Your endpoint: OpenAI, Anthropic and Bedrock

Your endpoint lets Jennefer apps use a model from a cloud provider with your own API key. Each address needs a short request and our approval, because agents can write files, run commands and use tools.

Request access#

1. 1

   Open the request

   In Hub > Models, switch the top card to Your endpoint, then choose Request an endpoint.

2. 2

   Pick the address

   Choose a Provider, and the Address fills in. For Amazon Bedrock, also pick a Region. Only https addresses on the internet work: localhost, IP addresses and local network addresses are refused.

3. 3

   Fill in the request

   Enter your Full name, Contact email, Company or project (optional), your Use case (at least 40 characters) and Models you plan to use (optional). Tick the three agreements and choose Send request.

4. 4

   Wait for the decision

   The tab says Your request is being reviewed. We email you when it is decided.

An approval covers one address. For another provider or region, choose Request another endpoint.

Providers#

Bedrock regions: Europe (Frankfurt, Ireland, Paris, Stockholm), US East (N. Virginia, Ohio), US West (Oregon) and Asia Pacific (Tokyo, Sydney). Each region is its own address with its own approval. The two Bedrock providers in one region share one approval.

Add your models#

Once an address is approved, Your endpoint shows the form:

• Endpoint: pick one of your approved addresses.
• API key: paste it once. It then shows saved, and you type a new one only to replace it. Models on the same address share one key.
• Model: the model name the provider expects. Check connection lists the models it offers, except on Bedrock, which does not list them.
• Effort: Default, Low, Medium or High, plus XHigh and Max on Anthropic and Bedrock Claude. Higher effort is slower and uses more output tokens.
• Context window: from 8K to 1M.
• Choose Save, or Add model for a new one.

Saved models#

• Saved models lists each model with its provider, context window and effort. Use selects one, Edit opens it in the form, the trash icon removes it and Add a model adds another.
• The selected model shows In use while the Hub default is your endpoint, and Selected while it is not. Choose Use your endpoint on the top card to make it the Hub default.
• Apps and windows can pick any saved model in their own picker, in blue. See Model and effort per app.
• Remove endpoint and key, or Remove all models and keys, deletes your saved models and keys.

Where your key is kept#

• The key is kept in Windows Credential Manager on this computer. Apps never see it: their requests go through the Engine, which adds the key.
• Connected apps always use the model on this computer, never your endpoint.
• The key leaves this computer only when you turn on Use from your phone without this computer for a model.

Prompt caching for Claude#

With Anthropic and Amazon Bedrock (Claude), the Engine marks the parts of a request that repeat, such as the agent's instructions and tools, for Claude's prompt cache. The cache lasts 5 minutes. A cache read costs about a tenth of normal input, and a cache write costs a quarter more than normal input. Agent runs send the same parts on every step, so caching lowers their cost. Hub > Usage shows cache reads and writes.

Use from your phone without this computer#

By default, your phone reaches your endpoint through this computer, so the computer must be on. Each saved model has a switch, Use from your phone without this computer.

• When you turn it on, the Hub sends the address, the model, its settings and the key to Jennefer once. The key is stored encrypted and used only for your phone chats and your Cloud sandbox builds from the phone (Build by chat).
• These then work while this computer is off. In Jennefer Mobile the model shows under In your account with Works when your computer is off.
• It needs a Pro plan, and the address must still be approved. Otherwise the phone shows Needs a Pro plan or Address not approved.
• Editing the model sends it again. Turning the switch off, or removing the model, deletes the copy.

Costs#

Your provider bills every request. Jennefer shows an estimated cost from list prices in the apps and in Hub > Usage. Your provider's bill is the real number.

FAQ#