Open Canvas (Local-First Edition)

Open Canvas is an open source web application for collaborating with agents to better write documents. This fork is configured for a fully local, privacy-focused stack using PostgreSQL + pgvector and Ollama as the primary LLM provider.

Key differences from the upstream version:

• All data stays local (PostgreSQL + pgvector)
• Ollama is the default LLM provider (no API keys required)
• No Supabase or cloud authentication dependencies
• Direct access to the canvas interface (no login required)

Features

• Memory: Built-in reflection agent stores style rules and user insights in a shared memory store across sessions.
• Custom quick actions: Define your own prompts that persist across sessions, invoked with a single click.
• Pre-built quick actions: Common writing and coding tasks are always available.
• Artifact versioning: Travel back in time and see previous versions of your artifact.
• Code, Markdown, or both: View and edit both code and markdown artifacts.
• Live markdown rendering & editing: View rendered markdown while editing.
• Local vector search: Semantic document search powered by Ollama embeddings and pgvector.

Quick Start

Prerequisites

• Docker & Docker Compose
• Node.js 18+
• Yarn (v4.9.2)
• Ollama installed locally (or via Docker)

Setup

1. Clone the repository:

git clone https://github.com/langchain-ai/open-canvas.git
cd open-canvas

2. Start infrastructure services:

docker-compose up -d postgres ollama

3. Pull an Ollama model:

ollama pull llama3.3

4. Install dependencies:

yarn install

5. Copy environment files:

cp .env.example .env
cp apps/web/.env.example apps/web/.env
cp apps/agents/.env.example apps/agents/.env

6. Build the monorepo:

yarn build

7. Start the LangGraph server (Terminal 1):

yarn --cwd apps/agents dev

8. Start the frontend (Terminal 2):

yarn --cwd apps/web dev

9. Open http://localhost:3000 and start using Open Canvas.

For detailed setup instructions, see LOCAL_SETUP.md.

Available Ollama Models

The following models are pre-configured:

Model	Name	Description
Llama 3.3	ollama-llama3.3	Default model, good general purpose
Big Tiger 27B	ollama-big-tiger-27b-48k:latest	Large context window (48k tokens)
GPT OSS 32K	ollama-gpt-oss-32k:latest	OpenAI-compatible OSS model

To add more models: ollama pull <model-name>

Optional: Cloud Providers

You can optionally enable cloud LLM providers by setting the appropriate environment variables and feature flags:

• OpenAI: Set OPENAI_API_KEY and NEXT_PUBLIC_OPENAI_ENABLED=true
• Anthropic: Set ANTHROPIC_API_KEY and NEXT_PUBLIC_ANTHROPIC_ENABLED=true
• Google GenAI: Set GOOGLE_API_KEY and NEXT_PUBLIC_GEMINI_ENABLED=true
• Fireworks: Set FIREWORKS_API_KEY and NEXT_PUBLIC_FIREWORKS_ENABLED=true

Architecture

See ARCHITECTURE.md for the local-first architecture documentation.

Running Smoke Tests

yarn smoke

For offline testing:

OPEN_CANVAS_SMOKE=1 yarn workspace @opencanvas/agents test

Troubleshooting

• No text being generated: Clear the oc_thread_id_v2 cookie and refresh the page.
• 500 network errors: Ensure the LangGraph server is running on port 54367.
• "thread ID not found" errors: Check the LANGGRAPH_API_URL env variable.
• Model name is missing in config. error: Ensure customModelName is set in config.configurable.
• Ollama connection errors: Ensure Ollama is running (docker-compose up ollama or ollama serve).

Contributing

We welcome contributions! See the GitHub issues for feature requests and bug reports.

Labels: frontend (UI focused), ai (agent focused), fullstack (both).