openarche

OpenArche

Local developer memory plugin for Claude Code. Automatically learns from your conversations and recalls relevant knowledge when you need it — fully local, zero cloud dependency.

Why

OpenArche solves three problems every developer hits with Claude Code:

1. No memory between sessions — every conversation starts from scratch, repeating the same context every time
2. Re-solving the same problems across projects — a solution you figured out in project A is forgotten by the time you're in project B
3. No personalization — Claude never learns your tech preferences, your decisions, or the traps you've already fallen into

OpenArche fixes all three by silently building a local knowledge base from your conversations, and surfacing the right knowledge exactly when you need it.

Features

• Auto-extract — after each conversation, extracts reusable insights via Claude Haiku and stores them as local Markdown files
• Auto-inject — before each prompt, vector-searches relevant memories and injects them as hidden context for Claude (invisible to you, visible to Claude)
• Smart reranking — supports local weighted reranking (similarity+quality+recency+frequency) or remote rerank API
• Knowledge graph — memories link to each other bidirectionally; retrieval expands through BFS graph traversal for richer context
• StatusLine — real-time memory count and last match in the Claude Code status bar

Getting started

1. Add the plugin:

/plugin marketplace add Done-0/openarche

2. Install:

/plugin install openarche

Then reload the plugin:

/reload-plugin

3. Run setup:

/openarche:setup

Setup handles all configuration automatically and optionally bootstraps from your conversation history. Downloads the local embedding model (~120MB) on first run — one time only.

4. Done. OpenArche runs fully automatically in the background.

Updating

To update OpenArche to the latest version:

/plugin update openarche
/reload-plugin

Commands

Command	Description
/openarche:setup	Initialize plugin, batch-extract memories from all conversation history
/openarche:extract	Manually trigger extraction from qualifying conversation history
/openarche:config	View or update configuration
/openarche:save	Manually save an insight from the current conversation
/openarche:list	List recently stored memories
/openarche:search	Search the memory library with a query
/openarche:forget	Delete a memory by ID
/openarche:reindex	Rebuild vector index after switching embedding models

Configuration

Run /openarche:config to view and change settings interactively.

Config file: <home>/.claude/openarche/config.json

{
  "embedding": {
    "provider": "local",
    "localModel": "Xenova/multilingual-e5-small",
    "remoteModel": "",
    "remoteApiKey": "",
    "remoteBaseUrl": ""
  },
  "retrieval": {
    "threshold": 0.73,
    "topK": 3,
    "maxInjectChars": 3000,
    "reranking": {
      "enabled": false,
      "provider": "local",
      "remoteModel": "",
      "remoteApiKey": "",
      "remoteBaseUrl": "",
      "weights": {
        "similarity": 0.7,
        "quality": 0.2,
        "recency": 0.05,
        "frequency": 0.05
      }
    }
  },
  "extraction": {
    "model": "claude-haiku-4-5-20251001",
    "minQualityScore": 0.6,
    "bootstrapConcurrency": 3
  }
}