AnyPixel Next / Open-source alpha

Your agent.
Your design
practice.

Give your coding agent the context to make better design decisions—and the tools to check its work. Edit the guidance. Keep the evidence. Carry decisions into the next task.

CLI + MCP · Node 24+ · Apache-2.0

Works with your existing agent Guidance you can edit Evidence in your project Open to contributions

The workflow

From a screen to a decision worth keeping.

Your agent does the reasoning and edits. AnyPixel brings the project context, captures the evidence, and helps check what changed.

01 / CONTEXT

Start with your project

Load the brief, design guidance, relevant examples, and accepted decisions for the task at hand.

02 / OBSERVE

Look at the real thing

Import a screenshot or reference. Add the browser extension to capture a live page, DOM facts, and axe results.

03 / REVIEW

Check the change

Read the evidence, run checks, and compare results in an HTML report. Keep measured findings separate from design judgments.

04 / RECORD

Remember the choice

Save an accepted decision with its reason and scope. Bring it into the next relevant task instead of starting over.

Five shared operations, through CLI or MCP: context / observe / review / record / read

Make it yours

A practice you can read.
And rewrite.

Your team's way of working belongs in files you control. Start with the included guidance, change it, and contribute what you learn.

your-project/
└── design/
    ├── brief.md       # the job to do
    ├── system.md      # your patterns
    ├── project.json   # packs + tools
    └── decisions/    # choices + reasons
Read the starter practice

Teach it with a pack

Add Markdown guidance and examples for your product, brand, or domain. A useful contribution doesn't need TypeScript.

Add a check or an observer

Use the extension API to collect new kinds of evidence or check your team's conventions. Keep those additions outside the core engine.

Build in the open

The new runtime and original packs are Apache-2.0. Read the source, fork it, share a pack, or help improve a fixture.

Build an extension

Start here

Try one useful loop.

The local demo captures the invitation form, finds two accessibility issues, applies a scripted fix, and checks again. Then it carries a decision into the next task.

It prints the before-and-after reports so you can inspect the evidence yourself.

Set it up in your own project
TERMINAL / NODE 24+
git clone https://github.com/justinrhowell/anypixel-next.git
cd anypixel-next
npm ci
npx playwright install chromium
npm run demo

Already have Chrome? Skip the browser download and use BROWSER_CHANNEL=chrome npm run demo.

Early alpha, installed from source. npm publication and testing in individual agent hosts are still ahead.

A few details

Small engine.
Room to grow.

How does it fit into my existing agent?

Use the CLI or register the MCP server, then give your agent the starter skill. The same five operations are available through both. Your agent keeps responsibility for reasoning, edits, and permissions. The MCP SDK integration is tested; setup in specific agent hosts still needs validation.

What does “remember” mean here?

Project decisions are scoped Markdown files with a reason and an acceptance record. Relevant accepted decisions appear in later context. You can read and edit them. This is explicit project knowledge, not automatic model training.

Does everything need a browser?

No. Guidance, decisions, and importing PNG, JPEG, Markdown, text, or JSON work without browser tools. Live capture and accessibility checks use the optional Playwright and axe extension.

Where does the project data go?

AnyPixel stores guidance, decisions, and evidence locally. The runtime makes no model-provider calls and has no telemetry. Your host agent has its own data handling, and live pages can make normal network requests. Only install native extensions you trust.

What's ready, and what's next?

The CLI, MCP server, packs, decisions, evidence reports, file import, and optional browser extension work today. Next come real agent-host trials, authenticated capture, and independent design tasks. See the current status and roadmap.

Bring your own point of view.

A better example. A sharper check. A practice that helps your team. Start small and help shape what comes next.

Start contributing