers Tailwind over CSS modules. Likes to deploy early and iterate.
Skills — What It Can Do
Skills are plug-and-play capabilities: email, calendar, smart home, code, search, anything with an API. Install what you need. Your agent reads the skill file and knows how to use it. Your files are yours. Your soul reflects your values. Your memory reflects your life. Over weeks, the agent becomes yours.
$ openclaw skills list ● email — Read, send, organize ● calendar — Events, scheduling ● smart-home — Lights, climate, locks ○ code — Build, test, deploy ○ search — Web, docs, knowledge
What makes an AI agent different?
A chatbot answers questions. An AI agent can keep context, use tools, read documents, browse the web, operate software, and stay grounded in your world. OpenClaw adds that missing layer between language and action.
The key difference is not just model quality. It is architecture. When identity, memory, tools, and safety rules live outside the prompt, the system becomes far more stable, reusable, and personal.
That is why OpenClaw Bulgaria focuses on real usage instead of abstract AI talk. We care about inboxes, research pipelines, coding flows, home systems, and business operations that actually work after the demo ends.
OpenClaw in one loop
- 1. You describe intent in plain language.
- 2. The agent reads its identity and recent memory.
- 3. It selects the right skills and tools.
- 4. It acts, checks results, and adapts.
- 5. Important outcomes get written back into memory.
Questions people ask before they get started
Can I run OpenClaw on my own machine?
Yes. That is one of the main points. OpenClaw is built for people who want an AI agent framework they can run, shape, and connect to their own tools and files.
Is OpenClaw only for developers?
No, but builders benefit fastest. Developers, founders, operators, researchers, and curious technical teams usually get value first because they already have workflows worth automating.
Why open source instead of a closed assistant app?
Because control matters. Open source AI agents are easier to inspect, extend, host, and adapt to real life. You are not stuck waiting for one company to decide what your assistant can do.
Where can I see this working in real life?
At OpenClaw Bulgaria meetups in Sofia, on the Events page, and inside the Academy where the ideas are turned into practical workflows and demos.
Keep going
If you want the practical side, start with installation. If you want the local human side, explore the Sofia community and upcoming meetups.