Handmade
GitHubDownload

Design in code.
No round trip.

A Mac app for designers. Select anything in your running React app and change it. Handmade writes the change to your code.

Download for Mac
agent-evals
evals-staging
Main
explain codebase
explain codebase

This is the evals workspace. Teams connect their agents, replay them against datasets and score every run with rubrics and checks.

  • Routes live in src/app/. Each page loads its data in a server component and passes it down.
  • Shared UI lives in src/components/, and you can browse each of them on the live /components route.
  • Scorers are plain functions in src/scorers/. Each takes a run's transcript and returns a score from 0 to 1, so they stay easy to test.
  • Traces arrive through workers/ingest/. Pages read only the rolled-up tables, never raw traces, so a slow query belongs in the rollup, not in the page.
  • Design decisions are recorded in docs/adr/, DESIGN.md and PRODUCT.md. Planned work queues in plans/README.md.

Commands

bun run dev starts the app on port 3000. bun run test runs the tests, bun run evals replays the golden runs, and bun run check must stay green.

I can go deeper on any one module. Scoring is the biggest and has the most logic.
Opus 5.5 (1M) High
agent-evals 790 × 940

Runs

Score every agent run against your test suites

Run
Pass rate
Refund flow regression
64%
9 Oct 2026
Tool-call accuracy v3
71%
2 Oct 2026
Multi-turn booking suite
78%
8 Oct 2026
SQL agent nightly
59%
6 Oct 2026
Retrieval grounding check
81%
5 Oct 2026
Support tone audit
69%
4 Oct 2026
Long-context recall
75%
10 Oct 2026
Showing 7 of 7 runs

Your production code is the canvas

  1. Select and adjust

    Change spacing, font, color, and corners on any element.

  2. Explore in canvas

    Save a version when you want to keep it.

  3. Write changes to code

    Simple style changes go straight into your React code.

  4. Hand bigger changes to agents

    Send your edits to an agent, or describe the change in words.

  5. Comment on elements

    Notes stay with the element they're pinned to.

  6. Open a pull request

    Your team reviews the change on GitHub.

Try it on your app

Download for MacStar on GitHub