ltcai 0.2.2

Lattice AI local MLX/cloud LLM workspace server

Private local AI workspace that turns your files, chats, and folders into a searchable knowledge graph.

---

Why Lattice AI?

Most AI tools forget everything after each conversation. Your files sit in folders, your chats vanish, and nothing connects.

Lattice AI remembers. It reads your local files, indexes your conversations, and builds a knowledge graph that links people, projects, concepts, and documents — all on your machine, with zero data leaving your PC.

• Your data stays local — everything lives in ~/.ltcai/, never sent to external servers
• Your AI gets smarter over time — every chat and file builds your personal knowledge graph
• One install, works everywhere — web UI, VS Code, Telegram, MCP clients, all connected to the same brain

---

3-Minute Workflow

1. Install             pip install ltcai && LTCAI
2. Detect hardware     Auto-detect CPU, GPU, RAM → recommend the best local model
3. Connect folders     Select local folders to index into your knowledge graph
4. Build knowledge     Files and chats auto-analyzed → nodes (people, concepts, files) + edges (mentions, contains, depends on)
5. Ask anything        "What did I discuss about the auth migration last week?" → Graph RAG retrieves context
6. Work from anywhere  Web UI · VS Code · Telegram · MCP — all connected to the same knowledge

---

Product Preview

Workspace Chat Chat with local/cloud LLM, upload files, pipeline controls

Knowledge Graph Auto-built from chats and documents — nodes = nouns, edges = verbs

Admin Dashboard User management, audit log, security monitoring

---

Quick Start

Python / PyPI

pip install ltcai
pip install "ltcai[local]"   # + Apple Silicon MLX models
LTCAI
# → http://localhost:4825

Node / npm

npm install -g ltcai
LTCAI

VS Code / Cursor

1. Install Lattice AI from VS Code Marketplace or Open VSX
2. Start the local server: LTCAI
3. Cmd+Shift+A to open the chat panel

First run: open http://localhost:4825 → sign up → first account auto-becomes admin → pick a model → start chatting.

---

How the Knowledge Graph Works

Lattice AI automatically analyzes your chats and files, extracting meaningful structure:

Nodes (nouns) — the things in your world:

Type	Examples
Document	PDF, PPTX, DOCX, code files, images
Person	You, mentioned colleagues
Concept	Technologies, frameworks, ideas
Chat	Conversation sessions
Task	Action items, TODOs
Decision	Choices made in discussions

Edges (verbs) — how things relate: mentions · contains · resolves · depends on · explains · uses · replaces · supports · related to

Local folder indexing:

1. Browse your drives and folders from the UI
2. Preview file counts, types, and sizes before indexing
3. Approve which folders to connect (sensitive files auto-excluded)
4. Files are parsed, chunked, and linked into the graph
5. Optional: enable file watcher for real-time updates

All data stays in a local SQLite database. Nothing leaves your machine.

---

Comparison

Based on public product behavior as of 2026-05.

	Lattice AI	Open WebUI	Continue.dev	GitHub Copilot
Local model (offline, Apple Silicon)	Yes	Yes	Yes	No
Cloud models (OpenAI, Groq...)	Yes	Yes	Yes	Yes
Knowledge graph (auto from files + chats)	Yes	No	No	No
Local folder indexing + file watcher	Yes	No	No	No
VS Code extension	Yes	No	Yes	Yes
Telegram bot	Yes	No	No	No
MCP registry (one-click install)	Yes	Partial	Yes	No
Admin + audit log	Yes	Yes	No	No
Zero telemetry, self-hosted	Yes	Yes	Yes	No
One-command public tunnel	Yes	No	No	No
Free	Yes	Yes	Yes	No

---

Supported Models

Local (Apple Silicon MLX):

Model	Best for	Size	Min RAM
Qwen3-VL 4B	Multimodal / low spec	~2.7 GB	8 GB
Qwen3-VL 8B	Multimodal / balanced	~4.8 GB	16 GB
GPT-OSS 20B	Reasoning / open-weight	~12.1 GB	32 GB
Gemma 4 26B	Multimodal / large	~15.6 GB	32 GB
Gemma 4 31B	Multimodal / latest Gemma 4	~18.4 GB	48 GB
Qwen3-VL 30B A3B	Multimodal / top	~18 GB	48 GB
GPT-OSS 120B	Reasoning / top open-weight	~62.3 GB	128 GB
Phi 4 Mini	Coding (fast)	~2.2 GB	8 GB
Llama 3.1 8B	General	~4.7 GB	8 GB
Mistral 7B v0.3	General / Apache	~4.1 GB	8 GB

Cross-platform (Ollama / LM Studio / vLLM / llama.cpp): Same models via Ollama pull, LM Studio download, vLLM serve, or llama.cpp GGUF where available.

Cloud (any platform): OpenAI GPT-5.5 · Claude Opus 4.7 / Sonnet 4.6 / Haiku 4.5 via OpenRouter · Groq · Together · xAI · any OpenAI-compatible endpoint

The setup wizard auto-detects your hardware and recommends the best model for your specs.

---

Data Privacy

Storage	All data in ~/.ltcai/ on your machine
Telemetry	None — no analytics, no tracking, no phoning home
File access	Approval-token gated — explicit consent per folder
Cloud models	When using cloud APIs, prompts are sent to the provider. Local models keep everything offline.
Sensitive files	.env, credentials, keys, certificates auto-excluded from indexing
Delete	Clear chat history, delete graph nodes, remove indexed folders at any time

---

All Features

Core Experience

Feature	Description
Web UI	Responsive chat, file upload, model picker, knowledge graph
Auto Setup Wizard	Detect hardware → recommend model → install dependencies → verify
Graph RAG	Chats and files auto-indexed into SQLite knowledge graph
Local folder indexing	Browse, audit, and index local folders with file watcher

Developer Tools

Feature	Description
VS Code / Cursor	Chat panel, Edit Selection, Explain, Generate command
Multi-step agent	File edit/create, grep, todo, terminal (25 steps, human-in-the-loop)
Multi-LLM pipeline	Plan → Execute → Review with different models
MCP server	Use Lattice tools in Claude Desktop / Cursor
MCP registry	One-click install from registry.modelcontextprotocol.io
Skills marketplace	77 official skills (Anthropic + verified third-party)
Plugin directory	Browse 149 open-source plugins

Access & Communication

Feature	Description
Telegram bot	Chat, upload files, manage models from anywhere
PWA	Install on iPad / Android home screen
Public tunnel	LTCAI --tunnel — Cloudflare HTTPS, no account needed

Administration

Feature	Description
User management	Roles, permissions, disable/enable accounts
SSO	Entra ID / Okta OIDC
Audit dashboard	Per-user AI usage, sensitive data detection, event log, TXT/CSV/Excel export
Security monitoring	Rate limits, file access approvals, MCP install audit trail

Security

Property	Detail
Binding	Default 127.0.0.1:4825 — local only
Auth	Session required when network-exposed or public mode
Cookies	HttpOnly + SameSite=Lax — no localStorage token
Local file access	Approval-token gated (path + user + action scope)
Package install	Admin-only with audit trail (MCP, skills, pip, npm)
CORS	Localhost only by default; configure via LATTICEAI_CORS_ALLOWED_ORIGINS
File upload	Magic-number signature check (blocks extension spoofing)
Rate limits	/chat 30/min · /agent 6/min · /upload 12/min per user
Telemetry	None — all data in ~/.ltcai/

Report vulnerabilities: SECURITY.md

Setup & Configuration

VS Code shortcuts

Shortcut	Action
Cmd+Shift+A	Open chat
Cmd+Shift+E	Edit selected code
Cmd+Shift+M	Load / switch model
Right-click	Explain / Save to Knowledge Garden

Telegram bot

LATTICEAI_TELEGRAM_BOT_TOKEN=your-token LTCAI

Public server (Docker / Render / Fly.io)

LATTICEAI_MODE=public \
LATTICEAI_PUBLIC_MODEL=openai:gpt-4o-mini \
OPENAI_API_KEY=sk-... \
LATTICEAI_INVITE_CODE=my-secret \
LTCAI

Public tunnel (Cloudflare, no account)

LTCAI --tunnel
# → https://xxxx.trycloudflare.com

Auto-start (Mac)

cat > ~/Library/LaunchAgents/com.ltcai.plist << 'EOF'
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
  <key>Label</key><string>com.ltcai</string>
  <key>ProgramArguments</key><array><string>/usr/local/bin/LTCAI</string></array>
  <key>RunAtLoad</key><true/>
  <key>KeepAlive</key><true/>
  <key>StandardOutPath</key><string>/tmp/ltcai.log</string>
  <key>StandardErrorPath</key><string>/tmp/ltcai.err</string>
</dict>
</plist>
EOF
launchctl load ~/Library/LaunchAgents/com.ltcai.plist

API Reference

Method	Path	Description
GET	/health	Server status & current model
GET	/models	Model list + load state
POST	/models/load	Load a model
POST	/chat	Chat (stream=true/false)
POST	/agent	Multi-step file agent
GET	/knowledge-graph/stats	Graph statistics
GET	/knowledge-graph/search?q=	Search the knowledge graph
GET	/knowledge-graph/local/roots	Discover local drives & folders
POST	/knowledge-graph/local/audit	Audit a folder before indexing
POST	/knowledge-graph/local/index	Index a folder into Graph RAG
GET	/mcp/installed	Installed MCP servers
POST	/mcp/install	Install MCP server (admin)
GET	/skills/marketplace	Skills marketplace
POST	/skills/install	Install a skill (admin)
GET	/admin/audit	Audit report
GET	/permissions/pending	Pending file-access approvals

Full reference: docs/mcp-tools.md

Troubleshooting

Symptom	Fix
Port 4825 in use	lsof -i :4825 → kill <PID> or LTCAI --port 4826
ModuleNotFoundError: mlx	pip install "ltcai[local]" (Apple Silicon only)
Python < 3.11	Upgrade Python: python3 --version
No API key warning	OPENAI_API_KEY=sk-... LTCAI or set in admin panel
Can't reach from iPad	LATTICEAI_HOST=0.0.0.0 LTCAI or use --tunnel

---

Platform Support

Feature	macOS Apple Silicon	macOS Intel / Windows / Linux
Web UI + cloud models	Yes	Yes
VS Code / Cursor extension	Yes	Yes
Telegram bot	Yes	Yes
MLX local models	Yes	--
Ollama / LM Studio / vLLM	Yes	Yes

---

Distribution

Channel	Link
PyPI	pypi.org/project/ltcai
npm	npmjs.com/package/ltcai
VS Code Marketplace	marketplace.visualstudio.com
Open VSX	open-vsx.org

Current version: 0.2.2 — Changelog

---

Contributing

See CONTRIBUTING.md. All PRs welcome.

License

MIT — TaeSoo Park

---

한국어 안내 (Korean)

Lattice AI

내 PC의 파일, 대화, 프로젝트를 기억하고 연결하는 로컬 AI 워크스페이스

대부분의 AI 도구는 대화가 끝나면 모든 것을 잊습니다. Lattice AI는 다릅니다. 로컬 파일을 읽고, 대화를 기록하고, 사람·프로젝트·개념·문서를 연결하는 지식 그래프를 자동으로 만듭니다. 모든 데이터는 내 PC에만 저장됩니다.

3분 사용 흐름

1. 설치          pip install ltcai && LTCAI
2. 하드웨어 감지   CPU, GPU, RAM 자동 감지 → 최적 로컬 모델 추천
3. 폴더 연결      로컬 폴더를 선택하여 지식 그래프에 연결
4. 지식 구축      파일과 대화 자동 분석 → 점(사람, 개념, 파일) + 선(언급함, 포함함, 의존함)
5. 질문           "지난주 인증 마이그레이션 논의 내용은?" → Graph RAG가 컨텍스트 검색
6. 어디서든 작업   웹 UI · VS Code · Telegram · MCP — 같은 지식에 연결

설치

pip install ltcai                    # 클라우드 모델
pip install "ltcai[local]"           # + Apple Silicon MLX 로컬 모델
LTCAI                                # 서버 실행 → http://localhost:4825
LTCAI --tunnel                       # + Cloudflare 공개 URL 자동 발급

핵심 차별점

• 내 데이터가 AI의 기억이 된다 — 채팅과 파일이 자동으로 지식 그래프로 구조화
• 로컬 폴더를 연결하면 프로젝트 전체를 이해 — 파일 변경 시 실시간 업데이트
• 모든 데이터는 내 PC에 — ~/.ltcai/에 저장, 텔레메트리 없음, 외부 전송 없음
• 설치 한 번으로 어디서든 — 웹 · VS Code · Telegram · MCP 클라이언트

추천 로컬 모델 (M-series Mac)

모델	용도	크기	최소 RAM
Qwen3-VL 4B	멀티모달 / 저사양	~2.7GB	8GB
Qwen3-VL 8B	멀티모달 / 균형 추천	~4.8GB	16GB
GPT-OSS 20B	추론 / 오픈가중치	~12.1GB	32GB
Gemma 4 26B	멀티모달 / 대형	~15.6GB	32GB
Gemma 4 31B	멀티모달 / 최신 Gemma 4	~18.4GB	48GB
Qwen3-VL 30B A3B	멀티모달 / 최고급	~18GB	48GB
GPT-OSS 120B	추론 / 최고급 오픈가중치	~62.3GB	128GB

자세한 내용: docs/CHANGELOG.md · 보안 · 기여