better-auth

name

better-auth

description

Authentication and authorization with better-auth in Spiceflow and TypeScript apps. Covers server config with Drizzle adapter (Postgres and SQLite), Spiceflow middleware for forwarding auth requests, client setup, session middleware, social and email/password auth, server-side session checks, and React client hooks (useSession, signIn, signOut, signUp). Also covers device authorization (CLI device flow), bearer token auth, and server actions with auth. ALWAYS load this skill when a project uses better-auth.

better-auth

better-auth is the most comprehensive authentication framework for TypeScript. It provides email/password, social OAuth, session management, 2FA, and more out of the box. It works with any backend that uses standard Request/Response objects.

Full docs: https://better-auth.com/llms.txt

When you need docs for a better-auth feature not covered in this skill (specific plugin API, config options, edge cases), use WebFetch to fetch https://better-auth.com/llms.txt. It contains the full better-auth documentation in a single file optimized for LLMs.

URL construction

Always use new URL(path, base) instead of string concatenation or template literals for building URLs:

// GOOD
const url = new URL('/api/auth', process.env.BETTER_AUTH_URL)

// BAD
const url = `${process.env.BETTER_AUTH_URL}/api/auth`
const url = process.env.BETTER_AUTH_URL + '/api/auth'

new URL handles trailing slashes, normalizes paths, and avoids double-slash bugs.

Installation

Recommended: better-auth-drizzle-adapter (works with drizzle v0 and v1)

Always use better-auth-drizzle-adapter (npm) instead of the official @better-auth/drizzle-adapter. Three reasons:

1. drizzle-orm v1 support. The official @better-auth/drizzle-adapter only works with drizzle-orm v0 (^0.45). It crashes on drizzle-orm v1 (1.0.0-beta) with "model 'user' was not found in the schema object". The community adapter is vendored from PR #9489 which adds relations-v2 support.

2. SQL null bug fixed. Both the official adapter and the upstream PR code use eq(column, null) which generates column = NULL in SQL. This is never true (SQL null semantics). It silently breaks device authorization, refresh-token rotation, and any operation using { value: null } WHERE clauses. better-auth-drizzle-adapter@>=1.0.3 fixes this with isNull()/isNotNull().

3. postgres-js deleteMany fix. The official adapter's deleteMany returns 0 on postgres-js because Result extends Array and res.length is 0 for DELETE without RETURNING. Fixed in >=1.0.4.

pnpm add better-auth better-auth-drizzle-adapter

import { betterAuth } from 'better-auth/minimal'
import { drizzleAdapter } from 'better-auth-drizzle-adapter'

export const auth = betterAuth({
  database: drizzleAdapter(db, { provider: 'sqlite' }), // or 'pg'
  // ...
})

Use better-auth/minimal on Cloudflare Workers to avoid bundling Kysely (~400KB). The /minimal entry strips the built-in database layer, so a drizzle adapter is required.

Source: https://github.com/remorses/better-auth-drizzle-adapter

Server config

Create src/lib/auth.ts (or lib/auth.ts). Export the auth instance as auth.

Drizzle adapter (Postgres)

// src/lib/auth.ts
import { betterAuth } from 'better-auth'
import { drizzleAdapter } from 'better-auth-drizzle-adapter'
import { db } from 'db' // drizzle instance from your db workspace package

export const auth = betterAuth({
  database: drizzleAdapter(db, {
    provider: 'pg',
  }),
  secret: process.env.BETTER_AUTH_SECRET!,
  baseURL: process.env.BETTER_AUTH_URL!,
  emailAndPassword: {
    enabled: true,
  },
  socialProviders: {
    google: {
      clientId: process.env.GOOGLE_CLIENT_ID!,
      clientSecret: process.env.GOOGLE_CLIENT_SECRET!,
    },
  },
  session: {
    expiresIn: 60 * 60 * 24 * 365, // 1 year
    updateAge: 60 * 60 * 24, // refresh expiry every 1 day of activity
    cookieCache: {
      enabled: true,
      maxAge: 5 * 60, // 5 minutes
    },
  },
})

Drizzle adapter (SQLite / Cloudflare D1)

import { betterAuth } from 'better-auth/minimal'
import { drizzleAdapter } from 'better-auth-drizzle-adapter'
import { db } from 'db'

export const auth = betterAuth({
  database: drizzleAdapter(db, {
    provider: 'sqlite',
  }),
  secret: process.env.BETTER_AUTH_SECRET!,
  baseURL: process.env.BETTER_AUTH_URL!,
  emailAndPassword: { enabled: true },
  session: {
    expiresIn: 60 * 60 * 24 * 365, // 1 year
    updateAge: 60 * 60 * 24, // refresh expiry every 1 day of activity
    cookieCache: {
      enabled: true,
      maxAge: 5 * 60, // 5 minutes
    },
  },
})

Environment variables

BETTER_AUTH_URL is always a secret, never a plain env var or hardcoded value. It differs per environment: dev uses http://localhost:3000, preview uses the preview deploy URL, production uses the real domain. Treat it the same as BETTER_AUTH_SECRET.

BETTER_AUTH_URL must match the Origin header the browser sends. BetterAuth validates the Origin header on every /api/auth/* request against the configured baseURL. If they don't match, you get 403 {"message":"Invalid origin","code":"INVALID_ORIGIN"}. This commonly happens when secrets management tools (Sigillo, Doppler) inject the production URL during local dev. Set it correctly in sigillo or doppler with BETTER_AUTH_URL secret. Or use a wrangler.json vars variable. Also make sure sigillo is configured to the dev env locally, check with sigillo me

For Cloudflare Workers, put both in secrets.required in wrangler.jsonc:

{
  "secrets": {
    "required": [
      "BETTER_AUTH_SECRET",
      "BETTER_AUTH_URL",
      "GOOGLE_CLIENT_ID",
      "GOOGLE_CLIENT_SECRET"
    ]
  }
}

For Doppler/Sigillo, set per-environment values:

Variable	development	preview	production
BETTER_AUTH_URL	http://localhost:3000	https://preview.example.com	https://example.com
BETTER_AUTH_SECRET	(random 32+ chars)	(random 32+ chars)	(random 32+ chars)

BETTER_AUTH_SECRET=  # min 32 chars, generate with: openssl rand -base64 32
BETTER_AUTH_URL=     # MUST be set per env — never hardcode
GOOGLE_CLIENT_ID=
GOOGLE_CLIENT_SECRET=

Schema generation

better-auth manages its own tables (user, session, account, verification). Generate the Drizzle schema for them:

pnpm dlx auth@latest generate

This outputs a Drizzle schema file. Add the generated tables to your src/schema.ts and run drizzle-kit generate + drizzle-kit migrate as usual.

When you add plugins that require new tables (2FA, organization, etc.), re-run pnpm dlx auth@latest generate to update the schema.

Spiceflow integration

Auth route middleware

In Spiceflow, mount better-auth using a .use() middleware that forwards requests with the /api/auth prefix to auth.handler(). If auth returns a 404 (no matching auth endpoint), fall through to your own routes instead of returning the 404:

import { Spiceflow } from 'spiceflow'
import { auth } from './lib/auth'

export const app = new Spiceflow()
  .use(async ({ request }, next) => {
    if (request.parsedUrl.pathname.startsWith('/api/auth')) {
      const response = await auth.handler(request)
      // Return auth responses (200, 401, 403, etc.) directly.
      // Only fall through on 404 (no matching auth endpoint).
      if (response.ok || response.status !== 404) return response
    }
    return next()
  })
  // ... rest of your routes

Use res.ok || res.status !== 404 instead of just res.status === 404 so auth error responses (401, 403, 400) are returned directly instead of falling through to your app routes:

.use(async ({ request }, next) => {
  if (request.parsedUrl.pathname.startsWith('/api/auth')) {
    const response = await auth.handler(request)
    if (response.ok || response.status !== 404) return response
  }
  return next()
})

This handles ALL better-auth endpoints (sign-in, sign-up, OAuth callback, session, etc.). The middleware short-circuits for auth paths and returns the auth response directly. Non-auth paths and unmatched auth paths fall through to next().

Session state + loader

Use .state() to resolve the session once in middleware, then expose it to all pages and client components via a /* loader. This is the recommended pattern — it's fully type-safe and avoids prop drilling:

import { Spiceflow, redirect } from 'spiceflow'
import { auth } from './lib/auth'

// Session type — includes both session and user, with plugin-extended fields
type AuthSession = typeof auth.$Infer.Session | null

export const app = new Spiceflow()
  // 1. Auth middleware — forward /api/auth/* to better-auth
  .use(async ({ request }, next) => {
    if (request.parsedUrl.pathname.startsWith('/api/auth')) {
      const response = await auth.handler(request)
      if (response.ok || response.status !== 404) return response
    }
    return next()
  })

  // 2. Session state — resolved once per request via middleware
  .state('session', null as AuthSession)
  .use(async ({ request, state }) => {
    state.session = await auth.api.getSession({ headers: request.headers })
  })

  // 3. Session loader — exposes session to all pages and client components
  // Matched by every page/layout via wildcard. Loader data is merged,
  // so pages can add their own loaders and session is always available.
  .loader('/*', ({ state }) => {
    return { session: state.session }
  })

This runs on every request including landing pages. When no session cookie is present, getSession returns null immediately (no DB query). When a session exists and cookie caching is enabled (which it should always be), getSession reads the signed cookie and skips the database entirely. The DB is only hit once every maxAge interval (default 5 minutes) to refresh the cache.

Now every page, layout, and client component can access the session type-safely:

In server components (pages/layouts) — via loaderData:

.layout('/*', async ({ loaderData, children }) => {
  return (
    <html>
      <body>
        {loaderData.session && <nav>{loaderData.session.user.name}</nav>}
        {children}
      </body>
    </html>
  )
})

.page('/dashboard', async ({ loaderData }) => {
  if (!loaderData.session) return redirect('/login')
  return <Dashboard user={loaderData.session.user} />
})

In client components — via useLoaderData hook from spiceflow/react:

'use client'

import { useLoaderData } from 'spiceflow/react'

export function UserMenu() {
  // Type-safe when SpiceflowRegister is declared in the app entry file
  const { session } = useLoaderData('/*')

  if (!session) return <a href="/login">Sign in</a>

  return (
    <div>
      <span>{session.user.name}</span>
      <button onClick={async () => {
        await authClient.signOut()
        window.location.href = '/login'
      }}>Sign out</button>
    </div>
  )
}

The /* loader matches all pages, so session is always available in useLoaderData. When multiple loaders match (e.g. /* and /dashboard), their return values are merged into a single flat object — more specific loaders override less specific ones on key conflicts.

Protecting API routes

For API routes (not pages), use state.session directly since loaders only run for pages:

.route({
  method: 'POST',
  path: '/api/posts',
  request: z.object({
    title: z.string(),
    content: z.string(),
  }),
  async handler({ request, state }) {
    if (!state.session) {
      return new Response('Unauthorized', { status: 401 })
    }
    const body = await request.json()
    // use state.session.user.id or state.session.session.userId
    return { id: '1', authorId: state.session.user.id }
  },
})

Server actions with auth

Spiceflow server actions ('use server' functions) run in a different request context than the page render. You cannot access the page's request or state directly. Use getActionRequest() from spiceflow to get the action's request, then call requireSession() on it:

import { getActionRequest, parseFormData } from 'spiceflow'

async function deletePost(formData: FormData) {
  'use server'
  const request = getActionRequest()
  const session = await requireSession(request) // throws 401 if not signed in
  const { postId } = parseFormData(z.object({ postId: z.string() }), formData)
  await db.delete(posts).where(eq(posts.id, postId))
  throw redirect('/posts')
}

Always call requireSession(getActionRequest()) at the top of every server action that mutates data. The action request carries the user's cookies/auth headers, so getSession works the same as in route handlers.

Full Spiceflow app example

import { Spiceflow, redirect } from 'spiceflow'
import { auth } from './lib/auth'
import { z } from 'zod'

type AuthSession = typeof auth.$Infer.Session | null

export const app = new Spiceflow()
  // Auth middleware
  .use(async ({ request }, next) => {
    if (request.parsedUrl.pathname.startsWith('/api/auth')) {
      const response = await auth.handler(request)
      if (response.ok || response.status !== 404) return response
    }
    return next()
  })
  // Session state
  .state('session', null as AuthSession)
  .use(async ({ request, state }) => {
    state.session = await auth.api.getSession({ headers: request.headers })
  })
  // Session loader — available to all pages and client components
  .loader('/*', ({ state }) => {
    return { session: state.session }
  })
  // Pages
  .page('/login', async ({ loaderData }) => {
    if (loaderData.session) return redirect('/')
    const { LoginButton } = await import('./components/login-button')
    return <LoginButton />
  })
  .page('/dashboard', async ({ loaderData }) => {
    if (!loaderData.session) return redirect('/login')
    return <div>Hello, {loaderData.session.user.name}</div>
  })
  // API routes use state.session directly
  .get('/api/me', ({ state }) => {
    if (!state.session) return new Response('Unauthorized', { status: 401 })
    return state.session.user
  })

declare module 'spiceflow/react' {
  interface SpiceflowRegister { app: typeof app }
}

Client setup

React client

// src/lib/auth-client.ts
import { createAuthClient } from 'better-auth/react'

export const authClient = createAuthClient({
  // omit baseURL if client and server share the same domain
  baseURL: process.env.NEXT_PUBLIC_URL,
})

export const { signIn, signUp, signOut, useSession } = authClient

With plugins

import { createAuthClient } from 'better-auth/react'
import { twoFactorClient } from 'better-auth/client/plugins'

export const authClient = createAuthClient({
  baseURL: process.env.NEXT_PUBLIC_URL,
  plugins: [
    twoFactorClient({
      twoFactorPage: '/two-factor',
    }),
  ],
})

export const { signIn, signUp, signOut, useSession } = authClient

Vanilla client (non-React)

import { createAuthClient } from 'better-auth/client'

export const authClient = createAuthClient({
  
})

Client usage patterns

useSession — reactive session in components

This is discouraged. Prefer passing down session via spiceflow loaders or props instead.

import { useSession } from '@/lib/auth-client'

function UserProfile() {
  const { data: session, isPending, error } = useSession()

  if (isPending) return <div>Loading...</div>
  if (!session) return <div>Not signed in</div>

  return <div>Hello, {session.user.name}</div>
}

Sign in with email/password

import { signIn } from '@/lib/auth-client'

await signIn.email(
  {
    email: '[e-mail]',
    password: 'password123',
    callbackURL: '/dashboard',
    rememberMe: true,
  },
  {
    onRequest: () => setLoading(true),
    onResponse: () => setLoading(false),
    onError: (ctx) => toast.error(ctx.error.message),
  },
)

Sign in with social provider (Google)

import { signIn } from '@/lib/auth-client'

await signIn.social({
  provider: 'google',
  callbackURL: '/dashboard',
})

Sign up

import { signUp } from '@/lib/auth-client'

await signUp.email({
  email: '[e-mail]',
  password: 'password123',
  name: 'John Doe',
  image: '', // optional, base64 or URL
  callbackURL: '/dashboard',
  fetchOptions: {
    onRequest: () => setLoading(true),
    onResponse: () => setLoading(false),
    onError: (ctx) => toast.error(ctx.error.message),
  },
})

Sign out

import { signOut } from '@/lib/auth-client'

await signOut({
  fetchOptions: {
    onSuccess: () => router.push('/login'),
  },
})

Using with Spiceflow typed fetch client

When calling authenticated Spiceflow API routes from the client, use createSpiceflowFetch with credentials: 'include' so cookies are sent:

import { createSpiceflowFetch } from 'spiceflow/client'

// Type safety comes from SpiceflowRegister declared in the app entry file
const safeFetch = createSpiceflowFetch(new URL('/', process.env.NEXT_PUBLIC_URL!).href)

const me = await safeFetch('/api/me', {
  fetch: { credentials: 'include' },
})
if (me instanceof Error) {
  console.error(me.message)
  return
}
console.log(me.name, me.email) // fully typed from the route handler return type

Server-side session checks

With the /* loader pattern above, session is already available in loaderData for all pages. For standalone server code that needs a session outside of Spiceflow (scripts, cron jobs, etc.):

import { auth } from './lib/auth'

const session = await auth.api.getSession({
  headers: request.headers,
})
if (!session) {
  // handle unauthenticated
}

Session caching

Always enable cookie caching. Without it, every getSession call hits the database. With cookie caching, the session is stored in a signed cookie and getSession just verifies the signature; zero database queries on most requests. This is especially important in Spiceflow apps where the /* loader calls getSession on every single page load.

export const auth = betterAuth({
  // ...
  session: {
    expiresIn: 60 * 60 * 24 * 365, // 1 year
    updateAge: 60 * 60 * 24, // refresh expiry every 1 day of activity
    cookieCache: {
      enabled: true,
      maxAge: 5 * 60, // 5 minutes
      strategy: 'compact', // smallest size, signed, default
      // 'jwt' for JWT compatibility
      // 'jwe' for full encryption
    },
  },
})

Every betterAuth() config in this skill and in new projects must include session.cookieCache.enabled: true. Omitting it means a database round-trip per request, which adds latency and load for no reason.

To bypass the cache for sensitive operations (e.g. before a destructive action):

const session = await auth.api.getSession({
  headers: request.headers,
  query: { disableCookieCache: true },
})

Session expiration — always set to 1 year

Always set session.expiresIn to 1 year in every better-auth project. The default is only 7 days, which forces users to re-login every week. This is especially painful for CLI tools using the device flow, where re-authenticating means opening a browser and approving again.

session: {
  expiresIn: 60 * 60 * 24 * 365, // 1 year
  updateAge: 60 * 60 * 24, // refresh expiry every 1 day of activity
},

updateAge means the session expiry timestamp gets pushed forward on every day of activity. Active users effectively never expire; only truly idle sessions (no API call for a full year) will need to re-authenticate.

This applies to all session types: browser cookies, CLI device-flow bearer tokens, and any other session created by better-auth. There is no per-auth-method session config in better-auth; expiresIn is global.

If you omit expiresIn, better-auth defaults to 60 * 60 * 24 * 7 (7 days). Never rely on this default.

Plugins

better-auth has a plugin system for adding features. Common plugins:

Two-factor authentication

Server:

import { betterAuth } from 'better-auth'
import { twoFactor } from 'better-auth/plugins'

export const auth = betterAuth({
  // ...
  plugins: [twoFactor()],
})

Client:

import { createAuthClient } from 'better-auth/react'
import { twoFactorClient } from 'better-auth/client/plugins'

export const authClient = createAuthClient({
  plugins: [twoFactorClient({ twoFactorPage: '/two-factor' })],
})

After adding plugins, re-run pnpm dlx auth@latest generate to generate updated schema, then run drizzle migrations.

Device authorization (CLI device flow)

Use the deviceAuthorization plugin when your app has a CLI companion that needs to authenticate via a browser. The CLI displays a user code, opens a browser to your verification page, and polls until the user approves.

Server:

import { betterAuth } from 'better-auth'
import { deviceAuthorization, bearer } from 'better-auth/plugins'

export const auth = betterAuth({
  // ...
  plugins: [
    deviceAuthorization({ verificationUri: '/device', schema: {} }),
    bearer(), // needed so the CLI can use the session token as a Bearer header
  ],
})

IMPORTANT: pass schema: {} to deviceAuthorization(). In better-auth@1.6.9+, the plugin's Zod options schema has schema: z.custom(() => true) which is non-optional. Without passing it, the plugin throws a ZodError at init time: "expected": "nonoptional", "path": ["schema"]. The schema field is only for user-provided table overrides and the plugin merges it with its built-in schema via mergeSchema(). Passing an empty object is safe and satisfies the validator. No as any cast needed; the published types accept {}.

// Error without schema field:
// ZodError: [{ "code": "invalid_type", "expected": "nonoptional",
//   "path": ["schema"], "message": "Invalid input: ..." }]
//   at deviceAuthorization (better-auth/dist/plugins/device-authorization/index.mjs)

**Schema:** The plugin requires a `device_code` table. Generate it with `pnpm dlx auth@latest generate`. The table stores device codes, user codes, expiry, and approval status.

```ts
// import * as s from 'drizzle-orm/sqlite-core'
export const deviceCode = s.sqliteTable('device_code', {
  id: s.text('id').primaryKey().notNull().$defaultFn(() => ulid()),
  deviceCode: s.text('device_code').notNull().unique(),
  userCode: s.text('user_code').notNull().unique(),
  userId: s.text('user_id').references(() => user.id, { onDelete: 'cascade' }),
  expiresAt: epochMs('expires_at').notNull(),
  status: s.text('status', {
    enum: ['pending', 'approved', 'denied', 'expired'],
  }).notNull().default('pending'),
  lastPolledAt: epochMs('last_polled_at'),
  pollingInterval: s.integer('polling_interval', { mode: 'number' }),
  clientId: s.text('client_id'),
  scope: s.text('scope'),
})

Verification page (Spiceflow):

The device flow verification page must:

1. Check the user code is valid via auth.api.deviceVerify() with request headers
2. Require the user to be signed in (redirect to login if not)
3. Provide approve/deny actions via auth.api.deviceApprove / auth.api.deviceDeny

deviceVerify must receive headers: request.headers. Without headers, better-auth cannot claim the device code for the authenticated session. The subsequent deviceApprove or deviceDeny call will fail with "Device code has not been claimed by a verifying session". Make sure the user is signed in before calling deviceVerify, so the headers carry a valid session cookie.

Requires better-auth-drizzle-adapter >= 1.0.5 which implements consumeOne and incrementOne. Older versions silently fail with "Invalid device code" errors because the device plugin depends on these methods.

import { getActionRequest, parseFormData, Spiceflow, redirect } from 'spiceflow'
import { router } from 'spiceflow/react'
import { z } from 'zod'

const devicePageQuerySchema = z.object({
  user_code: z.string().optional(),
  status: z.enum(['approved', 'denied']).optional(),
})

const deviceUserCodeSchema = z.object({ userCode: z.string().min(1) })

export const app = new Spiceflow()
  // ... auth middleware ...
  .page({
    path: '/device',
    query: devicePageQuerySchema,
    handler: async ({ request, query }) => {
      const userCode = query.user_code ?? ''
      const status = query.status

      if (!userCode) {
        return <div>Open this page from the CLI login flow.</div>
      }

      if (status === 'approved') {
        return <div>CLI approved. You can close this page.</div>
      }
      if (status === 'denied') {
        return <div>CLI denied. You can close this page.</div>
      }

      // 1. Require sign-in BEFORE deviceVerify so headers carry session cookie
      const session = await getSession(request)
      if (!session) {
        throw redirect(router.href('/login', {
          callbackURL: `${request.parsedUrl.pathname}${request.parsedUrl.search}`,
        }))
      }

      // 2. Validate AND claim the device code for this session
      const auth = getAuth()
      const device = await auth.api.deviceVerify({
        query: { user_code: userCode },
        headers: request.headers, // REQUIRED: links device code to the authenticated session
      }).catch(() => null)

      if (!device) {
        return <div>Invalid or expired device code.</div>
      }

      // 3. Server actions for approve/deny
      async function approveDevice(formData: FormData) {
        'use server'
        const actionRequest = getActionRequest()
        await requireSession(actionRequest)
        const { userCode: code } = parseFormData(deviceUserCodeSchema, formData)
        const actionAuth = getAuth()
        await actionAuth.api.deviceApprove({
          body: { userCode: code },
          headers: actionRequest.headers,
        })
        throw redirect(router.href('/device', { user_code: code, status: 'approved' }))
      }

      async function denyDevice(formData: FormData) {
        'use server'
        const actionRequest = getActionRequest()
        await requireSession(actionRequest)
        const { userCode: code } = parseFormData(deviceUserCodeSchema, formData)
        const actionAuth = getAuth()
        await actionAuth.api.deviceDeny({
          body: { userCode: code },
          headers: actionRequest.headers,
        })
        throw redirect(router.href('/device', { user_code: code, status: 'denied' }))
      }

      return (
        <div>
          <p>A CLI is requesting access. Code: {userCode}</p>
          <form action={approveDevice}>
            <input type="hidden" name="userCode" value={userCode} />
            <button type="submit">Approve</button>
          </form>
          <form action={denyDevice}>
            <input type="hidden" name="userCode" value={userCode} />
            <button type="submit">Deny</button>
          </form>
        </div>
      )
    },
  })

CLI side (polling loop):

Always use createAuthClient with deviceAuthorizationClient plugin instead of hardcoding endpoint URLs. Endpoint paths change between better-auth versions (e.g. /device-authorization/request → /device/code). Hardcoded URLs silently break on upgrades and are too easy to get wrong.

import { createAuthClient } from 'better-auth/client'
import { deviceAuthorizationClient } from 'better-auth/client/plugins'

const client = createAuthClient({
  baseURL: 'https://myapp.com',
  plugins: [deviceAuthorizationClient()],
})

// 1. Request a device code
const { data, error } = await client.device.code({
  client_id: 'my-cli',
})
if (error || !data) throw new Error(`Failed to request device code: ${error?.message}`)

const verificationUrl = data.verification_uri_complete
  || `https://myapp.com/device?user_code=${data.user_code}`
console.log(`Open: ${verificationUrl}`)
console.log(`Code: ${data.user_code}`)

// 2. Open the browser for the user
open(verificationUrl)

// 3. Poll until approved
const pollInterval = (data.interval || 5) * 1000
const deadline = Date.now() + (data.expires_in || 300) * 1000

while (Date.now() < deadline) {
  await new Promise((r) => { setTimeout(r, pollInterval) })
  const { data: tokenData, error: pollError } = await client.device.token({
    grant_type: 'urn:ietf:params:oauth:grant-type:device_code',
    device_code: data.device_code,
    client_id: 'my-cli',
  })
  if (tokenData?.access_token) {
    // Store access_token for Bearer auth on subsequent API calls
    break
  }
  // authorization_pending and slow_down mean keep polling
  const errorCode = (pollError as { error?: string })?.error
  if (errorCode === 'authorization_pending' || errorCode === 'slow_down') continue
  if (pollError) throw new Error(`Device auth failed: ${pollError.message}`)
}

The client proxy maps method names to endpoint paths via kebab-case conversion: client.device.code() → POST /device/code, client.device.token() → POST /device/token, client.device.approve() → POST /device/approve. The pathMethods from the plugin tell the proxy which HTTP method to use.

Bearer token auth

The bearer plugin lets clients authenticate with Authorization: Bearer <session-token> instead of cookies. Essential for CLI tools, API clients, and mobile apps.

Server:

import { betterAuth } from 'better-auth'
import { bearer } from 'better-auth/plugins'

export const auth = betterAuth({
  // ...
  plugins: [bearer()],
})

No client plugin needed. The CLI or API client just sends the session token as a Bearer header:

const response = await fetch('https://myapp.com/api/me', {
  headers: { Authorization: `Bearer ${sessionToken}` },
})

auth.api.getSession({ headers }) automatically checks both cookies and the Authorization header when the bearer plugin is enabled. No code changes needed in your session resolution logic.

Other plugins

• organization — multi-tenant orgs with roles and teams
• passkey — WebAuthn/passkey authentication
• magic-link — passwordless email links
• email-otp — one-time password via email
• username — username-based auth
• admin — admin panel and user management
• bearer — Bearer token auth for APIs
• api-key — API key authentication

See https://better-auth.com/llms.txt for full plugin docs.

Error handling with onAPIError

Use onAPIError.onError to capture auth errors with your observability stack. Without this, auth errors (failed OAuth callbacks, expired sessions, DB issues) are silently logged to console and never reach your error tracker.

import { betterAuth } from 'better-auth'
import { captureException } from '@strada.sh/sdk' // or Sentry, etc.

export const auth = betterAuth({
  // ... your config ...
  onAPIError: {
    onError(error) {
      captureException(
        error instanceof Error ? error : new Error(String(error)),
        { tags: { source: 'better-auth' } },
      )
    },
  },
})

The onAPIError config is a top-level betterAuth() option (not nested under advanced). Available fields:

• onError?: (error: unknown, ctx: AuthContext) => void — called on every API error (except redirects). The error is unknown, so wrap non-Error values. When set, this replaces better-auth's default error logging, so include your own logging if needed.
• throw?: boolean — re-throw the error instead of swallowing (for frameworks that catch at a higher level)
• errorURL?: string — redirect URL for OAuth error pages (defaults to /api/auth/error)
• customizeDefaultErrorPage? — style the built-in error page

If using the strataBetterAuth() plugin from @strada.sh/sdk/better-auth, error capture is already wired up automatically. You don't need to add onAPIError manually.

Spiceflow page examples

Login page

A standalone login page that redirects to the dashboard if already authenticated. Uses loaderData.session from the /* loader — no need to call getSession again:

// In your app entry (src/main.tsx or src/app.tsx)
// Assumes auth middleware + session state + /* loader are registered (see above)

  .page('/login', async ({ loaderData }) => {
    if (loaderData.session) return redirect('/')
    const { LoginButton } = await import('./components/login-button')
    return (
      <div className="flex justify-center items-center min-h-[60vh]">
        <div className="text-center max-w-sm">
          <h1 className="text-2xl font-bold tracking-tight mb-2">My App</h1>
          <p className="text-muted-foreground mb-6">Sign in to continue</p>
          <LoginButton callbackURL="/" />
        </div>
      </div>
    )
  })

// src/components/login-button.tsx
'use client'

import { useState } from 'react'
import { authClient } from '../lib/auth-client'

export function LoginButton({ callbackURL = '/' }: { callbackURL?: string }) {
  const [loading, setLoading] = useState(false)

  return (
    <button
      onClick={async () => {
        setLoading(true)
        await authClient.signIn.social({
          provider: 'google',
          callbackURL,
        })
      }}
      disabled={loading}
      className="h-10 px-6 rounded-lg bg-primary text-primary-foreground font-semibold"
    >
      {loading ? 'Redirecting...' : 'Sign in with Google'}
    </button>
  )
}

Dashboard redirect for authenticated users

Never redirect / to a dashboard automatically. The landing page should always render for all users (authenticated or not). Instead, add a /dash or /dashboard link in your navbar that resolves the user's default destination.

The /dash route should resolve the full target path in a single query and issue one redirect directly to the final URL. Never chain redirects (e.g. /dash → /orgs/{id} → /projects/{id} → /projects/{id}/envs/dev). Each redirect is a separate worker invocation with its own DB queries, and on Cloudflare Workers with D1 the latency compounds fast; users far from the database region will see multi-second page loads.

// Resolve org → project → env in one hop, redirect to the final URL
.get('/dash', async ({ state, request }) => {
  if (!state.session) return redirect('/login?redirect=/dash')
  const db = getDb()
  const org = await db.query.orgMember.findFirst({
    where: { userId: state.session.user.id },
    with: { org: true },
    orderBy: { createdAt: 'desc' },
  })
  if (!org) return redirect('/dash/new-org')
  const project = await db.query.project.findFirst({
    where: { orgId: org.orgId },
    with: { environments: true },
    orderBy: { createdAt: 'desc' },
  })
  if (!project) return redirect(`/dash/orgs/${org.orgId}`)
  const envSlug = project.environments?.[0]?.slug ?? '_'
  return redirect(`/dash/projects/${project.id}/envs/${envSlug}`)
})

Protected layout with session

Use a layout to enforce auth for a group of pages. The session is available from loaderData (provided by the /* loader), so the layout just checks it and renders:

.layout('/app/*', async ({ loaderData, children }) => {
  if (!loaderData.session) return redirect('/login')
  const { user } = loaderData.session

  return (
    <div className="flex min-h-screen">
      <aside className="w-64 border-r p-4">
        <div className="text-sm text-muted-foreground">{user.email}</div>
        <nav>{/* sidebar links */}</nav>
        {/* Use a client component with authClient.signOut() for sign-out */}
      </aside>
      <main className="flex-1 p-6">{children}</main>
    </div>
  )
})

Protected page

Pages under a protected layout don't need to re-check auth — the layout already redirected unauthenticated users. Session data is still available via loaderData:

.page('/app/settings', async ({ loaderData }) => {
  const { user } = loaderData.session!
  return (
    <div>
      <h1 className="text-2xl font-bold">Settings</h1>
      <p>Signed in as {user.name} ({user.email})</p>
    </div>
  )
})

Protected API route

API routes don't use loaders — use state.session directly:

.route({
  method: 'POST',
  path: '/api/posts',
  request: z.object({
    title: z.string().min(1),
    content: z.string(),
  }),
  async handler({ request, state }) {
    if (!state.session) {
      return new Response(JSON.stringify({ error: 'unauthorized' }), {
        status: 401,
        headers: { 'content-type': 'application/json' },
      })
    }
    const body = await request.json()
    const post = await createPost({ ...body, authorId: state.session.user.id })
    return { ok: true, id: post.id }
  },
})

Sign out button

Do NOT use <a href="/api/auth/sign-out"> — the GET sign-out endpoint does not work reliably. Always use the client method which POSTs to the correct endpoint:

'use client'
import { createAuthClient } from 'better-auth/react'

const authClient = createAuthClient()

function SignOutButton() {
  return (
    <button onClick={async () => {
      await authClient.signOut()
      window.location.href = '/login'
    }}>
      Sign out
    </button>
  )
}

Reading session in any client component

Any client component can read the session via useLoaderData without props — it's type-safe and always available from the /* loader:

'use client'
import { useLoaderData } from 'spiceflow/react'

export function AuthGuard({ children }: { children: React.ReactNode }) {
  const { session } = useLoaderData('/*')

  if (!session) return <a href="/login">Please sign in</a>
  return <>{children}</>
}

Auto-join organization by email domain

Pattern for automatically adding users to an org when their verified email matches a domain (e.g. all @acme.com users join the Acme org). This is application-level logic built on top of better-auth sessions, not a better-auth plugin.

Schema: add a nullable autoJoinDomain column on the org table, indexed for lookup. A blocklist of common public domains (gmail.com, outlook.com, etc.) prevents misuse.

// Schema (SQLite example)
const org = s.sqliteTable('org', {
  id: s.text('id').primaryKey().$defaultFn(() => ulid()),
  name: s.text('name').notNull(),
  autoJoinDomain: s.text('auto_join_domain'), // e.g. 'acme.com'
}, (table) => [
  s.index('org_auto_join_domain_idx').on(table.autoJoinDomain),
])

Auto-join function: runs on every authenticated page load (e.g. in a Spiceflow loader). Checks emailVerified, skips public domains, queries matching orgs, inserts memberships with onConflictDoNothing. No pre-read of existing memberships needed; the unique index handles duplicates.

async function autoJoinOrgsByDomain(session: Session): Promise<void> {
  if (!session.user.emailVerified) return
  const domain = getEmailDomain(session.user.email)
  if (!domain || COMMON_EMAIL_DOMAINS.has(domain)) return

  const db = getDb()
  const matchingOrgs = await db.query.org.findMany({
    where: { autoJoinDomain: domain },
    columns: { id: true },
  })
  if (matchingOrgs.length === 0) return

  const queries = matchingOrgs.map((o) =>
    db.insert(schema.orgMember)
      .values({ orgId: o.id, userId: session.userId, role: 'member' })
      .onConflictDoNothing({ target: [schema.orgMember.orgId, schema.orgMember.userId] }),
  )
  await db.batch(queries as [any, ...any[]])
}

Where to call it: in the authenticated layout loader, before querying the user's org memberships. The function is idempotent, so calling it on every page load is safe.

.loader('/dash/*', async ({ request }) => {
  const session = await requirePageSession(request)
  await autoJoinOrgsByDomain(session) // join matching orgs before listing them
  const members = await db.query.orgMember.findMany({
    where: { userId: session.userId },
    with: { org: true },
  })
  // ...
})

UI for enabling: add a checkbox in the org creation form (hidden for public email domains like gmail.com) and a toggle in org settings. Both the create action and the update action should require emailVerified: true on the admin before allowing auto-join to be enabled.

Key rules:

• The domain blocklist and getEmailDomain() helper must live in a client-safe module (no server imports). Client components need them to show/hide the checkbox. Server code can re-export them.
• Require emailVerified: true on the admin who enables auto-join and on users who get auto-joined.
• Use onConflictDoNothing on the (org_id, user_id) unique index so the function is idempotent.
• No domain ownership verification (DNS TXT, Google Admin SDK) is needed for self-hosted tools. First user to claim a domain gets it.

Drizzle ORM v1 (beta) compatibility

The official @better-auth/drizzle-adapter does not work with drizzle-orm@beta (v1.0.0-beta). It relies on v0 APIs (db._.fullSchema, db.query) that changed in v1 and crashes with "model 'user' was not found in the schema object".

Use better-auth-drizzle-adapter instead. It's vendored from better-auth PR #9489 (relations-v2 support) with additional bug fixes for eq(null) SQL generation and postgres-js deleteMany row counts.

pnpm add better-auth better-auth-drizzle-adapter drizzle-orm@beta

import { drizzleAdapter } from 'better-auth-drizzle-adapter'

export const auth = betterAuth({
  database: drizzleAdapter(db, { provider: 'pg' }), // or 'sqlite'
  // ...
})

No subpath import needed. Works with both drizzle-orm v0 and v1. Source and bug tracker: https://github.com/remorses/better-auth-drizzle-adapter

TS2742 from exported getAuth() in app packages

If a private app exports a BetterAuth instance factory like getAuth() and tsc fails with TS2742, first check whether the app package is emitting declaration files.

The error looks like this:

The inferred type of 'getAuth' cannot be named without a reference to '../node_modules/better-auth/dist/types/auth.d.mts'. This is likely not portable. A type annotation is necessary.

This usually happens when declaration, declarationMap, emitDeclarationOnly, or composite forces TypeScript to emit .d.ts files for an app package. The inferred BetterAuth return type is large and can include transitive pnpm/pkg-pr-new internals, so TypeScript cannot print a portable public type.

For private application packages, prefer disabling declaration emit instead of writing a fake wrapper type:

{
  "compilerOptions": {
    "noEmit": true
  }
}

Then keep getAuth() exported normally:

export function getAuth() {
  const db = getDb()
  return betterAuth({
    database: drizzleAdapter(db, { provider: 'sqlite' }),
    // ...
  })
}

Only keep declaration emit for packages that are actually consumed as libraries. If a package is only a Vite/Spiceflow app, Vite emits the runtime build and tsc --noEmit is the right typecheck path.

If the package must emit declarations, add an explicit real exported BetterAuth type annotation instead of hand-writing a partial auth shape. Also try pnpm dedupe better-auth @better-auth/core better-auth-drizzle-adapter, but do not expect dedupe to fix TS2742 when declaration emit is the root cause.

Server-side API calls and cookies

When calling auth.api.* methods server-side (e.g. auth.api.signInSocial()), the response cookies (state cookies, session cookies) are not automatically sent to the browser. If you extract just the URL and create your own Response.redirect(), all Set-Cookie headers are lost.

This causes state_mismatch errors on OAuth callbacks because BetterAuth stores a signed state cookie for CSRF protection. Without it, the callback validation fails.

Always use returnHeaders: true and manually forward cookies when the result is a browser redirect:

// BAD — cookies lost, causes state_mismatch on callback
.get('/sign-in', async ({ request }) => {
  const auth = getAuth()
  const res = await auth.api.signInSocial({
    body: { provider: 'google', callbackURL: request.url },
  })
  return Response.redirect(res.url, 302) // ← bare redirect, no cookies!
})

// GOOD — returnHeaders + manual cookie forwarding
.get('/sign-in', async ({ request }) => {
  const auth = getAuth()
  // signInSocial returns JSON { url, redirect } on server calls.
  // Use returnHeaders to get both the parsed body AND Set-Cookie headers.
  const { response: result, headers } = await auth.api.signInSocial({
    body: { provider: 'google', callbackURL: request.url },
    headers: request.headers,
    returnHeaders: true,
  })
  if (!result?.url) {
    return new Response('Failed to initiate sign-in', { status: 500 })
  }
  const redirect = new Response(null, { status: 302, headers: { Location: result.url } })
  for (const cookie of headers.getSetCookie()) {
    redirect.headers.append('Set-Cookie', cookie)
  }
  return redirect
})

Do NOT use asResponse: true for signInSocial — it returns a JSON Response with { url, redirect: true } body, not a 302 redirect. The redirect is client-side behavior. On the server you must build the redirect yourself.

When to use each pattern:

• returnHeaders: true — when you need Set-Cookie headers from the response (OAuth redirects, sign-in flows that set state cookies). Returns { headers, response } where response is the parsed body and headers is a Headers object with getSetCookie().
• asResponse: true — when you want the raw Response object (rarely useful for signInSocial since it returns JSON, not a redirect).
• Default (no flag) — when you just need the data (e.g. getSession). Returns the parsed body directly. Cookies are not forwarded.

headers parameter — always pass the original request.headers when the API call needs request context (cookies, user agent, IP). Without it, BetterAuth can't read existing cookies or set new ones with the correct domain.

Server-side social sign-in redirect (protected layouts)

Never redirect to /api/auth/sign-in/social?provider=discord&callbackURL=/dashboard as a GET URL. There is no GET endpoint at that path. Better-auth's social sign-in is a POST endpoint consumed by the client SDK internally. Redirecting the browser to it returns a 404.

When a server-side layout or route needs to redirect an unauthenticated user to a social provider (e.g. Discord OAuth), call auth.api.signInSocial() programmatically and redirect to the returned OAuth URL:

// In a Spiceflow layout that guards /dashboard/*
.layout('/dashboard/*', async ({ children, request, state }) => {
  const auth = createAuth({ env: state.env, baseURL: new URL(request.url).origin })
  const session = await auth.api.getSession({ headers: request.headers })
  if (!session) {
    // Use the current path so users return to the page they tried to visit
    const callbackURL = new URL(request.url).pathname
    const { response: result, headers } = await auth.api.signInSocial({
      body: {
        provider: 'discord',
        callbackURL,
      },
      headers: request.headers,
      returnHeaders: true,
    })
    if (!result?.url) {
      throw new Response('Failed to initiate sign-in', { status: 500 })
    }
    const redirect = new Response(null, {
      status: 302,
      headers: { Location: result.url },
    })
    for (const cookie of headers.getSetCookie()) {
      redirect.headers.append('Set-Cookie', cookie)
    }
    throw redirect
  }
  return <DashboardLayout>{children}</DashboardLayout>
})

returnHeaders: true is required to capture the CSRF state cookie that better-auth sets. Without forwarding those cookies, the OAuth callback fails with state_mismatch.

SQLite/D1 date binding issue

BetterAuth passes Date objects for timestamp columns (createdAt, updatedAt, expiresAt). This crashes on Cloudflare D1 because D1's .bind() only accepts string | number | null | ArrayBuffer.

Do not use new Proxy to wrap D1. Instead, use a drizzle customType called epochMs for all timestamp columns. It stores epoch milliseconds as integers (same SQL type, no migration needed) but converts Date → date.getTime() in drizzle's toDriver hook before values reach D1. See the drizzle skill's "Timestamps" section for the full implementation.

// import * as s from 'drizzle-orm/sqlite-core'
// Use epochMs instead of integer({ mode: 'number' }) for timestamps
const user = s.sqliteTable('user', {
  createdAt: epochMs('created_at').notNull().$defaultFn(() => Date.now()),
  updatedAt: epochMs('updated_at').notNull().$defaultFn(() => Date.now()),
})

// Then pass env.DB directly to drizzle, no wrapper needed
export function getDb() {
  return drizzle(env.DB, { schema, relations: schema.relations })
}

Why not supportsDates: false? When this flag is set, BetterAuth converts Date → toISOString() (a string). If your columns are integer (storing epoch ms), this stores ISO strings in integer columns, corrupting data and breaking sorting/comparisons.

Why not integer({ mode: 'timestamp_ms' })? This changes the TypeScript type from number to Date, requiring changes across the entire codebase. API JSON responses would serialize as ISO strings instead of epoch numbers, breaking CLI clients.

This issue is tracked in https://github.com/better-auth/better-auth/issues/8882 (PR #8913 adds supportsDates: false for SQLite but converts to ISO strings, not epoch numbers, so it doesn't help for integer timestamp schemas).

Cloudflare Workers

better-auth uses AsyncLocalStorage. Enable it in wrangler.jsonc:

{
  "compatibility_flags": ["nodejs_compat"]
}

Or for just AsyncLocalStorage: ["nodejs_als"].

Bundle size: always use better-auth/minimal

For Cloudflare Workers and edge runtimes, always import from better-auth/minimal instead of better-auth. The default entrypoint bundles Kysely (~400 KB) for when no database adapter is provided. Since Cloudflare projects always use an explicit adapter (drizzle, prisma, etc.), Kysely is dead code.

import { betterAuth } from 'better-auth/minimal'

The API is identical. Plugins are imported from better-auth/plugins as usual. Saves ~400 KB from the bundle.

Social providers

Google

Always set prompt: 'select_account' so Google shows the account picker every time. Without it, users with a single Google session are silently signed in with no way to switch accounts.

socialProviders: {
  google: {
    clientId: process.env.GOOGLE_CLIENT_ID!,
    clientSecret: process.env.GOOGLE_CLIENT_SECRET!,
    prompt: 'select_account', // always show account picker
  },
}

GitHub

socialProviders: {
  github: {
    clientId: process.env.GITHUB_CLIENT_ID!,
    clientSecret: process.env.GITHUB_CLIENT_SECRET!,
  },
}

Multiple providers can be enabled simultaneously. Each provider needs its own OAuth app credentials.

Core schema

better-auth creates and manages these tables:

• user — id, name, email, emailVerified, image, createdAt, updatedAt
• session — id, token, userId, expiresAt, ipAddress, userAgent, createdAt, updatedAt
• account — id, userId, accountId, providerId, accessToken, refreshToken, expiresAt, etc.
• verification — id, identifier, value, expiresAt, createdAt, updatedAt

Generate the Drizzle schema for these with pnpm dlx auth@latest generate. Do not define these tables manually. Plugins add additional tables (e.g. twoFactor adds a twoFactor table).

Table name customization

If your Drizzle schema uses different table names (e.g. users instead of user):

export const auth = betterAuth({
  database: drizzleAdapter(db, {
    provider: 'pg',
    schema: {
      ...schema,
      user: schema.users, // map better-auth's "user" to your "users" table
    },
  }),
})

Or configure via modelName:

export const auth = betterAuth({
  database: drizzleAdapter(db, { provider: 'pg' }),
  user: { modelName: 'users' },
  session: { modelName: 'sessions' },
})

Or if all tables are plural:

drizzleAdapter(db, {
  provider: 'pg',
  usePlural: true,
})

Always typecheck before building

Always run tsc before vite build in build and deploy scripts. Vite does not typecheck; it only transpiles. Auth misconfigurations (wrong types on session, missing plugin fields, adapter type mismatches) will slip through to production silently without tsc. Your build script should be tsc && vite build.

Testing with vitest

Test better-auth apps by calling app.handle() directly with vitest. No browser, no build, sub-second feedback. This tests the real business logic: user creation, session validation, protected pages, server actions with auth, resource CRUD with ownership checks.

Much faster than browser e2e tests and covers the important things: auth flows, authorization guards, data isolation between users, redirect behavior after mutations.

Setup: set AUTH_DB=:memory: in vitest env so tests run against an in-memory SQLite database. Add a setup file that applies drizzle migrations before tests start. Enable the bearer() plugin so tests can authenticate with Authorization: Bearer <token> headers.

Pattern: create real users via auth.api.signUpEmail, get bearer tokens, pass them to createSpiceflowFetch(app, { headers }). Call server actions with runAction + authed request. Assert on page renders, loader data, and redirect responses.

Full working example: https://github.com/remorses/spiceflow/tree/main/example-better-auth

Use cases to test:

• Public pages render without auth (landing, login, marketing)
• Protected pages redirect unauthenticated users to login
• Protected API routes return 401 for unauthenticated requests
• Authenticated users see their own data (dashboard renders user name/email)
• Multi-step resource creation: signup → create org → redirect → dashboard → create project → verify render
• Mutations via server actions with auth (update profile, create/delete resources)
• Redirect behavior after mutations (action creates resource, redirects to its page)
• Security: unauthenticated users cannot access protected resources
• Security: users cannot access resources owned by other users (user B cannot see user A's org dashboard)
• Security: ownership checks on mutations (user B cannot create/delete projects in user A's org)
• Multiple users with separate sessions see isolated data
• Loader data contains correct values for the authenticated user