Rabbitholes Review: capable workspace, costly friction (67/100)

Rabbitholes is a canvas-first multi-model AI workspace for research and long-context knowledge work.

Fact	Value
Domain	rabbitholes.ai
Category	Canvas-first multi-model AI workspace
Pricing	$89/year and $249 lifetime options reported in the supplied evidence; confirmation unavailable
Pages crawled	40
Crawl date	2026-10-08

Rabbitholes presents a clear product proposition: a canvas-first, multi-model workspace for research and other long-context work. The supplied evidence gives it strong scores for audience and messaging (86/100), first impressions and positioning (84/100), accessibility (82/100), writing quality (76/100), decision-support surfaces (72/100), review-content integrity (68/100), and docs and self-serve help (88/100).

The score is held back by execution. Performance is the sharpest weakness at 35/100: mobile largest content appeared after about 6.4–7.2 seconds, versus a 2.5-second target, while desktop LCP was about 1.4 seconds. Design execution and technical SEO each scored 52/100.

Rabbitholes is legible as a canvas-first, multi-model workspace for research and long-context work within five seconds. The hero claim, “Branch any AI chat. See every idea on one canvas,” is specific and supported by copy about forking answers, wiring sources, and comparing models. Researchers, learners, strategists, writers, polymaths, and business owners are named, while Product Hunt evidence includes a 4.9/5 score and 2,500+ users. The material weakness is brand continuity: the title says Slashspace, the FAQ calls RabbitHoles its previous name, and the canonical points to slashspace.ai while the crawl domain remains rabbitholes.ai. Make the migration story consistent across title, domain, social profiles, and support surfaces, then place the strongest proof beside the hero claim.

A researcher, student, polymath, or business owner can identify the product as an AI workspace for branching, source-grounded work without decoding insider language. The site explains the canvas model, the difference from linear chat, pricing, trial terms, platform support, privacy, and how to start. The FAQ usefully distinguishes hosted models, BYOK, subscriptions, and local Ollama models; pricing states $89 yearly or $249 lifetime, with 500 credits and two device seats. The remaining friction is breadth: several audiences and model paths are named at once, so understanding the product does not always make the next choice obvious. Keep the audience-led entry points, add segment-specific CTAs, and place the local-first privacy explanation near the first trial or download action.

The core conversion paths are easy to find: the homepage reaches Pricing in one click, exposes Docs and Contact, and the support page provides support paths. Pricing states $89/year after a 3-day trial and offers a $249 lifetime option, while the docs provide Quickstart, canvas, node, AI-feature, settings, billing, privacy, and troubleshooting paths. The first interaction is still vulnerable on mobile: lab data reports 7.2 seconds LCP and 1,790 ms blocking time. Trial, lifetime, and download actions also compete for emphasis while Rabbitholes and Slashspace coexist. Reduce hero and JavaScript cost first, rank one recommended CTA, and add task-oriented shortcuts to the docs landing page.

The sampled pages have a solid structural baseline: most expose header, navigation, main, and footer landmarks; the declared language is en; sampled images were checked for alt treatment; and no sampled form controls lack a label or ARIA name. PSI reports accessibility scores of 96–97. CSS evidence reports 11 animation declarations and zero prefers-reduced-motion rules. Correct computed color pairs, add a first-focusable skip link, repair heading structure, and add reduced-motion support before manual keyboard and focus checks.

The homepage has a deliberate product-led hierarchy, but mobile execution is the weak point. At 390px, the layout facts report 162 small tap targets, including 36×36px header controls, 24×24px icon buttons, and 21×21px toolbar buttons; expand interactive hit areas to at least 44×44px. Several canvas boxes also sit outside the viewport, with x positions from 392 to 636 and others at −96, so the demo needs containment or mobile reflow. PSI fails color contrast on mobile and desktop. The stylesheet reports 234 colors, 136 font sizes, 48 radii, 92 spacing values, 59 shadows, and 28 font families, indicating substantial value drift. Establish semantic tokens, fix contrast, and simplify the mobile canvas composition.

Mobile performance is the clearest execution weakness: lab runs measured LCP at 6.4–7.2 seconds against Google’s 2.5-second target, while desktop reached about 1.4 seconds. The homepage made 136 requests and transferred 12,472 KiB; 33 requests were blocking and www.slashspace.ai represented 11,737 KiB across 111 requests. PSI recorded up to 1,786 ms blocking time, 4,010 ms boot-up, and 6,363 ms main-thread work. CrUX data was unavailable, so these are lab findings, not field loss. Prioritize the mobile LCP request, split route bundles, defer non-critical hydration and vendor work, lazy-load below-fold media, and remove the estimated 780 ms mobile redirect overhead.

The strongest copy gives visitors a problem, audience, and action in sequence. “Chat scroll is not built for research” is followed by an AI-native canvas for researchers, polymaths, and learners; docs then use concrete imperatives such as “Ask a question,” “Fork a tangent,” and “Bring in a source.” Pricing is similarly explicit: “3 days free, then $89/year” and a seven-day lifetime refund statement. Editorial hygiene is less consistent. Public pages alternate between Rabbitholes and Slashspace, the changelog reports seven H1 elements, and sampled blog titles reach 80–86 characters. Keep the Quickstart’s action-led specificity, state the rename once in the hero, shorten article titles to the core query, and repair the changelog heading structure.

The pricing page presents a compact two-plan comparison: Yearly at $89/year and Forever at $249 one-time, with access, updates, wallet credits, BYOK, and device seats visible. The 3-day trial and seven-day refund language are also exposed. What is missing is a defended recommendation: the table tells visitors what differs, not which plan fits which use case. Add an overridable guide—Yearly for the lowest upfront commitment, Forever for expected multi-year use—beside the tradeoff. The alternatives hub also leads with “78% less than Poppy AI Base” and “59% less than Heptabase Premium” without a supplied date, source, or methodology. Date those claims, name the exact tiers and billing basis, and link the source pricing.

The comparison writing earns credibility by admitting tradeoffs. The linear-versus-non-linear article names “More complexity,” “Higher cognitive load,” and “Onboarding required” as disadvantages while also presenting context-control and reproducibility benefits. The evidence trail is thinner: supplied articles make claims about AI memory, retrieval, and context control without a visible methodology block, source notes, test protocol, or boundary between observation and synthesis. The sampled comparison surfaces report no disclosure text; no affiliate links were detected, so this is not evidence of undisclosed activity. Preserve the balanced tables, then label claims as manufacturer specification, observed behavior, owner-experience synthesis, or editorial analysis and disclose any future commercial relationship before relevant recommendations.

The public pages render independently, the nonexistent-URL probe returns a real 404, and the supplied crawl shows no sitewide robots block or raw-versus-rendered content gap. The meaningful stability risk is migration hygiene: Rabbitholes pages return 200 while declaring Slashspace canonicals, and the root path takes two hops before the final host. Complete a one-hop redirect map and make internal links, sitemap entries, and canonicals agree on slashspace.ai. There is no measured traffic drop in this evidence: GSC and analytics are not connected. Wayback provides 23 sampled snapshots from July 2024 through May 2026, while RDAP dates registration to June 2024. Recheck GSC coverage, clicks, and impressions after the migration is consistent.

Slashspace treats documentation as part of onboarding. The public Quickstart covers installation, sign-in, plan selection, a first canvas, asking a question, forking a tangent, bringing in a source, optional BYO AI, and help; it contains 12 code blocks and links into canvas, node, and AI-feature guides. The desktop changelog records v3.3.0 on September 28, 2026, giving users a visible freshness signal. Findability is the main gap: rendered /docs facts report no search box despite a substantial guide set with 35 inbound links to Quickstart. Add keyboard-accessible documentation search and make the short changelog landing page a dated release index rather than a thin set of platform links.

The public surface is crawlable and server-rendered: the 40-page sample returned 200 responses, robots.txt allows the public surface, the nonexistent-URL probe returned a real 404, and rendered text matches raw text in the supplied samples. Sign-in and checkout are noindexed appropriately. The dominant issue is host migration. Sampled Rabbitholes pages return 200 while canonicalizing to slashspace.ai; the root path records HTTP Rabbitholes to HTTPS Rabbitholes to www Slashspace, and the sitemap contains 3,344 Slashspace URLs while the crawl captured 40 Rabbitholes URLs. Make slashspace.ai the primary host, update internal links and sitemap references, emit final self-canonicals, and redirect each legacy URL directly to its matching destination. Then clean up missing alt text, duplicate query metadata, and multiple-H1 templates.

Rabbitholes is a credible choice for people who need a canvas-first environment for multi-source, multi-model work. Its positioning is legible, its audience and workflow are well explained, and its documentation gives prospective users a useful path into the product.

The fixable weaknesses are concentrated in delivery: mobile performance is poor, design execution has contrast and undersized-control problems, and technical SEO remains uneven.

This is a 13-dimension review based on supplied crawl evidence and the public dimension score table for 40 pages crawled on 2026-10-08. The usability result uses simulated cognitive walkthroughs against crawl data, not live user testing.

Dimension 19, Editorial QA of content, was excluded because the input marks it missing. Google Search Console was not connected, so related enrichment was unavailable. The supplied evidence reports $89/year and $249 lifetime options, while confirmation is unavailable. Read the scoring method.