Search

PythonReactSEOPDF
GitHub Skill

openmemory

Persistent long-term agent memory for storing and querying past work, patterns, and learnings.

5dlabs☆ 1 Repository stars0 Repository forks

Installation

Follow the repository instructions for this skill and your agent. Review the files and dependencies before adding it to your workflow.

View source ↗Guides

Skill instructions

OpenMemory (Persistent Agent Memory)

You have access to OpenMemory for persistent long-term memory across sessions.

Memory Tools

Tool Purpose
openmemory_query Semantic search across memories by similarity
openmemory_store Store new memories with sector classification
openmemory_list List recent memories for a user/agent
openmemory_get Retrieve specific memory by ID
openmemory_reinforce Boost salience of important memories

Memory Sectors

Memories are classified into sectors:

Sector Use Case Example
episodic Events, task history "Implemented auth flow for project X"
semantic Facts, learned patterns "Always add Context7 lookup before Rust implementation"
procedural How-to knowledge "Steps to deploy with ArgoCD"

Usage Patterns

Before starting a task:

openmemory_query({ query: "similar implementations", sector: "episodic" })

After completing a task:

openmemory_store({ 
  content: "Implemented OAuth2 with PKCE for React app using Effect",
  sector: "episodic",
  tags: ["auth", "react", "effect"]
})

For important learnings:

openmemory_reinforce({ memory_id: "mem_xyz", boost: 1.5 })

Best Practices

  1. Query before implementing - Check for similar past work
  2. Store after completing - Save successful patterns and solutions
  3. Reinforce important memories - Boost salience of critical learnings
  4. Tag memories well - Include relevant technologies and patterns

Examples

Read the usage examples in the source repository.

View source ↗

Source history

Indexed on
2026-01-20
Index updated
2026-03-04
Branch
main

Index and sync dates describe this directory. View the source repository for code history.

GitHub ↗

chat Comments (0)

chat_bubble_outline

No comments yet. Be the first to share your thoughts!