Design eval for agents
Agents write UI.
Rams judges it.
Kill slop in its tracks. Every change scored 0–100 against 308 design rules, fixes handed back as patches your agents apply.
One engine.
Four ways to run it.
The Skill for a free taste in your agent, the MCP for the real score before you commit, the GitHub App for review on every pull request, and the Action for the merge blocker in CI.
Every public repo Rams has reviewed.
Browse real Rams reviews across public repositories. No canned demos, just UI issues found in live codebases.
The whole review, right in your PR.
Every comment includes the issue, its severity, the affected UI area, and a concrete fix, directly inside the pull request where engineers already work.
Your team ships UI faster than it’s reviewed.
Design review does not scale with modern frontend velocity. Rams gives every UI change a consistent review before it reaches production.
Two primary-weight buttons, no visual ranking, no clear next step. The conversion click splits between actions that look identical.
Sub-4.5:1 body text, missing alt attributes, sub-44px targets, suppressed focus rings. Screen-reader and keyboard users get a half-broken page.
Inline colors that bypass tokens, padding outside the scale, !important overrides on system components. Theme changes silently miss these.
Purple-to-pink gradients, glow shadows, vague hero copy, identical filled buttons, the output of unedited prompts.
Submit buttons with no disabled state, double-fired handlers, clickable spans without keyboard support, missing loading states.
transition: all watches every property on every render. Animations without prefers-reduced-motion violate WCAG and unsettle users.
Trained on how senior designers actually review.
308 rules across 9 categories. Not linter rules. Review judgments: the bar is human-level design craft, and a whole category exists to catch AI antipatterns. New rules ship every week. Every repo Rams reviews gets sharper without changing a thing.
A design engine that keeps improving.
Eight specialist reviewers and 308 encoded judgments today, with new rules added every week. Re-reviews verify the fixes you ship, so the score climbs as the work lands. The engine finds the issue. AI explains the review.
New review judgments are authored and refined every week. As the engine encodes more of how senior designers think, every project it reviews gets sharper automatically, without you changing a thing.
Accessibility, color, type, spacing, motion, components, UX, and craft each review every PR with their own rules, the way a real design team splits the work.
Contrast ratios are calculated from real hex and token values, and every finding quotes the line it came from. Two passes: fast triage, then a deep, line-level review of the highest-risk files.
Issues are severity-rated and arrive with an inline suggestion you can commit straight from the pull request. No context-switching.
AI gives you an opinion. Rams keeps score.
Opinions are useful once. Scores become operational. Every review produces a consistent design score across PRs, repos, and releases, so teams track design quality over time instead of debating one-off opinions.
Re-reviews check every finding you fixed, by name. The score climbs as the work lands, and a clean pass says so: all flagged issues resolved.
The same 308 rules on every review, and criticals cap the score — one holds it to 59, two to 49, three or more to 39. A 60 or above always means zero critical issues.
Every pull request from every contributor gets reviewed, humans and agents alike. Nobody has to install anything.
The honest comparison: Rams vs design skills
The standards your product relies on.
Accessibility findings map to named W3C requirements, not house opinion. When Rams flags contrast, a missing label, or unguarded motion, it is pointing at the standard behind the call.
Contrast, labels, focus order, and heading structure: the quiet failures that lock real people out of your product.
Roles, names, and states on every interactive element, so assistive tech can announce each control your users reach for.
Animation that ignores motion preferences: the unguarded transitions and autoplay that leave motion-sensitive users dizzy or unwell.
Target size and focus appearance: the newest criteria, checked before most teams have adopted them.
How dialogs, menus, and tabs should behave for keyboard users, from focus trap to escape key.
“Rams brings the missing piece to AI tooling: taste.”
Cory Etzkorn · Founder at Soulmate,
previously Design Engineer at Notion
Start free. Upgrade when Rams becomes part of your merge process.
Free on public repos. Flat monthly tiers from $39, no per-seat fees. Every tier runs the full engine.
Free
For trying Rams on your open-source work. One public repo, no credit card, no commitment. Just install and go.
Solo
For your next big idea or side project. One private repository at a time, with the full review on every pull request.
Team
PopularFor teams shipping real product. Unlimited repos, unlimited people, one flat price. Go over and it keeps working.
25 reviews at $0.96 each, no subscription. Credits never expire and work on any plan. For launch weeks and backlog sweeps.
For the whole organization. Custom rules, SSO, dedicated SLA.
Catch design regressions before your users do.
Score your repo free in about two minutes, no account. Or install Rams and every pull request gets the review.
Frameworks