Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.linksee.app/llms.txt

Use this file to discover all available pages before exploring further.

​
What is Linksee Memory?

Linksee Memory is a local-first MCP server that gives every AI agent on your machine persistent, structured memory. Sessions end. Agents forget. Linksee Memory fixes that. It stores not just what happened, but why it failed, how you fixed it, and what to never do again — organized across 6 cognitive layers, shared via a single SQLite file.

Cross-agent

Claude Code, Cursor, Windsurf, OpenAI Codex, Gemini CLI — one memory, all agents.

Token-saving

AST-aware file diff cache. Up to 99% token reduction on re-reads.

WHY-layered

6 layers: goal / context / emotion / implementation / caveat / learning.

Local-first

No cloud. No account. One SQLite file on your machine.

​
Quick Start

Install with a single command: 
npm install -g linksee-memory
Add to your MCP client config:
claude_desktop_config.json
{
  "mcpServers": {
    "linksee-memory": {
      "command": "linksee-memory"
    }
  }
}
Restart your agent. Done.
Linksee Memory works with any MCP-compatible client — Claude Desktop, Claude Code, Cursor, Windsurf, Cline, and more.

​
Key Features

  • 8 tools — remember, recall, update_memory, list_entities, forget, consolidate, recall_file, read_smart
  • 4 static resources — memory://stats, memory://hot, memory://recent, memory://caveats
  • 3 resource templates — memory://entity/{name}, memory://layer/{layer}, memory://memory/{id}
  • 5 prompts — session summarization, caveat extraction, recall-and-write discipline, entity handoff, weekly consolidation
  • Ebbinghaus forgetting curve — cold memories fade; caveats and goals are protected forever
  • AST-aware chunking — TS/JS/Python files are split by function/class, not arbitrary line counts
  • Bilingual — full Japanese + English support with trigram FTS5 tokenizer

​
How It Works

  1. During a session, the agent calls remember to store decisions, caveats, and context
  2. On the next session, the agent calls recall to retrieve relevant memories ranked by relevance, heat, and importance
  3. For file re-reads, read_smart returns only changed chunks — saving 50-99% tokens
  4. Over time, consolidate compresses cold memories into learning-layer summaries while preserving caveats

​
System Requirements

  • Node.js 20+
  • Any MCP-compatible client
  • ~10 MB disk space for the SQLite database

​
Next Steps

Installation

Detailed setup for every MCP client

Quick Start

Your first remember → recall flow in 2 minutes

Memory Layers

Understand the 6-layer structure

Tools Reference

Full API reference for all 8 tools