About this skill

The Problem

Long-term memory for an AI Agent can become an engineering problem when it relies only on vector search or cloud session storage: entries keep growing, low-value context is hard to prune, and private data may have to leave the local machine. ES-Memory targets the personal-agent case by putting memory compression and local lifecycle management into a single CLI engine, writing data only to the supplied data_dir without network dependence.

How It Works and Where It Fits

The core path is built around a local plaintext JSON file, nodes.json:
- Ebbinghaus decay: low-weight memories are cleaned up automatically to free local space;
- Pin protection: important memories can be locked with pin() and are not deleted;
- Chinese segmentation: it prefers jieba exact mode and falls back when the dependency is missing;
- Local-only operation: data stays in the directory you provide and does not call external services.

Backup and migration are simple: copy the entire data_dir, then restart on another machine to reload automatically. The practical boundary is that it fits better as a local personal memory layer, prompt-context compression component, or agent state cache than as a multi-tenant database or high-availability backend. Because storage is plaintext JSON, sensitive entries should be protected by directory permissions and access controls.

Screenshots

Use Cases

Prune low-value context in a personal agent by letting decay remove expired memories while keeping high-weight entries.
Move local agent state by copying the whole data_dir to a new machine and restarting to reload nodes.json.
Extract Chinese memory keywords with jieba exact-mode segmentation, with automatic fallback when the dependency is missing.
Preserve critical user preferences by calling pin() so the selected memory is exempt from automatic decay cleanup.

Best For

Engineers maintaining local personal-agent memory who need offline context lifecycle control.
Chinese prompt-engineering researchers who need jieba-based keyword extraction and fallback testing.
Developers building offline agents who want to pin user preferences and manage local backup and migration.
Engineers managing AI agent state caches who need pin() protection and directory-level migration.