Claw Memory

What it does

Claw Memory is a service that allows different AI agent instances (like OpenClaw, KimiClaw, or NanoClaw) to share memory securely across those instances. It utilizes a Cloudflare Worker and TiDB Cloud backend, providing full data isolation by provisioning each token with its own dedicated TiDB Cloud Zero instance. Connection strings are encrypted at rest for enhanced security, and optional client-side encryption is supported for double protection.

When to use it

• Persistent State Across Agent Restarts: Share complex agent state (e.g., a long conversation history or ongoing task data) even if an agent instance crashes or restarts.
• Multi-Agent Collaboration: Enable multiple agents working on the same project to access and update shared information without needing to constantly retransmit data.
• Distributed Processing: Distribute parts of a larger task across different agent instances, allowing them to share intermediate results in a secure manner.
• Maintaining Context Across Scaling Events: Ensure that an agent’s context is preserved even when scaling up or down the number of agent instances.

Key capabilities

• Cross-instance memory sharing for Claw agents
• Full data isolation per token with dedicated TiDB Cloud Zero instances
• AES-256-GCM encrypted registry connection strings
• Optional client-side encryption using X-Encryption-Key header
• Zero-friction provisioning of TiDB Cloud Zero instances (~2 seconds)
• 30-day lifecycle for provisioned Zero instances

Example prompts

These are not prompts for the skill, but examples of what an agent might do with it:

• "Agent A, store this complex data structure in shared memory so Agent B can access it."
• "Agent C, retrieve the latest status update from shared memory and report it to me."
• "When I restart, load my previous session state from shared memory."

Tips & gotchas

Each token automatically provisions a TiDB Cloud Zero instance. Be mindful of the 30-day lifecycle; data will be lost after this period unless proactively migrated.