Verified project record

basketikun/infinite-canvas

infinite-canvas is an open-source infinite-canvas workbench that unifies canvas orchestration, AI image generation, reference-image editing, a chat assistant, a prompt library and asset management in a single interface. The browser connects directly to a user-configured OpenAI-compatible endpoint for generation, while canvases, assets and API keys are stored locally in the browser by default.

Project overview

It replaces the usual juggling of separate canvas, generation, editing and prompt-management tools with one interface, and adds a plugin system with a TypeScript SDK plus MCP-based agent integration that lets Codex and Claude Code operate the current canvas directly.

Project type: MCP · Workflow · Image & Vision · Prompt Engineering
Use cases: Image Processing
Deployment: Refer to project documentation
License: MIT

Best for

Creators doing iterative visual exploration who want canvas arrangement, AI image generation, reference-image editing, chat assistance, prompt libraries and asset management in one interface instead of separate tools.
Developers who want to extend the canvas with custom node plugins via the TypeScript SDK or adapt generation API calls to relay or self-hosted services.
Users who prefer API keys, canvases, assets and generation records stored locally in the browser by default.

Key capabilities

Multi-canvas projects with node drag/zoom, connections, a minimap, undo/redo and import/export.
The browser connects directly to a user-configured OpenAI-compatible endpoint for text-to-image, image-to-image, reference-image editing, Q&A, audio and video generation.
Chat and generate images around selected and upstream nodes, inserting results back into the canvas.
A Local Canvas Agent connects Codex / Claude Code so agents operate the current canvas via MCP.
A plugin for the Codex app that auto-registers MCP and tries to launch the local agent.
Install, enable, update and uninstall remote node plugins via URL, plus a TypeScript SDK for building custom canvas node plugins.
Users can customize how image/video generation APIs are called to adapt to relay services or self-hosted services.
Seven open-source prompt sources built in, plus custom standard JSON sources; fetched by the browser front end and cached in IndexedDB.

Limitations and risks

The project is in development; there is no guarantee of historical data compatibility and local storage formats may change.
Local storage formats may be adjusted directly, which risks compatibility of user data stored in the browser; factor this into adoption decisions before relying on accumulated canvases and assets.

Getting started

Clone the repository (git clone git@github.com:basketikun/infinite-canvas.git), enter the directory (cd infinite-canvas), run docker compose up -d, then open http://localhost:3000 and configure an OpenAI-compatible Base URL and API Key. Setup difficulty is rated easy: the Docker Compose route brings up the web app on port 3000 and the only remaining step is entering the Base URL and API Key.

Evidence and sources

README
README
README
README
README

Source updated at: 2026-09-25T05:00:40Z

Open original GitHub project