AI Tools AI apps and workflows

AI tooling that’s not the Claude-centric stack. (For everything Claude, see /claude/.) This section covers the local LLM playground, the Gemini API for embeddings/multimodal/agentic experiments, and one-off utilities like the AI file labeler.

Tools
Tool Use Where it lives
LM Studio Local LLM playground on the Mac Studio (MPS) — quick prompts, model swap-testing, MLX experiments LM Studio app, sometimes serving on 192.168.8.180:1234
Claude Desktop (legacy notes) Older app-specific notes that didn’t fit under /claude/ See /claude/setup/ for the current setup
Gemini API Google Gemini API — embeddings, multimodal, occasional Claude alternative Key in ~/.config/google-api-key (per ~/Sync/ED/SECRETS.md)
AI File Labeler Auto-classify files in inbox folders using a small LLM One-off utility
AI Workflow Cross-tool patterns — when to reach for which model Reference page
AI File Labeler

Your custom tool: label <folder> to preview AI-generated names, label-apply <folder> to rename + tag metadata.

AI Workflow

The AI stack has several layers that work together. This page maps how they relate.

The Stack
Claude (claude.ai / Claude Desktop)
    ↓ MCP tools
Desktop Commander → file system, terminal
Life Archive MCP  → 278K personal documents, knowledge graph
Tana Local MCP    → knowledge workspace
Apple MCP         → Calendar, Contacts, Reminders
Google Calendar   → gcal events
Gmail             → email search

LM Studio (Mac Studio — 192.168.8.180:1234)
    ↓ OpenAI-compatible API
TypingMind, AnythingLLM, Cherry Studio, Chatbox

Life Archive Embed Server (port 1235)
    ↓ gte-Qwen2-7B on Apple MPS
Life Archive RAG pipeline → LanceDB vector search
Claude

Primary AI interface. Used via:

  • claude.ai — web/desktop for general work
  • Claude Desktop — with full MCP server suite for tool-augmented sessions
  • Claude Code / Cowork — for agentic file, code, and desktop tasks

Key MCP servers attached: Life Archive, Desktop Commander, Tana, Apple, Google Calendar, Gmail, Exa, Firecrawl, Obsidian.

Local Models (LM Studio)

LM Studio runs on the Mac Studio (192.168.8.180:1234) serving an OpenAI-compatible API. Any app that supports a custom OpenAI endpoint can use it.

Connected clients: TypingMind, AnythingLLM, Cherry Studio, Chatbox, LM Studio chat UI itself.

Current issue: Multiple large models loaded simultaneously causes memory pressure → unresponsiveness. Load one model at a time for heavy use.

Config note: config.py in Life Archive uses LMSTUDIO_URL — set to http://192.168.8.180:1234, not localhost, so it works from other machines.

Life Archive

Personal RAG system with 278K documents. Accessed via:

  • Claude MCP tools (life_archive_search, life_archive_entity_lookup, etc.)
  • HTTP API at http://192.168.8.180:8900
  • Web graph explorer at http://192.168.8.180:1313/kg/

Full documentation: Life Archive →

Which Tool for What
Task Tool
Research, writing, analysis Claude (claude.ai)
File system work, coding Claude Desktop + Desktop Commander MCP
Search personal documents Claude + Life Archive MCP
Manage Tana nodes Claude + Tana Local MCP
Local model experimentation LM Studio
Multi-model chat management TypingMind
Voice dictation superwhisper
Screen OCR TextSniper / Shottr
Quick text transforms Boop / TextBuddy
Claude Desktop

Cowork mode for file management, code, and automation. Can access your folders and run tools via MCP. claude.ai

Gemini API

Google’s AI — powers the label command for file renaming. Uses gemini-2.0-flash for speed + vision.

LM Studio

Run open-source AI models locally on your Mac. No internet required after download. lmstudio.ai

Edit Markdown files in /content/ · Auto-index regenerated every deploy · / or ⌘K to search