Skip to content

Latest commit

 

History

165 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

ParamRig

ParamRig is an open-source workbench for human-tuning AI-built visual systems.

Instead of asking AI to own the finished creative result, ParamRig asks it to build the tools. The AI creates a purpose-built rig around a component, design system, SVG, animation, 3D scene, or procedural world. A human shapes the result in a live preview, then the AI applies the validated choices back to the product.

ParamRig also supports rapid proofs of concept: uncertain ideas become manipulable prototypes whose approved parameters can be carried into production.

Let AI build the tools. Keep creation human.

Official website: paramrig.com — planned, not yet launched.

Status

A local frontend is available for the library and SVG, HTML and 3D editors, with editable animation tracks, snapshots and action history. Discovery notes still apply to product strategy; they are not the application itself.

Run locally

The workbench runs as the Docker Compose project paramrig. Do not start Vite on the host.

docker compose up -d

Then open http://localhost:5174/.

See docs/local-docker-development.md for stop/start, rebuilds, and one-off commands.

  • Library: /
  • Example rig: /r/contour-bloom, /r/tidal-planet
  • Docs: /docs
  • Controllers: /docs/controls
  • Full controller lab: /r/controller-lab
  • QA fixtures: /?fixture=empty, /?fixture=error, /?fixture=loading (holds the skeleton), /?fixture=long
docker compose run --rm app npm test
docker compose run --rm app npm run lint
docker compose run --rm app npm run build

Example rigs come from src/rigs/registry.ts. See docs/ADDING-A-RIG.md to add another study without editing the workspace shell.

QA captures from the implementation pass live in workproduct/ui-implementation/qa/.

Workspace UI

Shared controls live in src/ui/: Button, NumberField, SliderField, ColorField, CurveField, GradientField, SelectField, SwitchField, Tooltip. The workspace shell (Inspector, Timeline, ExportAction, RigPreview) binds them through RigSession — do not branch on a rig name in the chrome.

ParameterField is the shared manifest-to-control renderer. The searchable catalog now contains 62 focused examples in 10 families: numbers, spatial controls, appearance, choices, typography, curves, resources, actions, collections and value sources. The same definitions run in the full Controller lab with the inspector, history, snapshots, export and timeline. See docs/CONTROLLERS.md for contracts and limits.

The timeline supports grouped tracks, zoom, keyframe selection and dragging, copy/paste/delete, precise time/value/easing controls, and a playback range. Undo/redo records a complete drag as one action. The History tab keeps the last 100 actions for the current session. Snapshots can be named, restored (including the active reference), and removed; these actions are undoable too. Reference and Current sample animated values at the same playhead position.

Right-click a control to reset it or add numeric animation. Shift+F10 opens the same actions from the keyboard; small screens expose an actions button. Timeline keys, track labels, track backgrounds, group headers and the ruler also have context menus. Right-clicking an already selected key keeps the multi-selection. Drag the grip above the timeline to resize it, or use Expand timeline. The grip supports arrow keys, End to expand, Home to collapse and Escape to cancel a drag. Its height is saved and bounded by the viewport. Drafts and snapshots use browser storage. History, selection, track visibility and the keyframe clipboard are session state, not permanent project data.

The key/track context panel uses the parameter's numeric instrument (including knobs, logarithmic sliders and discrete stops). Curve editor switches to an editable graph of the selected track, with real keys and their interpolation.

Known limits of this build: bundled examples only (no disk watch), export does not write back to source, 3D is a bounded study, and there is no auth, AI chat, or telemetry. The current UI includes intentional departures from assets/ui-mockups/; those mockups are not a pixel-perfect acceptance target.

Connected web projects

The local Web workspace opens a running project through a development SDK. It supports exposed controls, DOM selection, visual markup, snapshots, approved feedback batches and agent responses saved in the connected project. Start the optional Compose web profile and open /web. See docs/web-workspace.md for setup, the React example, the SDK build, source revisions and the file handoff contract.

Initial audience

The working hypothesis is creative developers, vibecoders, and technical artists who use coding agents to build complex visual systems and lose time translating perceptual feedback into precise code changes. This hypothesis has not yet been validated as the final target segment.

Language policy

English is the canonical language for the application, APIs, rig definitions, documentation, and error messages. Future translations must derive from the English source rather than introducing parallel product terminology.

Product framing

Internal product-discovery sessions are kept locally and are not published. The local UI uses Vite, React, and TypeScript. See package.json.

Brand

The approved Coform logo and the current branding study are available in assets/brand/. The application self-hosts Public Sans (WOFF2) from public/fonts/. The custom wordmark is vector geometry, not a font.

Contributions

Repository conventions, including the prohibition on AI signatures in commit messages, are documented in AGENTS.md.

Licensing status

No project license has been selected yet; public repository access alone does not grant an open-source license. Runtime dependencies and their licenses are listed in package.json. Bundled third-party fonts retain their accompanying license notices.

About

Let AI build the tools. Keep creation human. A workbench for fine-tuning visual systems.

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages