Open Canvas is an open-source web application designed to enhance the user experience for collaborating with Large Language Models (LLMs) in document writing, content creation, and coding. It targets users who want a more interactive and persistent way to iterate on existing content with AI assistance, offering features like memory, custom quick actions, and artifact versioning.

How It Works

Open Canvas leverages LangGraph for its agentic backend, enabling complex, stateful interactions with LLMs. A key differentiator is its built-in memory system, which stores user insights and style rules to personalize future interactions. The application supports starting sessions with existing documents (code or markdown) rather than just blank prompts, facilitating iterative refinement. Artifacts are versioned, allowing users to revert to previous states.

Quick Start & Requirements

• Install: git clone the repository, cd open-canvas, yarn install.
• Prerequisites: OpenAI API key (required), Anthropic, Google GenAI, Fireworks AI, Groq AI, FireCrawl, ExaSearch API keys (optional). Supabase account for authentication. LangGraph CLI and LangSmith for observability.
• Setup: Copy .env.example to .env in root and apps/web/, configure API keys and Supabase credentials. Run yarn build, then cd apps/agents and yarn dev for the LangGraph server. Finally, cd apps/web and yarn dev for the frontend.
• Links: Try it out, YouTube Guide, LangSmith Studio UI

Highlighted Details

• Supports multiple LLM providers including OpenAI, Anthropic, and Fireworks AI.
• Local Ollama models can be integrated by setting environment variables.
• Live markdown rendering and editing within the artifact view.
• Artifact versioning allows for time-travel debugging and history tracking.
• Customizable quick actions for common tasks.

Maintenance & Community

The project is actively developed by the LangChain AI team. Contributions are welcomed, with specific labels for frontend, AI, and full-stack issues. Contact information for contributions is provided via email.

Licensing & Compatibility

The project is MIT licensed, permitting commercial use and integration with closed-source applications.

Limitations & Caveats

Local LLMs may exhibit reduced instruction-following capabilities compared to proprietary models, potentially leading to errors. The setup process requires multiple API keys and services, which may incur costs.