Loading...
Loading...
OpenClaw is an open-source agent orchestrator that gives Claude Code a persistent brain — memory, agents, workflows, and skills — all in plain text files.
OpenClaw sits on top of Claude Code and adds structure. Without it, each Claude session starts fresh with no memory of what happened before. With it, your agent knows its role, remembers past decisions, and follows consistent workflows.
Everything lives in plain markdown files checked into your repository. No database, no proprietary format, no lock-in. If you stop using OpenClaw, the files stay and you can read every decision your agent ever made.
Specialist AI personalities with defined roles, memory, and tools. Each agent handles a specific type of work.
Ordered sequences of agent tasks. A workflow might run planner → coder → reviewer in sequence.
Persistent context that survives across sessions. Stored in markdown files, read at session start.
Reusable prompt libraries loaded on demand. Skills give agents specialized domain knowledge without bloating the base prompt.
npm install -g openclawopenclaw init my-workspaceopenclaw agent create --name planneropenclaw run planning --task "Build a login page"openclaw connect --runtime claude-codeAn agent is defined by a markdown file in openclaw/workspace/agents/. Here's a minimal example:
# Planner Agent ## Role You are a senior software architect. When given a feature request, you break it down into a concrete implementation plan with ordered steps. ## Memory Read MEMORY.md before every response. Write key decisions to MEMORY.md after completing a plan. ## Output Format Always return a numbered task list with: - Task description - Estimated complexity (S/M/L) - Dependencies on other tasks - Files likely to be modified
OpenClaw uses Claude Code as its execution runtime. Point OpenClaw at your Claude Code installation and it will spawn sessions automatically:
# openclaw/config.json
{
"runtime": "claude-code",
"model": "claude-sonnet-4-5",
"workspace": "./openclaw/workspace",
"memory": "./openclaw/workspace/MEMORY.md",
"agents_dir": "./openclaw/workspace/agents"
}Once connected, running openclaw run <workflow> launches a Claude Code session with the right agent context already loaded.
The full OpenClaw setup guide covers the 7-layer brain architecture, 12-agent army, 5-tier memory, and automated workflows — exactly how we configure it for customers.