All posts

28 Oct 2026 · Hardening & Ops

Next.js 15 AI SaaS Starter Kits, Self-Hosted vs Cloud-Locked

What you actually get from a self-hosted Docker starter versus a Vercel-shaped boilerplate, and how AgenticPacks Pro compares on auth, Stripe, RAG, and ops.

Most Next.js starters assume you will stay on a cloud app platform. That is fine until the product is streaming agents, Postgres with vectors, and a bill you can explain to a cofounder. Then the lock-in is not ideological. It is duration caps and a second vendor for every feature.

This is the buying guide. The hosting math is in self-hosting architecture and cost and serverless vs VPS at 25k MAU.

AgenticPacks OSS is the free MIT baseline: Agentic-Packs/agenticpacks-oss. Pro is the paid pack at $149. Live UI: production demo.

The problem

Search "Next.js AI boilerplate" and you get three families:

  1. Cloud-locked. Ship on Vercel. Auth from a MAU vendor. Vector DB from a third vendor. Fine for a launch weekend. Painful when a graph needs three minutes and the function dies.
  2. Generic SaaS kits. Auth and Stripe, maybe a blog. No LangGraph, no pgvector, no SSE story. You still write the AI product.
  3. Self-hosted AI kits. Docker, Postgres, your own Auth.js, optional Caddy. You run it on a VPS. That is this stack.

If you want Clerk plus Pinecone plus Vercel, plenty of templates exist. This post is for the other buyer.

What to compare, not logos

QuestionCloud-locked kitAgenticPacks OSS (MIT)AgenticPacks Pro ($149)
Where does it run?Platform firstDocker Compose, laptop or VPSDocker Compose on Hetzner via Terraform
AuthOften Clerk or similarAuth.js v5, credentials + GoogleAuth.js v5, magic links, workspace RBAC
DatabaseHosted Postgres + a vector vendorPostgreSQL 16 + PrismaSame, plus pgvector RAG
AgentsRarely a real graphStreaming chat onlyLangGraph supervisor, Tavily, summarizer
BillingStripe or the platformSingle-plan Stripe + credits3-tier Stripe, portal, proration, team pools
StreamingOften untested behind a proxyWorks locallyCaddy flush_interval -1
ObservabilityVendor dashboardsLogsPrometheus + Grafana token burn
LicenseVariesMIT, clone, no checkoutOne-time GitHub access via Creem

AgenticPacks OSS has no Caddy and no Terraform. If a comparison table on the internet says otherwise, it is wrong. Caddy, Terraform, LangGraph, and pgvector are Pro. See the OSS vs Pro page.

ShipFast and friends

ShipFast-class kits are good at shipping a marketing site, auth, and Stripe fast. They are not a LangGraph plus pgvector plus a Caddyfile. If your product is an AI workspace, you will rewrite the middle of that kit. If your product is a directory with a chat widget, you may not need Pro.

The honest alternative to Pro is AgenticPacks OSS plus your own time: add Caddy, add a graph, add vectors. The cost post puts a floor on that time.

What you should try before you pay

  1. Clone AgenticPacks OSS. docker compose and sign in.
  2. Hit the Pro demo for the graph and threaded chat.
  3. Read LangGraph streaming and pgvector vs Pinecone so you know what you are buying. The layer map is in the self-hosted AI SaaS stack.

Do not buy a kit because the landing page says AI. Buy it because the repo has the Route Handler, the Compose file, and the billing table.

Tired of a starter that assumes Vercel, Clerk, and Pinecone?

Start on MIT: Agentic-Packs/agenticpacks-oss.

The self-hosted Pro pack is on the self-hosted Pro comparison.


Where to Go From Here

You do not have to assemble five vendors to launch an AI SaaS.

Option A: Free open-source baseline (MIT)


Option B: Ship with AgenticPacks Pro ($149 one-time)

If you want the self-hosted product, not a cloud-locked skeleton:

  • Automated Hetzner provisioning: Terraform, firewall, Caddy SSL, Compose.
  • Production multi-agent core: LangGraph, pgvector RAG, Redis, threaded chat.
  • Usage & Unit Economics: 3-tier Stripe, team pools, token ledger, Prometheus and Grafana.