Quick Start
Run your first multi-agent swarm session after installing clnode.
1. Start the Daemon
npx clnode start
The daemon runs on http://localhost:3100 by default.
2. Open Claude Code
Start a new Claude Code session in your project directory. The hooks are already installed from clnode init.
3. Give a Multi-Agent Task
Ask Claude Code something like:
Create a REST API endpoint for user profiles.
Use a backend agent for implementation and a reviewer agent to check the code.
Claude Code will:
- Spawn a backend agent → clnode records it via
SubagentStarthook - Agent completes work → clnode saves summary via
SubagentStophook - Spawn a reviewer agent → clnode injects backend agent’s summary automatically
- Reviewer sees what was built without the Leader relaying it
4. Monitor in Web UI
npx clnode ui
See active agents, context entries, and task status in real-time.
5. Create Custom Agents
Use the /clnode-agents skill in Claude Code to:
- Discover installed agents, skills, and rules
- Create new agents with proper frontmatter
Key Concept
Agents communicate through time, not through the Leader:
Agent A finishes → summary saved to DuckDB
Agent B starts → receives A's summary via additionalContext
Leader → only makes high-level decisions
This keeps the Leader’s context minimal while agents share results automatically.