Quick Start

After installation, the plugin registers as OpenClaw tools automatically. You do not need additional setup.

Write a memory

memory_add("Project uses PostgreSQL 16 on AWS RDS, connection pooling via PgBouncer.")

On write, the plugin:

1. Checks for exact duplicates (hash dedup)
2. Runs LLM note construction (keywords, tags, context, category)
3. Retrieves candidate notes and generates bidirectional links
4. Evaluates memory evolution on linked notes

Search memories

memory_search("database configuration", limit=5)

Returns up to limit memories ranked by hybrid RRF score (BM25 + vector cosine) with heat boost and 2-hop BFS graph expansion.

List memory count

memory_list()

Returns the total active note count for the current agent namespace.

Manual consolidation

memory_consolidate()

Merges semantic duplicates (cosine ≥ 0.75) across the whole store, grouped by category, with link cascading. It skips knowledge notes.

---

Automatic consolidation

The plugin schedules daily consolidation automatically at 02:30 AM (in-process setTimeout). It:

• Groups notes by category
• Merges semantic duplicates (cosine ≥ 0.75) into one note, keeping the episodic type
• Cascades all link references to preserve graph topology

The plugin requires no configuration.

---

LLM CRUD gate

At agent conversation end (agent_end hook), the plugin:

1. Reads the last user message and the last assistant message, each truncated to 500 characters — not the whole conversation
2. Decides NEW / UPDATE / DELETE / NONE per note
3. Executes changes automatically

This keeps the memory store clean without manual intervention.