AI
AI in Tines Stories is available for all tenants, and can be configured by a tenant owner.
Learn more about AI in Tines.↗
Enable AI features
Control whether or not AI features are enabled across the whole tenant. These can be turned on via the AI settings in the settings center. By default, AI is turned on for newly created tenants.
For more granular control, expand Advanced options to enable or disable individual AI features. You can also choose whether newly released AI features are enabled by default.
AI providers
By default, AI features are powered by Anthropic's Claude, hosted securely through AWS Bedrock.
It is possible to bring your own AI provider to power all of our AI features. The following AI providers are configurable with the following customizations:
AI providers, observability tools, or custom proxies which are schema compatible with the following providers can also be configured as well. This includes but is not limited to:
OpenAI / Anthropic behind a proxy or service
Credentials and formulas in provider configuration
When configuring an OpenAI or Anthropic compatible AI provider, you can use credentials and formulas in the API key and header fields instead of pasting raw secrets.
To use credentials, associate the provider with a team. Once associated, any credential owned by that team is available to reference in the provider configuration. The provider itself remains available for use tenant-wide.
Multiple AI provider support
You can enable multiple AI providers within a single tenant.
Once enabled, Tines lets you choose models from any of your active providers when configuring:
AI Agent actions on the storyboard
The tenant's default smart and fast models in AI settings
The selected model for a Workbench conversation
This gives you the flexibility to mix and match providers and models while keeping control over exactly which models are exposed in your tenant.
Note: Tines AI features require at least one active AI provider. If a provider is being used as the default smart or fast model, it must remain enabled until you switch those defaults to another provider's models.
Team scoping
AI providers can be scoped to specific teams, giving tenant owners more granular control over who can access each provider and its models. This is useful for:
Managing costs: limit expensive models to teams that need them.
Enforcing compliance : restrict providers based on data handling requirements.
Staged rollouts: introduce new models to specific teams before enabling them tenant-wide.
The team that owns each provider's credential always has access. By default, all other teams also have access unless the provider is explicitly scoped. For AI Agent actions, access is based on the action's team. Workbench access is based on the user, not the selected preset's team, so users can access providers available through any team they belong to. Tenant owners can access all enabled providers in Workbench.
Schema compatibility
When using a custom AI provider that's not listed above, the provider must be schema compatible with the OpenAI API. Set the API type option (under the provider’s extra options) to match your endpoint. The endpoint must support streaming and tool use.
Additionally, if the provider supports the models list endpoint↗, Tines is able to auto discover models. If not, you can still add custom models manually from the models tab.
AI models
When configuring OpenAI compatible APIs, it is possible to select which models are usable within Tines. Once configured, the fast model will power features such as automatic transform. The smart model will be used to power Workbench and other similar features.
When a feature does not have a model selected directly, Tines uses the default smart or fast model. A model selected in Workbench or on an AI Agent action overrides these defaults.
The smart model is used by default for:
Workbench responses, including Workbench for Storyboard, Workbench for Cases, and App building
AI Agent actions in chat mode or task mode with agentic capabilities, such as tools, code analysis, web search, or skills
AI formula generation
Automatic mode
The fast model is used by default for:
Task-mode AI Agent actions without agentic capabilities
Lightweight Workbench tasks, including prompt suggestions and conversation titles
AI-assisted action and story names, story descriptions, change request descriptions, and App names
All custom models enabled on an AI provider are also accessible on the AI action.
Custom AWS Bedrock support for cloud
With custom AWS Bedrock support for cloud, you can choose any models that are enabled in your AWS region and permitted by your Bedrock account.
We recommend enabling the latest Anthropic Claude models for best performance and capabilities.
Model IDs can be obtained from the AWS Bedrock documentation↗.
Credentials for AWS Bedrock
To securely invoke AWS Bedrock APIs, Tines supports AWS authentication by connecting a credential to the provider or through assumed roles for the instance (self-hosted only).
Required IAM permissions
To use Bedrock within Tines, the IAM role must include the following permissions:
{
"Version": "2012-10-17",
"Statement": [
{
"Sid": "BedrockModelAccessPermissions",
"Effect": "Allow",
"Action": [
"bedrock:InvokeModel*",
"bedrock:GetInferenceProfile",
"bedrock:ListInferenceProfiles",
"bedrock:ListFoundationModels"
],
"Resource": "*"
}
]
}AWS Bedrock endpoints
When configuring your connection to AWS Bedrock, you’ll need to specify the correct Amazon Bedrock runtime API endpoint for your AWS region.
It will look something like: bedrock-runtime.<region>.amazonaws.com
View the list of region-specific Bedrock runtime endpoints.↗

Azure OpenAI
When deploying OpenAI models through the Azure AI Foundry↗, additional configuration is required to enable use of those models in Tines. Each model is deployed to a unique URL and must be manually added to the model list.
You will also need set the API Type in Extra Options to Azure
OpenAI models on Amazon Bedrock Mantle
Amazon Bedrock Mantle↗ serves OpenAI models over an OpenAI-compatible API. Configure them as a custom OpenAI provider with these settings:
API endpoint:
https://bedrock-mantle.<region>.api.aws/openai/v1API type: OpenAI Responses (under the provider's Extra options)
Model ID: added manually from the models tab, for example
openai.gpt-5.6-luna
Custom certificate authorities
When creating a custom AI provider, you can select the custom certificate authority to use for the connection. This can be helpful when connecting to your own private AI service.
Tunnel
You can also configure a custom AI provider to connect via a Tines Tunnel. This can be configured under Extra options when configuring your provider. Connecting via a Tunnel allows your cloud tenant to securely access AI services running in your internal networks. Only Tunnels which are accessible by all teams can be used.
For more information visit the Tunnel docs here.

AI credit management
In the AI settings screen you can open AI credit allocations & limits to allocate your AI credits↗ to runtime AI usage (AI action runs in your Stories) as well as set build-time usage limits for Workbench. You can also use this page to view how much of your tenant's monthly AI credit budget has been spent so far this month.

Note:
You can top up your AI credit balance from the Billing screen in Settings - you can learn more about that here↗. If you need to increase your monthly AI credits, contact your Tines account team.
The limits and allocations set from this page only apply when using one of the Tines-provisioned AI providers (i.e. Tines AWS Bedrock or Tines OpenAI).
Team allocations for Stories
Under the Stories tab, you can allocate an amount of credits to each of your teams. This reserves those credits for use in stories owned by that team, protecting their AI Action runs from everyone else's usage. Tenant owners and team admins will receive email alerts when a team reaches 80% and 100% of its allocated credits. To customize the thresholds and recipients for these alerts, take a look at the docs here.
By default, when a team reaches 100% of their allocation, any further AI Action runs for their stories will be stopped. From the Stories tab, you can optionally enable the "Fall back on shared credits" setting.

This setting is applied across all teams and allows any team that reaches 100% of their allocation to continue executing AI Actions in their stories. This extra usage will consume from your tenant's remaining Shared AI Credits (i.e. the amount of credits left unallocated).
Spending limits for Workbench
Under the Workbench tab, you can set spending limits for the users in your tenant. These limits apply across all Workbench surfaces (including Workbench for Storyboard, Apps, and Cases). Limits can be set for individual users, teams, and for the tenant. When setting limits for teams, note that the limit amount applies to each user within that team individually, it does not apply to the team as a whole. The tenant-wide Workbench limit can be set by selecting the Each user option from the dropdown, and likewise this applies to each user individually.
Note:
When a user is covered by multiple limits, the most specific limit is what applies. For example, if a user is covered by an team limit and the tenant-wide limit, then the team limit is their effective limit. In the case of a tie (i.e. multiple conflicting team limits), the highest limit wins out.
All Workbench conversations consume credits from your tenant's remaining Shared AI Credits, this ensures that build-time AI usage does not consume credits that have already been allocated to your teams for runtime usage.
Prompt caching
Tines automatically handles prompt caching when using Tines AI, OpenAI, Anthropic, or custom AWS Bedrock providers. Prompt caching reduces costs and latency by reusing previously processed prompt content. You can learn more about prompt caching here↗.
Prompt caching is ephemeral - cached content automatically expires within minutes and is not stored persistently. Caching does not affect data residency and does not change the security or compliance posture of your AI provider.

Prompt caching controls the amount of tokens that fall under Cached input. These tokens are processed faster and for a lower price.