AI providers

The AI chat layer is built on the Vercel AI SDK, so the provider is a config choice, not a rewrite. You pick one at scaffold time, and you can switch later.

Selecting a provider

Two env vars drive it:


AI_PROVIDER="openai"      # which provider
AI_MODEL="gpt-4.1-mini"   # which model (provider default if omitted)
OPENAI_API_KEY="sk-..."   # the key for that provider

Only the API key for your selected provider needs a value.

Supported providers

openai, anthropic, google, xai, groq, mistral, cohere, togetherai, fireworks, deepinfra, deepseek, cerebras, perplexity, azure, bedrock, vertex, baseten, openrouter, ollama.

Each provider expects its own key (or credentials):

Provider	Key / credential
`openai`	`OPENAI_API_KEY`
`anthropic`	`ANTHROPIC_API_KEY`
`google`	`GOOGLE_GENERATIVE_AI_API_KEY`
`xai`	`XAI_API_KEY`
`groq`	`GROQ_API_KEY`
`mistral`	`MISTRAL_API_KEY`
`cohere`	`COHERE_API_KEY`
`bedrock`	`AWS_ACCESS_KEY_ID` (+ AWS creds)
`vertex`	`GOOGLE_VERTEX_PROJECT` (+ GCP creds)
`openrouter`	`OPENROUTER_API_KEY` — also unlocks any upstream model via `AI_MODEL`
`ollama`	`OLLAMA_BASE_URL` (local, no key)

(The remaining providers follow the same <PROVIDER>_API_KEY pattern — see .env.example in your project for the exact names.)

Switching providers after scaffold

The scaffolder only installs the SDK for the provider you chose. To switch to a different one:

Install the matching package, e.g. npm i @ai-sdk/anthropic.
Change AI_PROVIDER and AI_MODEL in .env.local.
Set that provider’s API key.
Restart the dev server.

If you stay within the provider you picked, just change AI_MODEL — no install needed.

Local models with Ollama

ollama needs no API key. Point it at your local server:


AI_PROVIDER="ollama"
AI_MODEL="llama3.1"
OLLAMA_BASE_URL="http://localhost:11434/v1"

Deploying on Vercel

If you deploy to Vercel, you can route models through the Vercel AI Gateway for observability and failover instead of wiring a provider key directly — see Deployment → Vercel.