TanStack Ship vs next-forge: Premium Next.js vs Premium TanStack Start

Written by Huifer, solo developer and maintainer of TanStack Ship. next-forge is what I'd build if I were betting on Next.js. The premium experience is real — the team behind it has shipped a polished, comprehensive starter, and the Next.js ecosystem at large has reciprocated with strong community signal. TanStack Ship is the TanStack ecosystem's answer to the same question — same ambition, different stack. This comparison is for the developer who is choosing between the two ecosystems, not between two forks of the same framework.

Verified sources: next-forge official site · TanStack Ship GitHub org Last updated: 2026-07-21 · Changelog

TL;DR: TanStack Ship and next-forge are both premium SaaS starters at the same price tier, but they target different ecosystems. next-forge is the high-water mark of the Next.js premium starter segment — polished, comprehensive, and the answer if you are betting on Next.js. TanStack Ship is the TanStack Start premium equivalent — TanStack-native modules, Cloudflare Workers + D1, and 12+ modules shaped around real production defaults. If you are choosing the framework, choose the starter that matches.

What next-forge Is, In One Paragraph

next-forge is a premium Next.js SaaS starter maintained by a team I respect. It ships authentication, billing, a marketing site, an admin UI, an email layer, a blog, AI integration, observability, and a polished design system, all wired to a Vercel-friendly deployment. The code quality is high, the documentation is thorough, the design language is consistent, and the team listens to community feedback. It is the best of the Next.js premium segment, full stop. If you are starting a SaaS on Next.js tomorrow, next-forge should be on your shortlist.

What TanStack Ship Is, In One Paragraph

TanStack Ship is the TanStack Start premium SaaS equivalent. It ships auth (Better Auth), billing (Stripe + D1), email, UTM attribution, referrals, credits, admin, blog, pricing, i18n (Paraglide), AI integration, analytics, monitoring, and a CLI scaffold — 14 modules wired together on Cloudflare Workers. It is shaped by the patterns I needed repeatedly across 12+ production client projects. The whole platform runs at the edge, costs roughly $14/month at our load, and ships TanStack Intent Agent Skills for every module.

The dependency graph tells the platform story:

{
  "dependencies": {
    "@tanstack/react-start": "^1.0.0",
    "@tanstack/react-router": "^1.50.0",
    "@tanstack/react-query": "^5.50.0",
    "@tanstack/react-form": "^0.30.0",
    "@tanstack/react-table": "^8.20.0",
    "@tanstack/react-virtual": "^3.10.0",
    "drizzle-orm": "^0.30.0",
    "better-auth": "^1.0.0"
  }
}

next-forge's parallel slice is Next-first — TanStack does not appear in the dependency graph at all:

{
  "dependencies": {
    "next": "^14.2.0",
    "@next-auth/auth-js": "^5.0.0",
    "@vercel/postgres": "^0.10.0",
    "ai": "^3.0.0",
    "@vercel/analytics": "^1.0.0",
    "inngest": "^3.20.0"
  }
}

The structural difference is not subtle: TanStack Ship's runtime is TanStack; next-forge's runtime is Next.js.

The Honest Positioning

This comparison is unusual because both codebases are "right" — for different buyers. There is no clean winner across the board. The decision tree at the end of this post makes the choice concrete; until then, the next two sections walk through the dimensions where each starter is structurally better.

I am biased because I built TanStack Ship. The next-forge team and I have not collaborated on this comparison. Where I have measured something on my own deployment, I have used real numbers from my load. Where I have estimated next-forge behavior, I have been conservative — the team has earned the benefit of the doubt on quality, even where I have not personally benchmarked their stack.

Architecture and Edge Story

TanStack Ship's edge story

TanStack Ship runs only on Cloudflare Workers. The whole architecture is shaped around that constraint and its affordances:

• Compute — Workers V8 isolates. Cold start is ~5 ms median; p99 is ~280 ms under load.
• Database — D1 (SQLite at the edge), with the Queue-backed write path for high-throughput mutations and read replicas for read-heavy routes.
• Object storage — R2 for user uploads, exports, and large assets. No egress fees.
• Async — Cloudflare Queues for producer/consumer patterns, Workflows for batched writers and durable retries.
• Stateful edge — Durable Objects for real-time and locking (used in the rate-limit and live-cursors modules).

The cost profile is honest: at the load I run on TanStack Ship (~3,200 paying customers, ~250 req/s peak, ~700 GB egress/mo), the entire Workers + D1 + R2 + Queues + Durable Objects bill is approximately $14/month. That number includes all of the platform primitives named above.

next-forge's edge story

next-forge is multi-platform. The codebase targets Vercel (the default), Netlify, and self-hosted Node. The architectural choices follow from the multi-platform target:

• Compute — Node.js on Vercel Functions (or equivalent on Netlify / Node). Cold start is a few hundred milliseconds at the typical Functions configuration.
• Database — Postgres, typically Neon or Supabase. The boilerplate ships with Postgres-native assumptions: transactions, foreign keys, serverless drivers.
• Object storage — Vercel Blob / S3, depending on deployment target.
• Async — Vercel Cron + Inngest (or equivalent), depending on what you wire.
• Stateful edge — Vercel KV / Redis / Upstash, depending on configuration.

The multi-platform support is real engineering. Maintaining two or three deployment targets is more work than maintaining one, and next-forge does the work. For teams whose corporate IT or platform policies lock them to a particular host, the multi-platform support is decisive.

Edge benchmark — directional only

I have not run a controlled A/B benchmark of TanStack Ship's Workers pipeline against next-forge's Vercel pipeline at identical workloads. The numbers I quote for TanStack Ship (cold start p99 ~280 ms, edge request budget for static pages < 50 ms TTFB from LHR) are real, from the production deployments I run. The comparable numbers for next-forge's Vercel pipeline are publicly available via Vercel's published infrastructure docs and customer reports.

Treat the comparison as directional: Workers will give you a better cold start, Vercel will give you a richer build ecosystem.

Feature Matrix

The matrix is honest about coverage parity. Both codebases ship the SaaS essentials. The differences are in the depth of specific modules and the platform-target assumptions.

Where TanStack Ship pulls ahead

• UTM attribution and referral credits as defaults. Two modules that required multiple client engagements to get right, now baked in.
• Stripe billing on a Queue-backed write path. The D1 write-lock postmortem is the production fix; the queue pattern is the default.
• TanStack Intent Agent Skills. Versioned AI context shipped with the package, not a CLAUDE.md you maintain by hand. The TanStack Intent explainer walks through the mechanism.

Where next-forge pulls ahead

• Multi-platform deployment. If Vercel is not your host, next-forge has a path. TanStack Ship does not.
• Built-in marketing site polish. next-forge's marketing templates are more refined out of the box. TanStack Ship's marketing module is intentionally minimal — it is a starting point, not a finished brand site.
• Polished admin UI. next-forge ships a more complete admin UI; TanStack Ship ships the route structure and the patterns, leaving the visual design to your project.
• Vercel AI SDK integration. next-forge's AI integration is wired more tightly to the Vercel AI SDK than TanStack Ship's. If you are using Vercel's AI primitives, next-forge is the easier path.

AI Tooling Maturity

This is the dimension I have spent the most time on, because it is the one most affected by the "premium starter" claim.

TanStack Ship's approach

TanStack Ship ships TanStack Intent Agent Skills for every module — markdown prompts, tool definitions, and example workflows that get discovered by Claude Code, Cursor, Windsurf, and Codex when the package is installed. The Skill files are versioned alongside the package, so an AI agent working on tanstack-ship@1.4.2 sees the Skill for tanstack-ship@1.4.2, not a stale global prompt. When the package upgrades, the Skills upgrade with it.

The payload shape per Skill is small — ~3-5 KB of markdown plus a tool-definition JSON. Each Skill tells the AI what the module does, what its API surface looks like, what the canonical examples are, and what to avoid. The AI gets a structured, vetted context instead of guessing from the package's exports.

next-forge's approach

next-forge ships a polished CLAUDE.md and a cursor/rules folder. The codebase is well-organized, well-typed, and well-documented — meaning an AI agent works reasonably well with it. The agent's context is whatever is in the repo: the code, the README, the CLAUDE.md the team maintains.

The trade-off: this approach is simpler and travels well across any AI tool, but it requires ongoing maintenance. When the CLAUDE.md drifts from the actual codebase — when an API changes and the doc is not updated — the AI hallucinates against the outdated doc.

What that means in practice

In my hands-on experience, versioned Skills reduce hallucinated imports and stale-API failures more than CLAUDE.md-shaped approaches do. But the controlled study has not been done. Treat the claim as directional: if AI tooling quality matters to your day-to-day, the difference is worth a small prototype on each codebase before you commit.

Decision Tree

Pick next-forge if any of the following are true:

• You are committed to the Next.js ecosystem, and TanStack would be a step sideways rather than forward for your team.
• You need a polished admin UI, marketing templates, and design system out of the box, more than you need module depth.
• You are deploying to Vercel, Netlify, or self-hosted Node — not Cloudflare Workers.
• You are building a marketing-led SaaS where the brand surface matters more than the platform.

Pick TanStack Ship if any of the following are true:

• You want TanStack-native integrations (TanStack Router patterns, TanStack Query conventions, TanStack Form) wired by default, not bolted on.
• You are deploying to Cloudflare Workers and you want D1 + R2 + Queues + Workflows + Durable Objects as your primitives — with the cost profile that implies.
• You want versioned AI context (TanStack Intent Agent Skills) shipping with the package, not a hand-maintained CLAUDE.md.
• You want modules like UTM attribution, referral credits, and the Queue-backed Stripe billing defaults — modules that emerged from the postmortem series.

The decision tree deliberately has no middle branch. Both starters are right; the right starter depends on the buyer's ecosystem commitment.

Clone TanStack Ship's premium starter: the TanStack Ship repo ships a free starter for evaluation; the Pro tier unlocks the UTM attribution, referral credits, AI integration, and the postmortem-defaulted billing and email modules. For the full pricing, see TanStack Ship pricing.

If you are broadening the comparison, the vs-shipfast comparison covers the Next.js + Vercel incumbent with a smaller community footprint, the vs-makerkit comparison covers the B2B multi-tenant segment, and the vs-supastarter comparison covers the multi-language (React + Vue) alternative. Browse the full alternatives index for every comparison page in the series.