Decision. For an auth-gated, SEO-irrelevant internal admin console, ship a standalone router on Vite — no meta-framework, no RSC. React Server Components buy you almost nothing here (no crawlers to feed, no first-paint SEO) while imposing a Node/edge server, a build-complexity jump, and a client/server mental-model split that a dashboard’s client-heavy component tree fights against.[6][12] Pick TanStack Router ⭐ 15k (Jul 2026) if you want end-to-end type-safe routes and search-params, or React Router ⭐ 56k (Jul 2026) in declarative/data mode for the familiar, lower-ceremony option. Both stay a plain static bundle you can host anywhere.[17][2] A meta-framework only earns its keep if this app later grows a public marketing surface, needs SSR/SSG for SEO, or wants to fold its backend into the same repo — none of which describe an internal console today.[1][19]
The actual question
“Which router” collapses into one architectural fork: just a router (library mode) or a whole meta-framework. Every serious option in 2026 straddles both — React Router ships three modes, and TanStack Router has TanStack Start bolted on top — so the decision is how far up that ladder you climb, not which brand you buy.[1][8]
What RSC actually buys an auth-gated admin console: ~nothing
RSC’s headline wins — zero-JS server components, streamed HTML for fast first paint, SEO-ready markup, keeping data-access on the server — are almost all crawler-and-first-paint concerns. An internal console is behind a login wall (no crawlers) and its users tolerate a brief spinner, so none of those apply.[12][13] Meanwhile the cost is real and structural:
- RSC is an architectural boundary, not an optimization. It shifts rendering to the server, changes the cost model, and forces the team to reason explicitly about which environment each component runs in.[6]
- Dashboards are the documented anti-pattern. Highly interactive, client-heavy UIs (dashboards, design tools, real-time views) push nearly the whole tree behind
"use client", which collapses the server-first advantage while keeping the coordination overhead — “harder to reason about than a traditional client-driven React app.”[6] - It requires a running server. RSC output cannot be a static file — you now operate a Node/edge runtime, and inherit hydration mismatches and a larger attack surface (2026 saw an unauthenticated RCE class in App Router server actions).[16][5]
Verdict: RSC is a poor trade for this app. The security-boundary benefit (keep secrets server-side) is real but is already solved by a normal REST/RPC backend behind the SPA.
When a meta-framework does earn its keep
| Signal your app has… | Then reach for… |
|---|---|
| Public, SEO-critical pages (marketing, docs, catalog) | Meta-framework with SSR/SSG[1][19] |
| Fast first-contentful-paint as a hard requirement | SSR / streaming (RSC)[6] |
| Want backend + frontend in one repo/deploy, server functions | TanStack Start / RR framework mode / Next[8] |
| Want the build (code-split, typegen, FS routing, SSR wiring) handled for you rather than assembling it yourself | Framework mode[1][3] |
| Auth-gated internal admin, client-heavy, no SEO | None — router + Vite SPA[12][13] |
The RR docs put it bluntly: in data mode you end up wiring builds, HMR, and code-splitting yourself — “effectively building your own meta-framework.” Framework mode exists to do that assembly for you.[1] That labor-saving is the real pitch — not RSC, and not SSR you won’t use.
SPA / client-only escape hatches (2026)
Every framework now offers a way to turn the server off — but they differ in how first-class that path is.
| Option | ⭐ Stars | Client-only story | Fit for internal admin |
|---|---|---|---|
| React Router — declarative/data mode | ⭐ 56k | Pure library, no compiler; runs on Vite/webpack/Parcel; static bundle[2] | ✓ Ideal, most familiar |
React Router — framework mode ssr:false |
⭐ 56k | First-class SPA mode: prerenders a shell, ships as static files[4] | ✓ If you want typegen + FS routing but no server |
| TanStack Router | ⭐ 15k | Purpose-built type-safe SPA router; no server at all[17][8] | ✓ Best-in-class types/search-params |
| TanStack Start — SPA mode | ⭐ 15k | Prerenders a shell; keeps server functions available if wanted[7] | △ Only if you might add server later |
Next.js App Router — output: export |
⭐ 141k | Static export + client fetching works, but its page-based routing fights the SPA client-routing model[9][11] | ✗ Fighting the tool |
| Wouter | ⭐ 7.9k | 12 KB minimalist router, pure client[15] | △ Fine but thin for a large console |
Note that RR framework mode’s RSC preview does not yet support SPA mode / prerendering — so if you went framework mode today for SPA, you’d use ssr:false, not RSC, anyway.[5]
Is “just a router + Vite SPA” still legitimate? Yes.
The ecosystem’s marketing pushes frameworks, but the tooling and the React team both still bless the SPA path. Vite is the #1 build tool by satisfaction and the React team points SPA projects at it; the pairing of Vite + a client router + TanStack Query ⭐ 50k is an explicitly recommended, production-ready 2026 stack for authenticated apps where crawlers don’t matter.[12][13] “Next.js fatigue” is a named 2026 phenomenon — hydration debugging, RSC security incidents, and complexity are driving internal-tool and dashboard teams back toward Vite SPAs; teams report dev-server startup and CI build times dropping several-fold after moving off the framework.[16] Refine’s 2026 survey still names Next.js the default for full-stack apps — the operative word being full-stack, which an internal console is not.[19]
Router pick, once you’ve chosen “no framework”
- TanStack Router ⭐ 15k — state-first (URL → state → data → UI), fully type-safe routes/params/search-params, built by the TanStack Query team. Best DX ceiling; steeper initial setup.[14][17]
- React Router ⭐ 56k — UI-first (URL → component), largest install base, now the merged home of Remix (v7). Declarative mode for simple nav, data mode when you want loaders/actions without a framework.[15][1]
- Wouter ⭐ 7.9k — only if the console is small and you want a 12 KB footprint; it overlaps but doesn’t match RR/TanStack on nested routes, FS routing, and prefetch.[15]
For an itenium-ui-style admin console on React 19: TanStack Router if the team values type safety highly, otherwise React Router (data mode) — both on Vite, both a static bundle, neither dragging in a server you don’t need.