Continuum

A collaborative workspace designed for human-AI thinking.

Real-time co-editing with AI • Multi-provider support • Run models locally • Full privacy control

Get Started • The Loom • Why Continuum • Privacy

---

What Is Continuum?

Continuum is a desktop workspace where you and AI write, research, and think together in real-time.

Think Notion meets Claude, but actually built for collaboration from scratch.

• The Loom — A real-time collaborative editor where you see AI typing alongside you, with full diff control over every change
• Project Spaces — Separate contexts for different work (research, writing, development) with persistent conversation history
• True Multi-Provider — OpenAI, Anthropic, Mistral, OpenRouter, or run models locally via llama.cpp
• Privacy-First — Run completely offline with local models, or use cloud APIs. Your choice, always.

Use it for:

• Technical writing and documentation
• Research and synthesis
• Creative writing and ideation
• Code planning and architecture discussions
• Personal knowledge management

---

Why Continuum?

Real-Time Collaboration, Not Turn-Taking

Watch the AI's cursor move. See thoughts form character by character. Accept, reject, or edit suggestions as they appear. This isn't chat with copy-paste—it's actual co-authoring.

Every AI edit shows as a pending diff. You stay in control without breaking flow.

Separate Spaces for Different Thinking

Your novel doesn't belong in the same context as your startup research. Continuum gives you distinct project spaces, each with:

• Its own conversation threads
• Dedicated file references
• Persistent context across sessions
• Independent settings and tone

Switch projects and the workspace adapts. Clean mental boundaries.

Use Any Model—Cloud or Local

Cloud APIs:

• OpenAI (GPT-4o, GPT-4, o1)
• Anthropic (Claude 3.5 Sonnet, Opus, Haiku)
• Mistral (Large, Medium, Small)
• OpenRouter (300+ models)
• Custom OpenAI-compatible endpoints

Local Models:

• Run llama.cpp for complete offline operation
• Works with any GGUF model (Llama, Mistral, Qwen, etc.)
• Download models directly from HuggingFace through the UI
• Zero telemetry, zero data leaving your machine

Configure once, switch between providers instantly. The architecture is model-agnostic.

---

The Loom

The Loom is Continuum's real-time collaborative editor. This is where you work with AI, not just prompt it.

How it works:

1. You and the AI share a cursor in the same document
2. AI edits stream in character-by-character (watch it type)
3. Every change appears as a reviewable diff
4. Accept, reject, or manually edit any suggestion
5. Enable auto-accept for full creative flow mode

Features:

• Native Markdown with live preview
• Drag-and-drop file imports (images, docs, references)
• Export to PDF with configurable formatting (headers, footers, page numbers)
• Version history via persistent storage
• Multiple documents per project

This isn't a document editor with AI bolted on. It's a shared workspace where both participants can write.

---

Conversation + Context

Chat isn't just for commands—it's for thinking out loud.

• Explore ideas before committing to the Loom
• Ask questions about your documents
• Branch into multiple conversation threads
• Resume any past discussion with full context

Every project maintains its own conversation history. Switch projects, switch contexts. Your thinking stays organized.

---

Privacy First

Your drafts, ideas, and thinking shouldn't live on someone else's servers by default.

Continuum gives you control:

• Run completely offline with llama.cpp (zero data leaves your machine)
• Or use cloud APIs when you need more capability
• Zero telemetry — No tracking, no phone-home, no analytics
• Local-first storage — All data stored as plain files on your disk

For local AI:

1. Install llama.cpp
2. Download any GGUF model (built-in HuggingFace browser in settings)
3. Point Continuum at your local server
4. Work offline with complete privacy

For cloud APIs:

• API keys stored locally, encrypted
• Configurable per-project
• Switch providers anytime

The architecture is built to support both. You decide where your data goes.

---

Get Started

Quick Start (Web)

git clone https://github.com/vanta-research/continuum.git
cd continuum
npm install
npm run dev

Open localhost:3000.

Desktop App

npm run electron:dev

The desktop app is a native Electron application (not a browser tab). Ships with One Dark theme and five accent colors (Blue, Green, Purple, Red, Yellow).

Setting Up Local AI (Optional)

For complete offline operation:

1. Install llama.cpp

git clone https://github.com/ggerganov/llama.cpp
cd llama.cpp
make

2. Run a model server

./llama-server --model path/to/model.gguf --port 8082 --ctx-size 4096

Recommended starter model: Atom-Olmo3-7B

3. Configure Continuum

• Settings → API Keys → Add llama.cpp endpoint
• Or set LLAMA_SERVER_URL=http://localhost:8082 in .env.local

You can also download models directly through Continuum's UI (Settings → Download Models).

---

Configuration

Environment Variables

Optional .env.local:

LLAMA_SERVER_URL=http://localhost:8082

In-App Settings

Access via Settings menu:

• API Keys — Configure OpenAI, Anthropic, Mistral, OpenRouter (stored locally, encrypted)
• Model Selection — Choose which models appear in your dropdown across all providers
• Download Models — Browse and download GGUF models from HuggingFace
• General — Temperature, max tokens, accent color
• Project Settings — Per-project model and provider configuration

---

Building

Web

npm run build && npm start

Desktop

npm run electron:build        # Current platform
npm run electron:build:all    # All platforms

---

Under the Hood

Tech Stack

• Next.js 16 (App Router)
• React 19
• TypeScript
• Tailwind CSS 4
• CodeMirror 6
• Electron
• shadcn/ui

Architecture

continuum/
├── app/                    # Pages and API routes
│   ├── api/
│   │   ├── chat/          # AI completions (streaming)
│   │   ├── projects/      # Workspace management
│   │   ├── models/        # Model management
│   │   └── settings/      # Preferences
│   └── ...
├── components/
│   ├── loom/              # The Loom editor
│   ├── projects/          # Workspace UI
│   └── ui/                # Shared components
├── lib/                   # Core utilities
└── electron/              # Desktop wrapper

API Reference

Chat

POST /api/chat                              # Streaming AI completion

Projects (Workspaces)

GET|POST        /api/projects               # List / Create
GET|PUT|DELETE  /api/projects/[id]          # Read / Update / Delete

Sessions (Threads)

GET|POST        /api/projects/[id]/sessions
PUT|DELETE      /api/projects/[id]/sessions/[sid]

Files

GET|POST        /api/projects/[id]/files
PATCH|DELETE    /api/projects/[id]/files/[fid]

Models

GET   /api/models/available                 # Browse HuggingFace
GET   /api/models/local                     # List downloaded
POST  /api/models/download                  # Download model
GET   /api/models/openai                    # List OpenAI models
GET   /api/models/anthropic                 # List Anthropic models
GET   /api/models/mistral                   # List Mistral models
GET   /api/models/openrouter                # List OpenRouter models

---

Requirements

• Node.js 20+
• npm
• For local AI: llama.cpp + GPU (recommended)

---

Roadmap

Shipped:

• Real-time collaborative editing (The Loom)
• Multi-provider support (OpenAI, Anthropic, Mistral, OpenRouter)
• Local model support (llama.cpp integration)
• PDF export with formatting options
• Model selection and download UI

In Development:

• Voice input/output
• Plugin system for extensibility
• Collaborative multiplayer spaces
• Mobile apps (iOS/Android)
• Web clipper for research
• Enhanced RAG for long documents

---

Contributing

Built by VANTA Research.

Contributions welcome. Fork, branch, PR.

Development:

• Report bugs via Issues
• Discuss features in Discussions
• See architecture details in the Tech Stack section

---

License

MIT License — Use it, modify it, ship it.

---

Continuum is a space for thinking with AI, not just prompting it.

GitHub • Issues • Discussions