LaunchKit · 2026
Back to Skills

neverforget

Automates Sovereign Local Vector Memory and Gemma-300M Embeddings.

2
510 downloads
by @greatape42069

Setup & Installation

openclaw skills install @greatape42069/neverforget

Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:

npx clawhub install neverforget

Version History

v1.0.4Feb 18, 2026 - Version note by skill developer:

- Hardened exclusion rules for memory indexing by skipping secret directories such as .ssh, .aws, and .env - Improved compliance with ClawHub security audit requirements - Documentation and install script updated to reflect new exclusion logic - No changes to functional code; update focused on enhanced security configuration

What This Skill Does

Manages local vector embeddings for OpenClaw agents using the Gemma-300M model via node-llama-cpp. All embeddings and searches run on-device with no external API calls. Includes recursive loop protection to prevent the agent from indexing its own vector database.

Removes dependency on external embedding APIs, keeping all data local and eliminating per-request costs or privacy exposure.

When to use it

  • Searching past agent sessions without sending data to external servers
  • Indexing a local codebase for semantic retrieval across agent runs
  • Maintaining project context in offline or air-gapped environments
  • Preventing memory bloat by excluding credential files and node_modules from indexing
  • Monitoring vector memory health with a periodic heartbeat check

Example Workflow

Here's how your AI assistant might use this skill in practice.

INPUT

User asks: Set up local memory indexing for my OpenClaw environment

AGENT
  1. 1Checks for existing node-llama-cpp install and adds it if missing
  2. 2Configures the local memory provider with the Gemma-300M model path from Hugging Face
  3. 3Sets exclusion rules to skip the vector DB directory, node_modules, and credential files
  4. 4Restarts the OpenClaw gateway to apply the new configuration
  5. 5Runs initial memory index if chunk count is zero, otherwise reports existing chunk count
OUTPUT

Memory is active with the indexed chunk count confirmed via `openclaw memory status`