Optional providers and env

Optional providers activate from environment variables and Worker bindings. With their configuration absent, the starter uses an inactive or local adapter. packages/env/src/server.ts declares the variables and production config audit; BETTER_AUTH_SECRET and BETTER_AUTH_URL remain required for deployed auth.

Configure a provider

Use root .env for the web dev server, each Worker's .dev.vars for direct Wrangler development, and the deployment shell environment for Alchemy. alchemy.run.ts forwards declared variables and provisions the required bindings. Restart development processes after changing configuration.

Verify activation

• Email: send an invitation on a deployed stack and inspect delivery history. Local log delivery does not prove Cloudflare sending works.
• Turnstile: load /sign-up with both keys set. The widget should appear; missing or invalid challenge responses must be rejected. Verification outages fail closed.
• Social sign-in: the configured provider's button appears on /sign-in. Complete a sign-in and inspect /account for the linked method.
• Exports: request an archive as an owner, wait for its ready notification, then download it. Local D1 uses emulated R2 and Queues; the Seed adapter builds archives in memory.
• Billing: the billing page offers configured provider actions to an owner or admin. See Stripe billing for setup and recovery.
• Observability: see Cloudflare observability and Web Analytics for server and browser behavior.

Assistant provider selection

packages/ai selects enabled Workers AI first, then a configured OpenAI-compatible provider. A configured provider failure does not switch providers. Persistent conversations refuse new generation when neither is configured, while browsing and management remain available.

The stateless selectAssistantLayer retains its existing explicit fallback behavior. This API Token request makes one unsaved answer:

curl -s https://<api-host>/assistant/answer \
  -H 'authorization: Bearer <api-token>' \
  -H 'content-type: application/json' \
  -d '{"workspaceSlug":"<slug>","question":"Hello"}'

A configured response reports provider: "workers-ai" or provider: "openai-compatible" and assistantConfigured: true.

Persistent model limits use ASSISTANT_INPUT_TOKENS, ASSISTANT_OUTPUT_TOKENS, ASSISTANT_PROVIDER_CONTEXT_TOKENS and ASSISTANT_PROVIDER_OUTPUT_TOKENS. Admission uses ASSISTANT_DEADLINE_MS, ASSISTANT_ACTIVE_LIMIT and ASSISTANT_RATE_LIMIT. Set provider ceilings to the selected model's actual limits. The lower application/provider bound applies, with room reserved for the answer. Invalid configuration refuses generation.

For saved history and member-authorized REST, see Private Assistant Conversations.

Add a variable

Declare it in ServerEnv and classify it in the optional secret or plain key list in packages/env/src/server.ts. Alchemy forwarding and Worker string-binding types derive from those declarations. Add any required resource binding to the infrastructure configuration.