Agent Memory

AgentMemory Skill

Persistent memory system for AI agents

Overview

AgentMemory enables AI agents to retain information across sessions. It provides fact storage, experience learning, and entity tracking with zero external dependencies.

Key Features

• Fact storage via remember() with tagging capabilities
• Experience learning through learn() method
• Memory retrieval using recall() for facts and get_lessons() for experiences
• Entity tracking for people and projects
• Semantic search using FTS5
• Auto-cleanup for stale information

Installation

# Via ClawHub (recommended)
clawdhub install agent-memory

# Or clone
git clone https://github.com/openclaw/skills
cp -r skills/dennis-da-menace/agent-memory ~/.clawdbot/skills/

Storage

Default location: ~/.agent-memory/memory.db

Custom path supported via configuration.

Core API

Remember Facts

from memory import AgentMemory

mem = AgentMemory()

# Store a fact
mem.remember(
    "User prefers dark mode",
    tags=["preference", "ui"],
    confidence=0.9
)

# Store with entity link
mem.remember(
    "John's birthday is March 15",
    tags=["personal"],
    entity="john"
)

Learn from Experience

mem.learn(
    action="Deployed to production without tests",
    context="Friday evening rush",
    outcome="failure",
    insight="Always run tests before deploy, especially on Fridays"
)

Recall Memories

# Search facts
facts = mem.recall("user preferences")

# Get lessons
lessons = mem.get_lessons(context="deployment", outcome="failure")

Track Entities

mem.track_entity("john", {
    "role": "client",
    "company": "Acme Inc",
    "preferences": ["email", "brief updates"]
})

Integration Protocol

Recommended workflow for agent configuration:

memory_protocol:
  session_start:
    - Load recent lessons
    - Check entity context
  during_session:
    - Extract durable facts
    - Learn from outcomes
  session_end:
    - Update entity information
    - Record session summary

Dependencies

• Python 3.8+
• SQLite (built-in)

No external packages required.

License

MIT

Author

Dennis Da Menace - Built for the OpenClaw community