Omg.dev scores 74/100, with its strongest evidence in accessibility, usability, and technical SEO. Its most material weaknesses are audience definition and the limited performance, documentation, and editorial evidence available from a one-page crawl.
Reviewed by SiteList Engine · 11 of 13 dimensions · published Reviewed on October 23, 2026
Is this your site?
Claim itOmg.dev is a mobile control surface for AI agents that build apps, websites, and digital work.
| Fact | Detail |
|---|---|
| Domain | omg.dev |
| Category | Mobile control surface for AI agents |
| Pricing | Free to start; no card required; no paid plan or billing cadence captured |
| Platform | iOS |
| Core capabilities | AI-assisted digital project building; online publishing |
| Crawl | 1 page on 2026-10-17 |
| Overall score | 74/100 |
Omg.dev communicates a clear mobile workflow: direct AI-agent work from an iPhone, receive a notification, and continue on a larger screen. That clarity supports strong scores for usability (86/100), accessibility (100/100), design execution (78/100), risk and stability (86.2/100), and technical SEO (88.8/100).
The weaker scores identify where the product story needs more proof. Audience and messaging scored 51.7/100, while first impressions and positioning scored 59/100; the available evidence does not name the buyer precisely or establish a defensible distinction. Performance scored 58/100 because all eight PageSpeed Insights checks returned no lab or CrUX metrics. Docs and self-serve help scored 55/100 because only 1 of 40 pages was crawled, and editorial QA scored 59.5/100 because the prose extraction was empty.
The result is a promising, understandable product surface with a stronger interaction story than its buyer definition and evidence coverage.
Omg.dev makes its phone-to-computer agent workflow understandable within five seconds, but it does not name the buyer precisely. The hero says "Build your ideas on the go," while the supporting copy explains: "Give your agents a task from your phone. Get a notification when the result is ready. Your laptop can stay closed." That mechanism is more distinctive than the headline. Proof is thin: the testimonial is attributed to "Eugene, omg.dev user" without a role, company, or measurable outcome, and the screenshot shows a GitHub count of 548 without context. Only 1 of 40 pages was crawled, so competitive positioning is not assessable. Name the intended builder segment beside the hero, lead with the agent-control mechanism, and add attributable outcome-based proof.
A stranger can understand the central action quickly: pick or write a task on the phone, let agents work on the computer, and receive a notification when it is done. The sequence "Pick a task. Put the phone down" is more informative than the broad H1, and "Download on the App Store" plus "Or open on the web" gives visitors two starting routes. The page does not identify whether it is for developers, technical founders, hobbyists, or another segment. "Free to start, no card" lowers initial friction but does not explain limits, a paid path, or cancellation. The sole testimonial lacks a role, company, or measurable outcome, and failure handling, integrations, data access, and security are not assessable from this one-page sample. Add a buyer label, a concrete task example, and a plain limits explanation.
The homepage gives first-time visitors a strong route from product explanation to activation. Desktop and mobile captures show the phone-to-computer workflow, the three-step sequence, and visible "Download on the App Store" and "Open on the web" calls to action. The mobile screenshot preserves the sequence without visible clipping, overlap, or horizontal overflow, and the lower call to action repeats the entry path after the product story. The small friction is interpretive: "Free to start" appears later than the two activation labels, so a visitor must infer which route is the free-start path. The crawl supplied no rendered HTML, click graph, destination state, or post-click feedback, so activation behavior and validation remain unverified. Keep both routes, but label them with intent, such as "Start free on the web" and "Download the iPhone app."
No accessibility failure is confirmed in the supplied evidence, but the result is provisional because the rendered DOM and accessibility tree were not captured. Desktop and 390px mobile screenshots show the headline, calls to action, product imagery, and footer content rendered without a confirmed clipped or blank state. The CSS capture reports 16 focus rules, including 15 focus-visible rules, 10 prefers-reduced-motion matches, and two visually-hidden patterns. Those are useful implementation signals, not proof that every control has a usable name, focus treatment, reading order, or contrast. All supplied PSI accessibility scores are null with no individual audits, and only 1 of 40 pages was crawled. Capture the rendered DOM and accessibility tree, then run keyboard and automated checks across the header, both CTAs, the interaction sequence, and footer links.
Omg.dev presents a coherent landing-page system across desktop and mobile: the screenshots show a clear hero-to-product-demo sequence, one dominant H1, generous section separation, and a consistent cream, black, and orange palette. The heading outline has one H1 and four H2 sections, while the mobile H1 measures 42.9px against 16px body text. The main mechanical weakness is touch ergonomics. At 390px, the layout reports 26 small targets, including 13×13px header icon links, 32×32px social controls, and compact footer links. Supporting paragraph copy is sampled at 13px. Expand interactive hit areas to at least 44×44px with 8px separation, and raise primary explanatory text to at least 14px on mobile, preferably 16px. The CSS inventory lists 56 colors, 49 font sizes, 31 radii, and 27 spacing values; computed-style review is needed before calling that visible inconsistency.
The homepage performance verdict is unmeasured because all eight PageSpeed Insights checks returned no lab or CrUX metrics. The available network summary still records 1,542 KB across 33 requests and 8 blocking requests, putting payload and scheduling first on the investigation list. Reported external groups include connect.facebook.net at 674 KB, vibes-media.t3.storage.dev at 335 KB, and www.googletagmanager.com at 151 KB, but the crawl does not identify MIME types or which requests block rendering. The below-fold phone-home-390.webp image is marked eager at 19,090 bytes while inViewport is false; it is a concrete lazy-loading candidate. Re-run mobile Lighthouse or PSI with a waterfall, identify the 8 blocking requests, set a 1.5 MB compressed homepage budget, and keep third-party JavaScript below 100 KB. Do not infer LCP, INP, CLS, or TTFB until metrics exist.
The available naming is direct, but a full copy judgment is not possible because prose extraction is empty despite an N1 count of 317 words and 27 sentences. "Build your ideas on the go" is concise, while the meta description adds useful specificity: "Build apps, websites and anything digital from your iPhone, and put them online." Preserve that plain language while adding the intended audience to the hero support copy. One confirmed hygiene issue is mechanical: 3 images lack alt text. Add purpose-based descriptions to informational images and empty alt attributes only when imagery is decorative. Re-run rendered-DOM extraction before making claims about sentence rhythm, page structure, or closing copy; the current evidence cannot support those judgments.
The sampled homepage shows no confirmed catastrophic stability pattern: it returns 200, is self-canonical, appears in the verified sitemap, has no extracted noindex directive, and the deliberately nonexistent probe returns 404. robots.txt allows crawling, and the HTTP root makes one HTTPS redirect. The evidence does not support a sitewide conclusion: only 1 of 40 crawl slots was used while the verified sitemap lists 45 URLs, GSC is unavailable, and PSI returned no lab or field metrics. The homepage also has no JSON-LD or microdata, which is an entity-signaling exposure rather than evidence of traffic loss. Renew the certificate before 2026-11-19, correct the same-target ja hreflang annotation, add accurate entity schema, then crawl representative sitemap URLs and connect GSC. Those steps will separate real property-wide issues from the current evidence gap.
This is a partial QA result for one page, not a corpus conclusion. The supplied outline is orderly: one H1, "Build your ideas on the go," followed by four H2 sections covering self-serve use, mobile context, and the iPhone CTA. However, the prose field is empty even though N1 reports 317 words, so body-level checks for sourcing, voice, endings, and answer paragraphs cannot be confirmed. The heading text "Run it yourself,for free." should be checked against the rendered DOM before treating its spacing as a typo. Separately, the crawler confirms 3 images without alt text. Re-capture rendered text, preserve the one-H1/four-H2 hierarchy, verify that heading spacing, and add meaningful alt text where images convey information.
Documentation is clearly intended as a first-class product surface: robots.txt points to the docs sitemap, and llms.txt names platform guides, local deployment, SDK material, the Sandbox API reference, and MCP setup. Quickstart flow, navigation, examples, update dates, search, and support escalation therefore cannot be judged from this sample. Include docs.omg.dev in the rendered crawl, then check the quickstart, task guides, API reference, SDK, and MCP pages against the machine-readable index.
The sampled homepage is crawlable and indexable: robots.txt explicitly allows / and declares the sitemap, the verified sitemap contains 45 URLs, the homepage returns 200 with a self-canonical, and the nonexistent probe returns 404. The HTTP root makes one 308 hop to HTTPS, and the mobile layout reports a 390px scroll width at a 390px viewport. Three operational fixes stand out. The sampled TLS certificate expires on 2026-11-19. The x-default, en, and ja hreflang entries all target https://omg.dev/; create a distinct live Japanese target or remove the unsupported alternate. The 26 small mobile targets also merit correction. This remains a homepage-level result because only 1 of 40 page budget slots was crawled and PSI supplied no metrics.
Omg.dev earns 74/100 because the core product moment is easy to understand and the sampled homepage is technically sound. The strongest case is for mobile-first builders who want to direct AI agents from an iPhone while their computer executes the work.
Three improvements would make the decision easier: name the primary buyer more precisely, clarify cost limits and alternatives, and expand the evidence for performance, documentation, and editorial quality. These are fixable communication and coverage gaps, not a reason to dismiss the underlying workflow. The site is free to start with no card required, but the crawl did not capture a paid plan or billing cadence, so commercial fit remains open.
This 13-dimension review uses the supplied crawl evidence, product facts, site profile, visible desktop and mobile screenshots, and summarized technical checks. The crawl covered 1 page on 2026-10-17. It records a live HTTPS homepage, a self-canonical, a correct 404 probe, a robots.txt entry that allows /, a verified sitemap with 45 URLs, and a documentation index referenced by robots.txt and llms.txt.
Google Search Console access was not connected, and the available PageSpeed Insights checks returned no lab or CrUX metrics. Read How SiteList scores for the review framework and evidence rules.
Omg.dev lets people build apps, websites, and other digital projects with AI from an iPhone, then put those projects online.
The site is free to start and says no card is required. No paid plan or billing cadence was captured in the crawl.
The available profile points to mobile-first builders who want to direct AI agents from an iPhone while their computer executes the work.
The one-page crawl verified a live HTTPS homepage, a self-canonical, a correct 404 probe, and clear product messaging. Performance metrics were not available from the supplied checks.