Skip to main content
Status: Available. Connect TypeSafe with one API key from the TypeSafe console. Your AI agent can then ask Jev, TypeSafe’s model, typed questions about any text or record and act on calibrated answers. Nothing at TypeSafe changes. Explore the TypeSafe connector.

​
Connect and list your models

Before you start
  • A TypeSafe account at console.typesafe.ai.
  • An API key created in the TypeSafe console.
1

Open TypeSafe in Connections

Choose Connect or Add connection.
2

Enter the API key

Paste the key from console.typesafe.ai. ApexGenius supplies the API address and checks the key against the model list.
The Connect TypeSafe dialog in ApexGenius with the empty TypeSafe API key field, the How to get it link and the Connect button.The Connect TypeSafe dialog in ApexGenius with the empty TypeSafe API key field, the How to get it link and the Connect button.

Paste the API key from console.typesafe.ai and choose Connect. ApexGenius checks it against the model list before saving.

3

Check the saved connection

Save after validation succeeds. Review the Tools tab. Evaluations are reads that bill input tokens to your account.
Try a first question with your AI agent
Confirm jev-latest appears in the list. Listing models bills nothing; evaluations bill input tokens.

​
How Jev answers

Jev returns typed answers with probabilities, not text. Each question uses one of three primitives:
  • Yes or no (noul): returns the probability that the answer is yes.
  • Score (score): takes a list of level labels and returns a number with its confidence.
  • Choice (choice): takes options with their meanings and returns the pick, its confidence and the probability of each option.
Put several questions in one evaluation. They run against the same state, the text or record you want judged. Put your business definition inside the question, including who is not a fit, so Jev reads the state the way you do.

​
What your AI can do

  • Check which Jev models your key can use.
  • Classify search terms before you add negatives.
  • Score ad copy or landing page headlines before you spend.
  • Qualify inbound leads and support tickets.
Three ApexGenius tools reach these operations: one to search them, one to read and one to write. Both operations are reads, so an AI agent set to Read only can use them. The write route has nothing to call. Open Skills on the connection to read TypeSafeJev. It explains how to phrase each question type and when to batch them.

​
Cost and limits

  • Each evaluation bills input tokens to your TypeSafe account. Listing models bills nothing. ApexGenius adds no TypeSafe charge.
  • Keep the state and its questions under the model’s context limit.
  • Disconnecting stops future access. It does not refund an evaluation that already ran.
If a layer fails (the connection, the tools, or the skill), TypeSafe is partially ready. A saved key by itself does not make the connector usable.

​
Troubleshooting

  • TypeSafe did not accept the key: copy the key again from console.typesafe.ai and reconnect.
  • Reconnect required: the key was revoked. Create a new key and reconnect.
  • An evaluation fails for size: split the state or ask fewer questions in one call.
Explore the TypeSafe connector · Open TypeSafe Connections

​
Official TypeSafe documentation

Last modified on October 26, 2026