> For the complete documentation index, see [llms.txt](https://docs.umbraco.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.umbraco.com/ai-in-umbraco/19.latest/providers/typesafe.md). # TypeSafe AI Configure TypeSafe AI as a provider for the experimental Decision capability. TypeSafe AI provides access to Jev, a model for typed yes/no, pick-one, and score questions. The provider supports only the [Decision](/ai-in-umbraco/19.latest/using-the-api/decision.md) capability. It has no Chat or Embedding models. {% hint style="warning" %} Decision is **experimental** and disabled by default. Until you enable the `Umbraco:AI:Experimental:Decision` feature flag, TypeSafe AI does not appear in the provider list. See [Enabling Decision](/ai-in-umbraco/19.latest/using-the-api/decision.md#enabling-decision). {% endhint %} ## Installation {% code title="Package Manager Console" %} ```powershell Install-Package Umbraco.AI.TypeSafe ``` {% endcode %} Or via .NET CLI: {% code title="Terminal" %} ```bash dotnet add package Umbraco.AI.TypeSafe ``` {% endcode %} Then enable the feature flag in `appsettings.json` and restart the site: {% code title="appsettings.json" %} ```json { "Umbraco": { "AI": { "Experimental": { "Decision": true } } } } ``` {% endcode %} ## Connection Settings | Setting | Required | Description | | -------- | -------- | ----------------------------------------------------------------------------------- | | API Key | Yes | Your TypeSafe AI API key. Stored as a sensitive value and masked in the backoffice. | | Endpoint | No | Custom endpoint URL, for example a proxy. Defaults to `https://api.typesafe.ai`. | Like other sensitive fields, the API key can reference a configuration value instead of storing it in the database. See [Configuration References](/ai-in-umbraco/19.latest/reference/configuration/ai-options.md#configuration-references). {% hint style="warning" %} Keep your API key secure. Never commit it to source control or expose it in client-side code. {% endhint %} ## Models | Model | Description | | ------------ | ----------------------------------------------------------------- | | `jev-latest` | Jev (latest). The only model the provider lists, and the default. | Create a profile with the **Decision** capability and the `jev-latest` model. Decision profiles have no extra settings. ## How Requests Are Sent The provider sends each Decision request to TypeSafe AI as one HTTP call, however many questions it holds: * The request's state is sent as the content to judge. When there is no state, the first question's instructions are sent instead, because TypeSafe AI requires one. * Every question is sent in the same call, keyed by its question ID. The answers come back under the same IDs. * A yes/no question is answered with the probability of "yes". * Pick-one and score questions are answered with a choice or score, a confidence when reported, and a probability for each option or level. Score probabilities are keyed by level index. Options or levels that TypeSafe AI leaves out are filled in with a probability of 0. * Token usage is recorded once for the call. This means [asking multiple questions at once](/ai-in-umbraco/19.latest/using-the-api/decision.md#asking-multiple-questions-at-once) costs a single request to TypeSafe AI. The provider sets no limit on the number of questions in a request. ## Test Connection TypeSafe AI has no models endpoint to check. Instead, **Test Connection** sends one small, authenticated yes/no question to TypeSafe AI: * A valid key passes the test. * An invalid key, or any other error, fails the test. A successful check is cached for one hour for the same endpoint and API key. Repeated tests within that hour do not send a new request. A failed check is not cached. ## Retries When TypeSafe AI responds with `429` (rate limited) or `529` (overloaded), the provider retries the request up to 2 more times: * It waits for the time given in the `Retry-After` header, when present. * Without the header, it waits 1 second, then 2 seconds. * A single wait is never longer than 30 seconds. Other errors are not retried. For example, `401` (invalid API key) and `422` (request rejected) fail straight away. ## Related * [Providers Overview](/ai-in-umbraco/19.latest/providers/providers.md) * [Decision](/ai-in-umbraco/19.latest/using-the-api/decision.md) * [Managing Connections](/ai-in-umbraco/19.latest/backoffice/managing-connections.md) --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation by asking a question. Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters: ``` GET https://docs.umbraco.com/ai-in-umbraco/19.latest/providers/typesafe.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.