Per-agent memory isolation in OpenClaw
✅Solved
Hey folks — I’m trying to set up OpenClaw with two agents:
• main agent: my “everything” assistant (Telegram now)
• Slack agent: a separate personality for Slack where I treat channels as public by default
Routing is solved (bindings can route all Slack → agentId=slack). The problem is memory isolation.
What I want:
I want the Slack agent to have its own long-term memory (or at least not be able to recall private memories), while the main agent keeps full memory — ideally in a single gateway.
What I found so far:
• memory-lancedb plugin config looks global (single dbPath, autoCapture, autoRecall under plugins.entries.memory-lancedb.config)
• I couldn’t find a supported way to set autoCapture/autoRecall per agent
• I also didn’t find a clean ClawHub/skill solution that enables “separate LanceDB DB per agent” in one gateway
I could run two gateways, but I’m trying hard to avoid that operational complexity.
Questions:
1. Has anyone found a way to do per-agent memory isolation (especially memory-lancedb) inside one OpenClaw gateway?
2. Any recommended workaround patterns?
• custom/fork memory plugin that tags memories with agentId and filters recall?
• alternative memory backend that’s workspace-scoped per agent?
3. If you’ve done this successfully, what did your config/plugin approach look like?
Happy to share config snippets / more details if useful. Thanks!
• main agent: my “everything” assistant (Telegram now)
• Slack agent: a separate personality for Slack where I treat channels as public by default
Routing is solved (bindings can route all Slack → agentId=slack). The problem is memory isolation.
What I want:
I want the Slack agent to have its own long-term memory (or at least not be able to recall private memories), while the main agent keeps full memory — ideally in a single gateway.
What I found so far:
• memory-lancedb plugin config looks global (single dbPath, autoCapture, autoRecall under plugins.entries.memory-lancedb.config)
• I couldn’t find a supported way to set autoCapture/autoRecall per agent
• I also didn’t find a clean ClawHub/skill solution that enables “separate LanceDB DB per agent” in one gateway
I could run two gateways, but I’m trying hard to avoid that operational complexity.
Questions:
1. Has anyone found a way to do per-agent memory isolation (especially memory-lancedb) inside one OpenClaw gateway?
2. Any recommended workaround patterns?
• custom/fork memory plugin that tags memories with agentId and filters recall?
• alternative memory backend that’s workspace-scoped per agent?
3. If you’ve done this successfully, what did your config/plugin approach look like?
Happy to share config snippets / more details if useful. Thanks!
Solution
It appears that this is an open feature request in GitHub so I will mark this as resolved.
https://github.com/openclaw/openclaw/issues/2141
https://github.com/openclaw/openclaw/issues/2141
The AI that actually does things. Emails, calendar, home automation - all from your favorite chat app. New shell, ready to help. The lobster way.
85,605Members
Resources
Was this page helpful?
