Open source · Self-hosted · Single-admin

Self-hosted
AI creation canvas

Text, images, video, audio and agents share one infinite canvas. Get the feel of a commercial canvas at the lowest possible cost — no expensive subscription, and no need to chase the most expensive models. The choice stays with you.

macOS / Linux · Docker Compose v2
mkdir agenvas && cd agenvas && \
  curl -fL https://raw.githubusercontent.com/grayrepo-byte/Agenvas/main/docker-compose.yml \
    -o docker-compose.yml && \
  docker compose up -d --wait

When it is up, open http://127.0.0.1:8088/setup to create the admin account. The database password and credential key are generated automatically on first start — nothing to fill in.

  • 4 UI languages
  • 6 Media adapters
  • 1 Command to deploy
  • MIT Open source licence
Product preview

One canvas for the whole creative flow

Every asset is a card. Drag, box-select, connect and auto-arrange — text, images, video, audio and agents all reference each other on the same canvas.

Interface mock-up: the prompt card spawns text and images, the agent reads its bound context and submits media proposals, and approved proposals produce audio and video.

Core features

From first idea to finished asset, all on the canvas

Generation, reuse, editing and approval all happen on one canvas, so you never shuttle assets between tools.

Canvas creation

Drag, box-select, connect and auto-arrange. Upload local media and see how every card relates at a glance.

Multimodal generation

Run text, image, video and audio per card, with queuing, cancellation, regeneration and full version history.

Agent collaboration

The agent reads bound context, edits text and places cards. Media proposals run only after you approve them.

Asset reuse

Project resources, a personal asset library, prompt templates, media styles and mixed reference inputs — saved once, reused everywhere.

Image processing

Annotate, crop, rotate, mirror, upscale and extract depth, plus AI edits wherever the configured capability supports them.

Self-hosted administration

Model configuration, local or OSS / COS / S3 storage, task and call records, and project manifest export.

Creative flow

Five capabilities, one kind of card

What you can run depends on the capabilities your administrator publishes. The interface ships in Chinese, English, Russian and Japanese.

Text

Point it at any OpenAI-compatible endpoint and generate or rewrite per card. Keep what you like, regenerate what you do not, and revisit any earlier version.

OpenAI-compatible endpoint Custom model ID Version history
Image

Keep working on an image inside the canvas: annotate, crop, rotate, mirror, upscale and extract depth — or upload your own media directly.

GPT Image Google Nano Banana ComfyUI workflows Mixed reference inputs
Video

Start from text or an image and generate video, with reference media where you need it. Results in an unknown state must be retried explicitly, so an unfinished job is never mistaken for a finished one.

Volcano Ark Seedance RunningHub V2 Media relay
Audio

Generate speech and sound effects card by card, referencing the text and images already on the canvas, and pair them with video cards.

Seed Audio Reference audio Per-card runs
Agent

Reads bound context, edits text and places cards, and puts media proposals in front of you to approve or reject in one place. Administrators should run the tool-calling diagnostic before using agents.

Bound context Proposal approval Skill configuration Multi-turn
Integrations

Bring your own models and workflows

Adapters and workflows are published by the administrator, so what users see always matches what is actually available.

Language models

  • Any OpenAI-compatible endpoint
  • Your own model IDs and API keys
  • Tool-calling diagnostic to verify agent readiness

Media adapters

  • GPT Image and Google Nano Banana
  • Volcano Ark Seedance and Seed Audio
  • Media relay for references that need public access

Workflow engines

  • ComfyUI: import and map published API workflows
  • RunningHub V2 workflows and AI apps
  • Available operations follow what you publish

Storage and credentials

  • Local storage, or OSS / COS / S3
  • API keys encrypted at rest and kept out of logs
  • Database password and master key generated and persisted
Quick start

Three steps to your own canvas

Starting up and signing in needs no Git, no local build toolchain, no GPU, no .env and no model keys. You only need Docker Engine or Docker Desktop with Docker Compose v2.

Launch it

Pick your platform and run the whole command as one paste.

Terminal
mkdir agenvas && cd agenvas && \
  curl -fL https://raw.githubusercontent.com/grayrepo-byte/Agenvas/main/docker-compose.yml \
    -o docker-compose.yml && \
  docker compose up -d --wait

This prepares an agenvas directory, pulls the images and waits for the web, API and PostgreSQL services to become healthy. The first pull can take a few minutes. Keep the directory — you will run maintenance from it later.

Create the admin

On the same machine, open http://127.0.0.1:8088/setup — from another device, use http://<server-ip>:8088/setup. Fill in an admin login and password, then sign in.

Setup completion is recorded permanently in the database, so restarts or disabling the admin never reopen it. Finish setup on a controlled network before exposing the service to the internet.

Configure models and start creating

Under “Settings → Model configuration”, add the endpoint, model ID and API key for your OpenAI-compatible provider. Under “Media configuration”, create connections, publish capabilities and set the default models. Then create a project and add cards or upload media from the canvas context menu.

Everyday maintenance
docker compose up -d --wait   # start or update
docker compose ps             # service status
docker compose down           # stop, keep data

Back up the database, media, configuration and encryption key before updating. Do not run docker compose down -v — it deletes your data and keys.

Stack

Familiar tools, transparent implementation

Toolchain baseline: JDK 21, Node 24 LTS (24.12+) and pnpm. Exact versions live in the dependency baseline document in the repository.

Frontend

  • Vite · React · TypeScript
  • React Flow for canvas and connector interaction
  • TanStack Query · Zustand

Backend

  • Java 21 · Spring Boot
  • Spring AI for model and agent runtime
  • jOOQ for type-safe database access

Data and deployment

  • PostgreSQL · Flyway migrations
  • Docker Compose for one-command startup
  • REST / SSE for live task feedback

Local development, the mock environment, container builds from source and image tagging rules are covered in the repository README and development constraints.

Get started

Deploy the canvas on your own server

MIT licensed, with no membership tiers and no usage threshold. The repository ships the product spec, development checklist, and backup and restore docs. Issues and pull requests are welcome.

github.com/grayrepo-byte/Agenvas