Skip to main content
Available for self-hosted Business plan and above
Path: Admin Panel → AI Settings The AI Settings page is used to configure AI Chat, AI fields, AI automation, and App Builder. The Pending configuration panel on the right shows which required items are still missing. Complete them in order.

​
Before you start

Before you open the AI Settings page, make sure the following items are ready. Both AI Chat Runtime and App Builder rely on Vercel.
You only need one Vercel account to complete the setup.In the current version, some AI and App Builder capabilities are affected if the instance cannot access Vercel. Teable is working on full private deployment support for AI and App Builder, so future versions no longer depend on Vercel services.
RequirementRequired?Notes
Vercel accountRequiredYou need a working Vercel account
Vercel API TokenRequiredUsed for AI Chat Runtime
Team / ProjectRequiredYou need a usable Team and Project. If needed, you can create them in the page
AI Gateway API keyWhen using AI GatewayOnly needed if you choose AI Gateway as your model provider
OpenAI Realtime API KeyOptionalOnly needed if you want to enable Voice input in AI Chat
Public accessRequiredYour Teable instance and object storage (MinIO / S3) must be publicly accessible
Vercel-related costs usually have two parts:
ItemNotes
Vercel base costAbout $20/month for App deployment and AI runtime infrastructure
AI Gateway model usageCharged separately based on actual usage
Make sure your Vercel account has a credit card attached, that Credit status is healthy, and that auto top-up is enabled.

​
Setup steps

Follow the Pending configuration panel in order. Turn the required items green first, then test the features. If a button is disabled or a feature still does not work, it usually means an earlier requirement has not been completed yet. Follow these four steps:
1

Configure LLM API

Set up your model connection and complete the initial test.
2

Configure Recommended Models

Add a clear set of primary models for users to choose from.
3

Set chat model

Choose the default model for AI Chat.
4

Configure the runtime and App Builder

Complete the Vercel setup and confirm that public access is ready.

​
Configure LLM API

Start by choosing how you want to connect your models in LLM API. This step decides which models users can choose from. Start with a clear primary set so users can choose by task complexity. A common setup is:
ModelBest for
OpusComplex tasks and high-quality output
SonnetEveryday tasks with a good balance of quality and cost
HaikuLightweight tasks and fast responses

​
Set chat model

This step sets the default model for sidebar AI Chat. The model must come from the recommended model list, so if this step is disabled, the previous step is usually incomplete. You should choose a stable model that supports tool calling and has already been tested. Otherwise, Agent capability will be noticeably limited.

​
Optional. Configure Voice input

Voice input lets users enter messages in AI Chat with a microphone. You can adjust Max session duration (seconds) and Session creates per minute per user. After voice input is enabled and saved, users will see the microphone button in the AI Chat input box.

​
Configure the Runtime and App Builder

Complete AI Chat Runtime first, then complete App Builder.
1

Finish AI Chat Runtime

In AI Chat Runtime, enter and verify the Vercel API Token, select Team and Project, click Save, then click Test Connection. After the connection test succeeds, the matching item in the Pending configuration panel will turn green.
2

Finish the basic App Builder setup

In App Builder, enter the Vercel access token under Vercel Deploy. This is the core setting required to publish apps.
If you also need a custom domain, or if your server cannot access the Vercel API directly, you can add these optional settings:
Optional itemWhen you need it
Custom domainWhen you want to publish apps under your own domain
App AuthWhen generated apps need Email OTP or Google sign-in
Vercel API proxy URLWhen your server cannot reach the Vercel API directly, or when you use a proxy or gateway
In App Auth, you can configure sign-in providers for generated apps:
  • Email OTP uses an independent SMTP mailbox for app login verification emails.
  • Google OAuth requires a Client ID and Client Secret. Add the redirect URI shown on this page to the authorized redirect URI list in Google Cloud Console.
Teable sign-in is available by default. Email OTP appears in App Builder only after the SMTP settings are complete, and Google appears only after both OAuth credentials are configured.
Click Test Public Access in the page and confirm that both your Teable instance and object storage (MinIO / S3) are publicly accessible.
How to verify the setup is complete After the setup is done, first confirm that all required items in the Pending configuration panel are green. Then test AI Chat, AI fields, AI automation, and App Builder one by one.

​
FAQ

Go back and finish Configure recommended models first. The chat model can only be selected from the recommended model list.
Check the Pending configuration panel first to see which item is still missing. In many cases, the current step is saved correctly, but an earlier requirement is still incomplete.
This usually means AI Chat Runtime is not fully configured yet. Make sure all three items below are completed:
  • Vercel API Token
  • Team
  • Project
Entering only the token is not enough. After choosing Team and Project, save the selection before testing the connection.
Check your Vercel account Credit or balance first. Top up if needed, and make sure auto top-up is enabled.
If you use Vercel AI Gateway, request data will still be sent to the model provider you selected for processing, so third-party providers may be involved.If you have data security or compliance requirements, enable this option in Vercel AI Gateway:
  • Only allow providers with Zero Data Retention (ZDR)
After you enable it, available models will be limited to providers that support Zero Data Retention.
Even with ZDR enabled, requests are still sent to the selected model provider for processing. The difference is that usage is limited to providers that support zero data retention.
Last modified on June 4, 2026