Ori-Mnemos

Overview

Ori Mnemos implements a sophisticated knowledge graph-based memory system that mimics human cognition using mathematical models. It features activation decay, spreading activation along wiki-links, Hebbian co-occurrence learning, and reinforcement learning on retrieval patterns. The system operates locally with zero cloud dependencies, storing data in markdown files with SQLite indexing. Its Recursive Memory Harness framework (RMH) treats memory retrieval as navigation rather than search, enabling multi-hop reasoning through graph traversal and sub-question decomposition.

Try asking AI

After installing, here are 5 things you can ask your AI assistant:

• add_note

  Add a new note to the memory vault

  Try:Add a note about my meeting with John todayTry:Create a new note about the Q3 sales strategy
• search_memory

  Search through the memory vault for relevant information

  Try:Find information about the marketing campaignTry:Search for notes related to product development
• update_note

  Update an existing note in the memory vault

  Try:Update the note with my meeting notes to add action itemsTry:Revise the project timeline note
• explore_topic

  Perform recursive exploration of a topic using sub-question decomposition

  Try:Explore the topic of AI ethics in our industryTry:Deep dive into competitor analysis
• get_graph_stats

  Get statistics about the knowledge graph structure

  Try:Show me information about my knowledge graphTry:Get metrics on wiki-link connections
• promote_note

  Classify and promote a note from inbox to appropriate memory space

  Try:Promote this idea to my knowledge baseTry:Classify this note as a decision
• get_identity

  Retrieve the agent's stored identity information

  Try:What is my agent's identity?Try:Show me my stored goals and methodology
• list_notes

  List notes in a specific memory space with filtering options

  Try:Show me all my notes in the knowledge spaceTry:List recent operations notes
• get_note

  Retrieve a specific note by ID or title

  Try:Get my note about the quarterly reviewTry:Show me the note with ID 12345
• delete_note

  Delete a specific note from the memory vault

  Try:Remove the outdated meeting notesTry:Delete the draft that's no longer needed
• create_wiki_link

  Create a wiki-link connection between two notes

  Try:Link my product notes to the strategy documentTry:Connect this research to the relevant project
• prune_memory

  Analyze and archive notes based on activation topology

  Try:Clean up outdated notes from my memoryTry:Prune low-value notes to optimize storage

Comparable tools

Installation

npm install -g ori-memory
ori init my-agent
cd my-agent

{
  "mcpServers": {
    "ori": {
      "command": "ori",
      "args": ["serve", "--mcp", "--vault", "/path/to/brain"],
      "env": { "ORI_VAULT": "/path/to/brain" }
    }
  }
}

FAQ

How does Ori compare to other memory systems?

Ori significantly outperforms systems like Mem0 in benchmarks, with 3.1× better recall, 2.1× higher F1 scores, and 9.5× faster latency—all using local markdown files and SQLite, no cloud services required.

What storage backends does Ori support?

Ori uses markdown files on disk for storage with SQLite for indexing and vector embeddings. It has no database lock-in and works with any filesystem.

On Hacker News

Recent discussion from the developer community.

• Ori Mnemos – an open-source persistent memory for AI agents▲ 2💬 1
  Story by starro____ · 2026-03-03
• 3AM Coding:cracking persistent open-source memory for agents▲ 1
  Story by starro____ · 2026-03-09
• Open Source, Decentralized Memory Layer for AI Agents▲ 1
  Story by starro____ · 2026-03-04

Compare Ori-Mnemos with