Greptile is an AI code review agent. It reviews each pull request with context from a graph of your whole repository, learns from how your team reacts, and with TREX can run the branch in a sandbox. Design Bug Bot is 21st's GitHub App for UI design review: it compares changed React code with the components, tokens, spacing and type scale your repository already uses, and attempts to render the changed UI.
Codebase context and design context are different inputs. Greptile's graph maps how code connects; your design system also lives in visual decisions a general reviewer only learns once someone writes them down. Use Greptile for cross-file bugs, logic and written standards; add Design Bug Bot when frontend pull requests need a check against your components and mobile layout. Treat it as a second reviewer, not a Greptile alternative.
Disclosure: 21st publishes this guide and makes Design Bug Bot. It draws on Greptile's and 21st's official documentation, checked on September 14, 2026. We have not run the two tools against a shared benchmark, so this compares features and workflows, not quality.
Greptile vs Design Bug Bot at a glance
| Greptile | Design Bug Bot | |
|---|---|---|
| Built to find | Issues in each pull request, with full codebase context. Greptile docs | UI inconsistencies, layout risks and unnecessary dependencies in changed React code. Design Bug Bot |
| Context it builds | A graph of files, functions, classes, imports and calls. Graph context | The components, tokens, spacing and type scale in the repository. |
| Where your standards live | Dashboard rules or cascading .greptile/ folders (config.json, rules.md, files.json). .greptile/ configuration | Your existing UI code, which each review compares against. |
| Learning over time | Reactions, replies, comments and commits; it can infer rules. Memory and learning | Not a documented feature. |
| Running the code | TREX runs the branch in a sandbox, can click through UI with browser agents, and attaches evidence when something fails. TREX | Attempts a render in a React harness; screenshots at desktop and mobile widths when it succeeds. |
| Fixes | Suggested fixes and a Fix with your Agent button for Claude Code, OpenAI Codex, Cursor and others. Key features | Suggested fix code per finding; can push a branch for a fix pull request. |
| Code hosts | GitHub and GitLab. Quickstart | GitHub. |
| On the pull request | Summary, inline comments, optional confidence score, and a status check by default. Quickstart, config reference | One updated review comment and an advisory check per reviewed commit. |
| Cost model | Free Starter and per-seat Pro, metered in review credits. Greptile pricing | 5 free successful reviews, then AI credits. Billing details |
A "not documented" cell marks a question for your trial, not proof that a capability is missing.
Where Greptile is strong
Context beyond the diff. Greptile parses every file into a graph of files, functions, classes and variables, links calls, imports and dependencies, and uses it to find affected callers and code that departs from an established pattern. Graph context
Standards that follow your folders. The changelog has recommended the .greptile/ folder since May 2026. A rules.md file holds plain markdown scoped to its directory, config.json holds settings and structured rules with glob scopes and severity, and files.json points the reviewer at existing docs. Child folders inherit their parents' rules and context. .greptile/ configuration, config reference, changelog
Style guides as context. A style guide or CONTRIBUTING file can be supplied as context, and Greptile's guidance asks for specific, measurable rules. Custom standards
A reviewer that adapts. Reactions, replies and which comments get addressed in later commits feed its memory, and it can infer new rules from repeated feedback. Memory and learning
Runtime evidence with TREX. Announced on June 15, 2026, TREX runs the branch in a sandbox with services, dev servers, mocks or browser agents as needed, and attaches logs, screenshots, traces or video when something fails. TREX, TREX announcement
Code context and design context are different inputs
Greptile's graph can tell a reviewer that PricingCard imports Button and that three pages render it. That is structure. A design system is also a set of decisions: which Button variant is for destructive actions, the gap between stacked sections, the width a table must fit on a phone. Greptile's docs describe supplying that knowledge through rules, context files and learned feedback, so someone has to write and maintain it.
TREX adds execution, and its page frames that work around bugs that only show up at runtime. A card with a one-off 15px heading and padding that does not match its siblings renders and behaves correctly, so there may be nothing to fail.
Design Bug Bot uses the existing UI code as the reference. It checks changed UI against your components, tokens, spacing and type scale, and looks for horizontal overflow, fixed widths, hidden actions on mobile and unnecessary dependencies. A successful sandboxed render adds screenshots and visual checks at desktop and mobile widths; otherwise you get code-backed recommendations with fix code, pointed at the file and line when available.
It is also narrower: no logic, security or backend review, no visual baselines or cross-browser pixel comparison, no merge blocking, and only GitHub and React codebases (Next.js, Vite and similar, with Tailwind or plain CSS).
When to run Greptile, Design Bug Bot, or both
A hook change with ripple effects. A shared currency hook changes its return shape and three callers now format prices incorrectly. That is the cross-file case Greptile's graph is built for; Design Bug Bot does not review logic.
A testimonial card that re-implements Card. A marketing page adds a card with its own radius, padding and heading size, although an existing Card covers the need. Nothing fails at runtime. Design Bug Bot proposes the existing component and tokens in code. With Greptile, the outcome depends on whether a rule or context file covers card usage, so test that directly.
A checkout button pushed off-screen on mobile. TREX may exercise the flow with a browser agent, and Design Bug Bot looks for hidden mobile actions and attempts a mobile-width render. Compare which evidence gets the engineer to a verified fix faster.
A GitLab monorepo. Greptile supports GitLab; Design Bug Bot is GitHub only, so the pairing is not available.
Still weighing Greptile vs CodeRabbit? Both document text-based ways to teach standards, and CodeRabbit vs Design Bug Bot covers CodeRabbit's side.
Greptile pricing
Listed on Greptile pricing on September 14, 2026: Starter is free with 50 credits per month for one active developer. Pro is $30 per seat per month with 50 credits per seat, $1 per additional credit and a 14-day trial. Enterprise is custom. A standard review uses 1 credit and a TREX review uses 3.
Design Bug Bot gives new accounts 5 successful reviews free with no subscription. After that, reviews need a subscription that includes AI and use AI credits from the same balance as 21st AI. See the trial and billing details.
Test the pair on a fixture branch
- Write the design rules Greptile would need. Add a .greptile/ folder to your UI package with a rules.md naming your shared components and token rules, and a files.json pointing at your token docs. Note which rules you could not state precisely.
- Seed four pull requests. A duplicated card, a fixed-width table, a cross-file bug in a shared hook, and a clean UI change. Write down expected findings first.
- Log the evidence type. For every finding, note whether it rested on a code reference, a runtime trace or a screenshot, and whether an engineer could verify it from what was attached.
- Re-run after feedback. React to Greptile's comments for a few days, open a similar pull request, and see whether its feedback changed. Weigh that against keeping rules.md current.
Pair the trial with a design system consistency audit.
Frequently asked
- Does Greptile's full-codebase context include my design system?
- Greptile's graph covers code structure: files, functions, classes, imports and calls. Design decisions such as which Button variant to use or your spacing scale reach it through dashboard rules, .greptile/rules.md, context files such as a style guide, and learned feedback. Design Bug Bot compares changed React UI with your existing components, tokens, spacing and type scale directly.
- Does Greptile TREX take screenshots of UI changes?
- Greptile's TREX page says it runs the pull request branch in a sandbox, can use browser agents to click through a UI, and attaches logs, screenshots, traces or videos to a comment when something fails. A TREX review uses 3 credits on Greptile's pricing page. Design Bug Bot's screenshots come from rendering changed components at desktop and mobile widths as part of a design review, when the render succeeds.
- What is a good Greptile alternative for frontend design review?
- If the gap is design-system consistency on React pull requests, Design Bug Bot is built for that job. It does not review logic, security or backend code and works only on GitHub, so most teams would add it next to Greptile rather than replace Greptile.
- Does it matter whether I choose Greptile or CodeRabbit for UI consistency?
- Both document text-based ways to teach standards. Greptile uses .greptile rules, context files and learning from reactions; CodeRabbit uses path instructions, guideline files such as CLAUDE.md, and Learnings from chat. Choose between them on your wider code review needs, and evaluate design review as a separate job.