๐ŸŽ™๏ธ 1M+ creators use this voice AI โ€” free tier, no CC required
Try ElevenLabs Free โ†’
Skip to content
codingbutvibes
BREAKING ยท AI TOOLS

Flowith Launches Canvas Cowork: The First Shared AI Canvas for Agents and Humans

A new open-source skill turns AI agent output into a live, collaborative visual canvas โ€” and it works with Claude Code, Cursor, Copilot, Windsurf, and 30+ other agents right now.

Published March 31, 2026 ยท 6 min read

The Short Version

Flowith Canvas Cowork is a new MIT-licensed skill that connects your AI coding agents to a shared real-time visual canvas on Flowith. Install with one command: npx skills add flowith-ai/canvas-cowork

It supports 30+ agents (Claude Code, Cursor, Copilot, Windsurf, and more), tree-based branching for comparing outputs, batch generation with parallel subagents, and rich multimodal controls. No API keys. No server. Just install and go.

What Just Launched

On March 31, 2026, Flowith shipped Canvas Cowork โ€” a skill that solves a problem every developer using AI agents has run into: agent output is invisible to the rest of your team. It lives in a terminal, a chat thread, or a local file. Canvas Cowork changes that.

When any connected agent generates output โ€” code, images, video frames, text โ€” it appears in real time on a shared visual canvas that your whole team can see, annotate, and branch off of. Think of it as a whiteboard where your AI agents are participants, not just tools running in the background.

The Key Features

30+ Agent Connections via WebSocket

Canvas Cowork uses WebSocket to connect to agents including Claude Code, Cursor, GitHub Copilot, Windsurf, Cline, Roo Code, Goose, Gemini CLI, and more. Any agent that supports the skill protocol can push output to the shared canvas. The list is already at 30+ on launch day.

Real-Time Collaborative Canvas

This is the headline feature. Multiple teammates and multiple AI agents all work on the same live canvas. One agent is generating image variations while another is working on code โ€” and your designer is watching it happen and can comment or redirect in real time. It's the closest thing to pair programming with an entire AI-augmented team.

Tree-Based Node Structure

Instead of a linear output stream, Canvas Cowork organizes generation into a tree. Every output is a node. You can fork any node to explore alternatives โ€” run the same prompt with different parameters, compare two image styles side by side, or branch a code solution in two directions. It's version control for AI generation.

Batch Generation with Parallel Subagents

Need 10 image variations, 5 video clips, or 3 different code approaches simultaneously? Canvas Cowork spawns parallel subagents to run batch jobs concurrently. Results populate the canvas as they complete โ€” no waiting for sequential runs.

Cross-Canvas Recall

Every canvas session is searchable. You can recall outputs from past sessions directly into your current canvas โ€” no digging through chat history or file system. This is particularly useful for creative projects where you want to revisit earlier directions or combine elements from different sessions.

Rich Multimodal Controls

For image and video workflows, Canvas Cowork exposes controls that most agent UIs don't surface: image-to-image, image-to-video, aspect ratio selection, and resolution settings. These aren't buried in a config file โ€” they're right in the canvas interface.

One Command to Install

$ npx skills add flowith-ai/canvas-cowork

That's it. No API keys. No server deployment. No infrastructure to manage. The skill connects to Flowith's hosted canvas infrastructure. MIT licensed and open source โ€” you can inspect, fork, or contribute to the code.

Why This Matters for Developer Teams

The AI coding workflow has a visibility problem. A developer using Claude Code is generating real output in their terminal, but their designer, PM, and teammates can't see it until it's done and shared. Canvas Cowork collapses that gap.

The tree-based branching model is particularly well-suited for the way real teams actually work with AI: you don't want one output, you want to explore a space of possibilities and converge on the best one. Being able to compare branches side-by-side on a shared canvas is a genuinely different way of working than chat-based AI tools.

And the parallel subagent support changes the economics of batch generation. If you need 20 image concepts for a product launch, running them in parallel on Canvas Cowork is faster and more visible than any other current approach.

About Flowith

Flowith (flowith.io) is an agentic AI workspace built around an infinite 2D canvas. With 1M+ users, 40+ integrated AI models, and Agent Neo running a 10M token context window, it's been one of the faster-growing AI productivity tools of 2025โ€“2026. It hit #1 Product of the Day on Product Hunt and holds a 4.7/5 rating. Canvas Cowork is the latest addition to its platform.

Flowith pricing starts free, with paid plans from around $9.90/month up to approximately $30/month for power users. The Canvas Cowork skill itself is free and open source โ€” you just need a Flowith account to connect it to.

Get Started:

Install Canvas Cowork in under 60 seconds. Works with Claude Code, Cursor, Copilot, Windsurf, and 30+ other agents.

Try Flowith Free โ†’

Frequently Asked Questions

What is Flowith Canvas Cowork?

Canvas Cowork is a skill (installable via npx skills add flowith-ai/canvas-cowork) that turns the output of any AI coding agent into a shared, real-time visual canvas hosted on Flowith. Teammates and AI agents can view, branch, and collaborate on the same canvas simultaneously. It supports images, video, text nodes, and tree-based branching for comparing different generation paths.

Which AI agents does Canvas Cowork support?

Canvas Cowork connects to 30+ agents via WebSocket including Claude Code, Cursor, GitHub Copilot, Windsurf, Cline, Roo Code, Goose, Gemini CLI, and more. Any agent that can call the skill's WebSocket endpoint can push output to the shared canvas.

Does Canvas Cowork require an API key or server?

No. Canvas Cowork is MIT open source and requires no API keys and no server deployment. You install it with a single npx command and connect to your Flowith workspace. The canvas runs on Flowith's infrastructure.

How do I install Canvas Cowork?

Run: npx skills add flowith-ai/canvas-cowork. Then connect it to your Flowith workspace and attach it to your AI agent. Full setup guide at codingbutvibes.com/learn/flowith-canvas-cowork-setup.

What does 'tree-based node structure' mean in Canvas Cowork?

Instead of a linear chat history, Canvas Cowork organizes agent outputs into a branching tree. You can fork any node to explore alternative generations, compare branches side-by-side, and recall outputs from past sessions via cross-canvas memory. This is especially powerful for image-to-image and image-to-video workflows where you want to iterate in parallel.

Is Flowith Canvas Cowork free?

The skill itself is MIT open source and free. Using it requires a Flowith account โ€” Flowith offers a free tier and paid plans starting at around $9.90/month. Canvas Cowork's batch generation and parallel subagent features may consume credits depending on your plan.

What is Flowith?

Flowith is an agentic AI workspace with an infinite 2D canvas, 1M+ users, 40+ AI models, and Agent Neo (10M token context window). It's been the #1 Product of the Day on Product Hunt and has a 4.7/5 rating. Canvas Cowork is its newest feature โ€” a shared canvas that lets AI agents and human teammates collaborate visually in real time.

Related Articles