Vercel templates worth using
The five starters worth reaching for, and the architecture behind the choice.
The verdict only. What exists is inventoried in the Template Library on disk: 80 researched templates screened against 482 catalog listings, with factsheets and status tracking. Nothing here repeats it.
The filter
Most AI templates are a chat box with a model behind it. The ones worth taking separate agent state from execution, treat channels as interfaces into one agent, or solve a structural problem that is painful to retrofit later.
Top five
| Template | Best at |
|---|---|
| Open Agents | Agent system architecture |
| eve Personal Agent | Cross-channel agent product |
| Platforms Starter Kit | General SaaS architecture |
| Morphic | Usable AI product foundation |
| Liveblocks Starter Kit | Collaborative product |
The idea in each
| Template | The idea |
|---|---|
| Open Agents | The agent is not the sandbox |
| Platforms Starter Kit | Tenancy as a routing primitive |
| eve Personal Agent | Channels are one agent's surfaces |
| Liveblocks | Presence as infrastructure |
| next-forge, Multi-Zones | Org structure, not product |
Open Agents keeps agent state outside the isolated runtime. That one separation buys execution past a request timeout, replaceable model and sandbox providers, and reconnection to work already running. eve deploys the interface and the agent runtime as separate services, which is what makes a fifth channel a config change rather than a rewrite.
The combination worth building
Platforms Starter Kit for tenancy and auth, Open Agents or eve for durable AI, Stripe Subscription Starter for money, Liveblocks when people share an object. Together those are an AI SaaS rather than another chatbot demo.
Caveats
- MotherDuck auth is demo-grade. Replace it before production.
- Doc Search predates the agent era. Keep the ingestion, modernize the model layer.
- Hume EVI is a voice interface, not a product.
- Multi-Zones and next-forge solve organizational problems. Without the organization they only add coordination cost.
- Catalogs move. The architecture notes are the durable part, so re-check a link before committing to a starter.
See also
- LLM Wiki for the memory and approval loop eve ships as a product
- RAG transit planning for what grounded retrieval demands
- AI-Augmented Software Engineering on context as the binding constraint
Related pages
- LLM WikiBuild a second brain that maintains itself — an AI reads your sources and keeps a living, cross-referenced wiki of what you know.
- RAG Transit Planning ReportAI-synthesized walkthrough of IFP's proposal to use RAG over FTA planning/construction documents as agency institutional memory.
- AWS Bedrock AI StrategyConsolidated overview of Bedrock capabilities, use cases, architecture patterns, governance, and a sequenced project portfolio for a wealth management firm.
- AI-Augmented Software Engineering 2025The essential numbers, shifts, tools, and risks reshaping how software gets built — from agentic IDEs to the junior squeeze.
- Estimating an LLM billThe token arithmetic every platform bills on — where the money actually goes, and which levers move it.