Overview

What aBuck is

aBuck is a visual workspace for AI generation. Instead of one chat box, you get a canvas: a prompt is a node, the image it produces is a node, the video made from that image is a node, and the lines between them are what feeds what.

Nothing generates on its own. You wire a graph, pick a model on each generator, and press Play on the node you want. Everything that node needs upstream is gathered for you.

Projects and canvases

A project is a folder. A canvas is one graph inside it. Projects can hold as many canvases as you like, and the canvas navigator in the top-left switches between them without leaving the page.

Every canvas has a URL of the form /your-project/canvas-1, so a canvas is linkable and bookmarkable.

Nodes and edges

The vertical toolbar on the left adds nodes, and the number keys 1–8 are shortcuts for the ones you reach for most. Drag from the dot on the edge of a node to another node to connect them.

Edges are typed, so not every pair connects. A text node can feed the prompt of an image generator; an image can feed another image as a reference; a video exposes its first and last frame as images. When a connection is not allowed, the target handle will not accept the drop.

Graphs have to stay acyclic — a node cannot end up feeding itself, however long the loop.

See the node reference for what each node takes in and produces.

Running a node

Press Play on a generator and it collects its inputs, calls the model, and writes the result back into itself. Every generator shows a cost estimate from your provider next to its run button before you commit to the run.

Runs are independent. Playing a node does not re-run the nodes above it, so a long chain does not get regenerated because you wanted one new variation at the end.

Versions and history

Every run is kept. A node with more than one result grows a stack of cards behind it — up to four — and the version strip lets you flip through them, compare a version against the original, and keep the one you want.

Because versions are recorded rather than overwritten, the media behind an old version stays addressable. Nothing is lost when you try a different prompt.

Undo and redo cover canvas edits, with ⌘Z and ⇧⌘Z, from the controls in the bottom-left.

Editing media on the canvas

Images and videos can be edited in place without a round trip through another app. Images support crop, relight, restyle, recolour and reframe; videos support trim and upscale. An edit becomes its own version, so the result you started from is still there.

Lists and fan-out

A list node holds many items — text, images or videos. Wire a list into the prompt input of an image or video generator and that generator fans out: it runs once per item, in parallel, and writes one result per item into a list node on its output.

Re-running a fan-out only replaces the items that run produced. Anything you uploaded into the list by hand, and any per-item edit you made, survives.

Your model access and billing

aBuck does not resell inference. You connect your own OpenRouter account under Settings → API keys, and every generation bills that account at the provider's list price. The connection is a single OAuth authorisation — there is no key to paste — and only the last four characters of the resulting key ever come back out of the server.

Settings → Models controls which models appear in the node pickers, so a canvas only offers the ones you actually want to use.

The Stocks library

Brand images and videos you upload go into your Stocks library, reachable from the images button in the bottom-left controls. The library is workspace-wide: an asset uploaded once can be dropped into any canvas in any project.

The same canvas in two places

A canvas is yours alone, but it is often open more than once — a laptop and a desktop, two tabs, or a browser tab alongside an AI client building into it. Each connection shows as a badge in the top-right, next to your avatar.

Those views reconcile rather than overwrite. A change made elsewhere arrives within a few seconds, and saving from a tab that had fallen behind merges instead of clobbering, so nodes created somewhere else are never wiped out.

Connecting an AI client

The canvas speaks MCP. Claude Code, Claude Desktop, Codex and other MCP clients can create projects, lay out nodes, wire them together and read a canvas back — they build the workflow, and you press Play. See MCP setup.

What it costs

aBuck itself is a flat subscription; the name is the price. Generation is separate and goes straight to your own provider account, which is why there is no metering, no credit balance, and no markup between you and the model.

Accounts are invite-only. Subscribing issues an invite, and each invite can be used once.