#OpenClaw Multi-Agent

OpenClaw achieves multi-agent / multi-user collaboration through session isolation: one gateway can serve different people, groups, and workspaces with non-interfering agent instances.

#The Session Isolation Model

OpenClaw's session manager sorts out "who, where, and which context," enabling isolation:

Group A ──→ session A (isolated context)
Group B ──→ session B (isolated context)
Your DM ──→ main session
Colleague's DM ──→ their own session

This keeps "one OpenClaw serving many people/groups" from mixing or leaking each other's context.

#Queue: Serial Per Session, Parallel Across Sessions

The gateway's Queue subsystem serializes runs within the same session — avoiding conflicts from concurrent execution in one context — while different sessions stay independent and can progress in parallel.

Same session:  task1 → task2 → task3 (serial, avoids conflicts)
Different sessions: session A ∥ session B ∥ session C (parallel)

#Typical Multi-Agent Uses

• Personal alter-egos: a work assistant and a life assistant in different workspaces, no cross-contamination
• Team-shared gateway: one session per member, single deployment, isolated each
• Group bots: add OpenClaw to several groups, each with its own context and persona
• Dedicated workspaces: one workspace for DevOps, one for content — each with its own skills and HEARTBEAT.md

#Coordinating with Skills / Memory

Different sessions can reuse the same skills (global skills/) or have workspace-specific skills and memory:

Global ~/.openclaw/skills/   →  available to all sessions
<workspace>/skills/          →  only that workspace/agent
<workspace>/HEARTBEAT.md     →  that workspace's autonomy checklist

This shares common capabilities while customizing dedicated flows for each "alter-ego."

#Practical Tips

• Use workspaces to separate agents with different responsibilities, avoiding context interference
• For team sharing, combine allowlists + approval gates for permission isolation
• Put common capabilities in global skills/ and special flows in the workspace to reduce duplication
• Maintain a separate HEARTBEAT.md per autonomous workspace so each does its own job

#Next Steps

• Message Channels — how session isolation maps to chat platforms
• Skills — reusing global vs. workspace skills
• CLI & Automation — let multiple workspaces run autonomously