claw-memory_skill

Overview

This skill provides a shared memory service for AI agents (OpenClaw, KimiClaw, NanoClaw, etc.) so agents can store, retrieve, search, and share memories across instances. It exposes a simple HTTP API to create isolated memory spaces, store individual memories or bulk-import files, and search or manage stored entries. Optional client-side encryption is supported for sensitive data.

How this skill works

Each memory space is created via a token; every token provisions an isolated TiDB Cloud Zero instance for that space. Use the token as a Bearer Authorization header on all memory API calls. You can store single memories, bulk-import many entries (up to 200 per request), search by query/tags/source/date, update or delete items, and claim a token to make the underlying database permanent. An optional X-Encryption-Key header enables client-side encryption so the server cannot decrypt without the key.

When to use it

• When multiple agent instances need a shared, persistent memory store.
• To import existing memory files or daily notes into a searchable service.
• When you need per-agent or per-project isolated databases.
• To back up or archive agent knowledge and activity logs.
• When you require optional client-side encryption for stored data.

Best practices

• Create a dedicated token per agent or project to keep data isolated and manageable.
• Claim the provided claim_url to convert the temporary instance into a permanent TiDB Starter instance before the 30-day expiry.
• Include X-Encryption-Key on all calls if you generate the token with a client-side key — otherwise the server cannot decrypt your data.
• Use bulk import for large files (MEMORY.md, daily notes) to preserve content and tags in a single operation.
• Index important fields with consistent tags and source values to simplify searches and filtering.

Example use cases

• Store conversational context and facts for an agent so other agents can retrieve shared state.
• Bulk-import an existing MEMORY.md file or daily notes to make them searchable across agents.
• Create per-user memory spaces for personalized assistants without cross-contamination.
• Archive agent outputs and reasoning traces for auditing or model improvement.
• Spin up temporary experiment spaces and claim the instance if the experiment becomes production.