API Endpoints

Learn how to configure and manage API endpoint configurations in the Management Portal.

Overview

An API endpoint defines how FoundationaLLM connects to an external or internal service: the URL, the provider, the authentication method, and optional headers. AI models, data sources, data pipeline plugins, and platform services reference API endpoints by name.

MCP server endpoints are API endpoints with the MCP Server category. They are usually managed on the Agents > MCP Servers page, but they also appear in the API Endpoints list.

Open the API Endpoints Page

Sign in to the Management Portal.
In the sidebar, select Platform > API Endpoints.

Endpoint List

The list shows every API endpoint you can read, sorted by name, with 10, 25, 50, or 100 rows per page.

Column	Description
Name	Display name (or resource name when no display name is set)
Provider	Hosting provider, with a filter
Category	Functional role, with a filter
Subcategory	Optional refinement, with a filter
Actions	View JSON definition (`pi-code`), Edit (`pi-pencil`), Delete (`pi-trash`)

Use Search endpoints to filter by display name or resource name. The search text is kept when you return to the page.
Use the refresh button to reload the list.
Edit and Delete are disabled when you lack the FoundationaLLM.Configuration/apiEndpointConfigurations/write or /delete permission.

Create an API Endpoint

Select Create API Endpoint.
Complete the sections described below.
Select Create API Endpoint. Select Cancel to discard your changes.

Identity

Field	Description
Resource name *	Unique name. Letters, numbers, dashes, and underscores only. A ✔️ or ❌ shows whether the name is available. Cannot be changed after creation.
Display name *	Shown in lists and pickers. Defaults to the resource name.
Description	Purpose of the endpoint

Classification

Field	Description
Provider	The platform that hosts the API: Amazon Bedrock, Anthropic, Azure AI Inference API, FoundationaLLM, Generic, Google Agent Platform, Google VertexAI, LangChain, Microsoft, OpenAI, or TypeSafe
Category	AzureAIDirect, AzureOpenAIDirect, ExternalOrchestration, FileStoreConnector, Gatekeeper, General, LLM, MCP Server, or Orchestration. AI model endpoints use General.
Subcategory	None, OneDriveWorkSchool, Indexing, or AIModel. AI model endpoints use AIModel.
Operation type	Shown only for the Microsoft and OpenAI providers: Chat Completions (default), Completions (legacy), or Responses

Connection

Field	Description
Endpoint URL *	Service URL
API version	Optional dated API version (for example, Azure OpenAI). Hidden when the operation type is Responses.
Timeout (seconds)	Request timeout. Defaults to 60, or 300 when you choose Responses on a new endpoint.
Status endpoint	Relative path of the service health endpoint, if any (for example `/status`)
Project	Optional project name used by some providers, such as Google. Select the ✕ button to reset it to empty.
Location	Optional region used by some providers, such as Google, to pin the endpoint. Select the ✕ button to reset it to empty.

Authentication

Field	Description
Authentication type	API Key, Azure Identity, or Google Service Account
Authentication parameters	Key/value pairs required by the authentication type. Use Add Authentication Parameter to add a pair.

Common authentication parameters:

Authentication type	Parameter key	Value
API Key	`api_key_configuration_name`	App Configuration key that holds the API key (for example, a Key Vault reference)
API Key	`api_key_header_name`	Header that carries the key, when the service needs a specific one
API Key	`api_key_prefix`	Optional prefix added before the key value
Azure Identity	`scope`	Optional token scope override
Google Service Account	`service_account_credentials`	App Configuration key that holds the service account JSON credentials

HTTP headers

Headers sent with every request to the endpoint. Values can be raw values or config:// references to App Configuration keys. When you save, raw values that look like secrets are stored in Key Vault and replaced with config:// references automatically.

URL exceptions

Per-user URL overrides, typically used to route one developer's requests to a locally running service. Select Add URL Exception and enter the URL, the Principal Name (UPN), and whether the exception is Enabled. See Configuring URL exceptions for local development.

Provider-Specific Guidance

Azure OpenAI and Microsoft Foundry (Chat Completions)

Field	Value
Provider	Microsoft
Operation type	Chat Completions
Endpoint URL	`https://<resource>.openai.azure.com/`
API version	A dated API version, for example `2024-02-01`
Authentication type	Azure Identity or API Key

Responses operation type

Use Responses for models that require the OpenAI Responses API for tool calling, such as GPT-6 Astra.

One Azure OpenAI or Foundry resource that hosts both Chat Completions models and Responses models needs two endpoints with the same URL and authentication: one with Chat Completions and one with Responses. Each AI model points at one of them.
Enter the resource URL (https://{resource}.openai.azure.com, https://{resource}.cognitiveservices.azure.com, or https://{resource}.services.ai.azure.com), a Foundry project URL (https://{resource}.services.ai.azure.com/api/projects/{project}), or an Azure portal Responses URL. The form shows Will be stored as … and, on save, stores the /openai/v1/ URL. If Project is set and the URL has no project path, the project path is added.
For the OpenAI provider, the URL is stored as https://api.openai.com/v1/.
Do not use a dated API version. Use a timeout of 300 seconds or more for reasoning models.
With Azure Identity, the token scope defaults to https://cognitiveservices.azure.com/.default for *.openai.azure.com hosts and https://ai.azure.com/.default for *.services.ai.azure.com hosts. Set the scope authentication parameter only to override it.

OpenAI

Field	Value
Provider	OpenAI
Endpoint URL	`https://api.openai.com/v1`
Authentication type	API Key
Operation type	Responses for models that require it; otherwise Chat Completions

Google Agent Platform and Google VertexAI

Field	Value
Provider	Google Agent Platform (Google VertexAI is the older provider name)
Authentication type	Google Service Account
Authentication parameters	`service_account_credentials` = App Configuration key that holds the service account JSON
Project	Google Cloud project ID
Location	Google Cloud region, for example `us-central1`

Amazon Bedrock

Use a Bedrock endpoint for Claude completion models (workflow and tool LLMs). Set the AI model's deployment name to the Bedrock model ID or inference profile ID. Claude thinking, effort, and token budget are configured on the AI model, not on the endpoint. See Claude on Amazon Bedrock.

Anthropic endpoints (API key for api.anthropic.com) are used by the Claude Agent tool, not by workflow completions.

TypeSafe (Jev)

Jev is TypeSafe's System One decision model. It is not hosted on Azure OpenAI or Microsoft Foundry; FoundationaLLM calls the TypeSafe API directly (POST /v1/systemone).

Field	Value
Provider	TypeSafe
Endpoint URL	`https://api.typesafe.ai`
Authentication type	API Key
Deployment name (on the AI model)	`jev-latest`, or a pinned ID such as `jev-1.13.0`

Jev does not generate prose and does not call tools. Put the JSON map of questions in the agent prompt; the user message is the state, and the reply is the decision. Use an agent with no tools.

Edit an API Endpoint

Select the pencil icon for the endpoint.
Change the settings. The resource name is shown below Display name and cannot be changed.
Select Save Changes.

Delete an API Endpoint

Select the trash icon for the endpoint.
In the Delete API Endpoint dialog, select Yes.

Warning

Deleting an endpoint breaks any AI model, data source, or service that references it. Check dependencies first.

Access Control

Permission	Description
`FoundationaLLM.Configuration/apiEndpointConfigurations/read`	View endpoints
`FoundationaLLM.Configuration/apiEndpointConfigurations/write`	Create and edit endpoints
`FoundationaLLM.Configuration/apiEndpointConfigurations/delete`	Delete endpoints

Troubleshooting

Issue	Resolution
"Please provide a url for the API endpoint."	Enter the Endpoint URL.
Responses calls fail with 404	Use the resource URL, or a project URL only when the Foundry project exists on that AI Services account.
Authentication errors	Check the authentication type and parameters, and that the referenced App Configuration keys exist. For Azure Identity, check the identity's role assignments on the target resource.
Google model calls fail	Check `service_account_credentials`, Project, and Location.

Table of Contents