10 replies

[Architecture Question] Agent Memory & Isolated Workspaces

I'm running into a workspace architecture issue with my multi-agent setup and would love your input on the best approach.

Current Setup
I have 4 agents: Eva (orchestrator), Makima, Hera, and Lucy. I want them to auto-save memory after each task.

My desired folder structure:
workspace/
├── agents/
│ ├── hera/memory/ ← Hera's memory
│ ├── lucy/memory/ ← Lucy's memory
│ └── makima/memory/ ← Makima's memory

The Problem
When I spawn sub-agents using

sessions_spawn()

sessions_spawn()

, OpenClaw auto-creates isolated workspaces:
-

/root/.openclaw/workspace-hera/

/root/.openclaw/workspace-hera/

/root/.openclaw/workspace-lucy/

/root/.openclaw/workspace-lucy/

The agents write their memory logs to these isolated folders, which creates:
1. Duplication: Same files in two places
2. QMD indexing issues: QMD can't index the isolated workspaces easily
3. Cleanup overhead: Need to sync then delete the isolated folders

Questions
1. Is there a way to mount

workspace/agents/hera/

workspace/agents/hera/

into the isolated session instead of creating a separate folder?
2. Or should I use

sessionTarget: "main"

sessionTarget: "main"

to avoid isolated workspaces entirely? (Security concerns?)
3. Is auto-cleanup (delete workspace-hera/ after sync) the recommended approach?

What's the best practice for agent memory persistence in OpenClaw?

Thanks!
@Krill 🦐