ltcai 2.2.1

Lattice AI Workspace OS for local-first graph, memory, agent, workflow, and skill operations

Lattice AI

A local-first AI workspace — Plan → Execute → Review with multiple LLMs, on top of a durable Knowledge Graph.

Apple Silicon (MLX) local model management · Personal & Organization workspaces · Multi-agent workflows · SSO for teams

---

Lattice AI turns your files, documents, images, screenshots, conversations, decisions, notes, and work history into linked knowledge. AI then works on top of that Knowledge Graph to advise, analyze, generate documents, and automate work.

It is built around a simple product rule:

Do not control the user in the name of protection. Explain clearly, disclose sources and risks, then let the user decide.

What Lattice AI is

Lattice AI is four things working together, not an operating system and not a simple chat front-end:

• Local-first AI Workspace — runs on your machine first; your data and graph stay local by default.
• AI Pipeline Platform — a Plan → Execute → Review loop that can route across multiple LLMs for each stage.
• Knowledge Graph Platform — multimodal inputs become entities, relationships, evidence, and artifacts that outlive any single model.
• Workspace Platform — Personal and Organization workspaces, role-based access, and SSO for teams.

Highlights

• Multimodal ingestion — accepts PDFs, Word, spreadsheets, slide decks, images, screenshots, notes, code, web content, conversations, and work logs without asking you to pre-convert them.
• Knowledge Graph core — extracts entities, relationships, evidence, decisions, and artifacts so your work memory survives model changes.
• Multi-agent workflow — agents hand off work with structured context packets, review/retry loops, and replayable timelines.
• Local model management — MLX-VLM on Apple Silicon, with current-generation multimodal model recommendations based on your CPU, GPU, RAM, storage, and OS.
• Source disclosure — every recommended model shows its facts in plain language before use: maker country, maker company, run mode (local/cloud), internet requirement, and model name.
• Basic / Advanced / Admin modes — the same features everywhere; the modes differ only in how much is explained. Admin mode adds authority (user management, permissions, audit logs, org/security policy, model approval).

v2.2.1 — Frontend & UX overhaul

v2.2.1 is a UX-focused release. No features were removed; the interface was re-laid out and re-themed.

• Responsive, mobile-first UI — phone, tablet, laptop, desktop, ultrawide, and 4K. Content is re-laid out for each size, never hidden.
• Light / Dark themes via design tokens — a single source of truth in static/css/tokens.css. :root holds light values, [data-lt-theme="dark"] holds dark values. The theme toggle, OS preference detection, and persistence live in static/scripts/ux.js.
• Accessibility — 44px touch targets, :focus-visible rings, a keyboard-safe chat composer (uses visualViewport insets), iOS no-zoom inputs, and reduced-motion support.
• Knowledge Graph UX — responsive canvas that re-fits on resize, zoom buttons, fullscreen, minimap, relationship filtering, a mobile graph↔card view, and a theme-aware palette.
• Admin UX — wide tables reflow to cards on mobile, with larger touch targets and full dark/light support.
• File UX — drag & drop and screenshot paste to attach. Model cards explain country, company, run mode, and internet use in plain language.

Knowledge Graph flow

files / documents / images / conversations / work history
  -> multimodal understanding
  -> entity extraction
  -> relationship extraction
  -> evidence storage
  -> Knowledge Graph update
  -> advice / analysis / document generation / automation

The graph preserves your work memory even when the model changes.

Local model policy

Lattice AI recommends current-generation multimodal models for local use and keeps text-only recommendation paths out of the product:

Family	Default role	Example current recommendation
Gemma 4	Default Google multimodal family	mlx-community/gemma-4-12b-it-4bit
Gemma 4 large	Higher-quality local multimodal work	mlx-community/gemma-4-31b-it-4bit
Qwen3-VL	Smaller, balanced multimodal options	mlx-community/Qwen3-VL-4B-Instruct-4bit
Llama 4	Meta multimodal option	mlx-community/Llama-4-Scout-17B-16E-Instruct-4bit

Low-spec machines use smaller or quantized multimodal models rather than older text-only models. See MODEL_POLICY.md for the full policy.

Requirements

• Python 3.11+
• macOS, Windows, or Linux for the workspace server
• Apple Silicon for local MLX-VLM model execution (optional; cloud models work without it)

Quick start

Lattice AI is a Python + FastAPI app. The CLI entry point is ltcai_cli.py.

npm install        # installs the npm package and dev tooling
npm run dev        # runs: python3 ltcai_cli.py --reload

Then open:

http://127.0.0.1:4825

Prefer Python directly?

python3 ltcai_cli.py
# host/port: --host 127.0.0.1 --port 4825

For local model execution on Apple Silicon, install the optional extra:

pip install "ltcai[local]"   # adds mlx-vlm

Useful commands:

npm test                # python3 -m pytest tests/ -v
npm run check:python    # py_compile across core modules
npm run build           # python3 -m build

VS Code extension

A companion VS Code extension lives in vscode-extension/.

cd vscode-extension
npm install
npm run build
npm run package:vsix

Build & packaging

Lattice AI ships as a Python package (ltcai), an npm package (ltcai), and a VS Code extension.

python3 -m build        # Python wheel + sdist
npm pack                # npm tarball

Publishing is intentionally manual — no glob uploads. See RELEASE.md for the exact, version-scoped publish steps for PyPI, npm, the VS Code Marketplace, and Open VSX.

Documentation

• ARCHITECTURE.md — system shape and runtime policy
• docs/architecture.md — full architecture reference
• PROJECT_PRINCIPLES.md — product principles
• AI_PHILOSOPHY.md — how AI is used here
• MODEL_POLICY.md — local model recommendation policy
• KNOWLEDGE_GRAPH.md — graph model and behavior
• docs/MULTI_AGENT_RUNTIME.md — multi-agent runtime
• docs/WORKFLOW_DESIGNER.md — workflow designer
• docs/REALTIME_COLLABORATION.md — realtime collaboration
• docs/ENTERPRISE.md — Organization workspaces & SSO (Okta / Microsoft Entra ID)
• docs/security-model.md — security model
• docs/PLUGIN_SDK.md — plugin SDK
• RELEASE_NOTES.md · docs/CHANGELOG.md

Release history

Version	Theme
2.2.1	Frontend & UX overhaul — responsive, design-token theming, accessibility, graph/admin UX
2.2.0	Multimodal-first Knowledge Graph, source disclosure, Gemma-4 recommendation policy
2.1.0	Agent platform maturity
2.0.0	Agentic workspace platform
1.7.0	Graph and collaboration
1.6.0	Product experience deepening

License

MIT