agent-memory-improved

What This Skill Does

The agent-memory-improved skill provides a robust, locally hosted persistent memory subsystem for your OpenClaw agents. Unlike volatile memory that resets upon session termination, this skill utilizes a dedicated SQLite-backed service to maintain a persistent state. The core innovation of this version is the implementation of cryptographically secure Ed25519 signatures, which ensures that all read and write operations are authenticated and tamper-proof. By deriving unique keypairs from a standard BIP39 mnemonic, the agent gains a permanent identity that remains consistent across system restarts. This service operates entirely on your local machine, ensuring that no sensitive preference data or internal learning snapshots are ever transmitted to external servers.

Installation

To get started, first ensure your system has the required Python environment configured. Run the following command via your terminal or terminal interface: pip install fastapi uvicorn psutil pydantic cryptography mnemonic base58. Once dependencies are satisfied, use the ClawHub manager by executing clawhub install openclaw/skills/skills/lucaspdude/persistent-private-agent-memory. Navigate to the skill directory and execute ./scripts/setup.sh to initialize the database and identity keys. You can then trigger the background service using ./scripts/start.sh, which will expose the internal memory API to your agent agent processes.

Use Cases

This skill is designed for power users who require long-term continuity in their AI interactions. Common use cases include: 1) Maintaining a complex list of coding style preferences (e.g., specific language versions or documentation standards) that persist across months of development. 2) Tracking progress on multi-stage research projects where the agent needs to recall previous "aha!" moments or dead-ends from weeks prior. 3) Storing custom user goals, such as personal productivity targets, which the agent can reference to provide proactive suggestions. 4) Managing a repository of "knowledge gaps" that the agent tracks and attempts to fill during research tasks.

Example Prompts

1. "Check your memory and tell me what we decided regarding the coding standards for our Rust project last week."
2. "Update your memory store: mark the 'Master async patterns' goal as 60% complete and note that I prefer using Tokio over async-std."
3. "Store our current conversation history as a knowledge snapshot, focusing on the decision logic we used to arrive at this architectural solution."

Tips & Limitations

When using this skill, remember that the memory service is local-only. Back up your mnemonic phrase generated during setup, as losing it means losing access to your stored memory. For best results, trigger a 'store' operation at the end of every significant session to capture the latest context updates. The system currently favors structured JSON snapshots; while it is efficient for structured metadata, avoid storing large binary blobs directly in the memory store, as this can degrade retrieval performance over time.