Use a different provider

The agent talks to AI models in the OpenAI chat completions format, the format most providers and AI gateways accept. ApexAgent comes with a connection to OpenRouter already set up, so you can start with one key. You can switch to your own provider at any time.

To change model without changing provider, see Use a different model.

Which providers work

Provider	Status
OpenRouter	Preconfigured and tested
OpenAI, Azure OpenAI	Use the same format. Follow Connect your own provider
A company AI gateway, or another OpenAI-compatible service	Follow the same steps
Google Vertex AI	Not tested by us. Uses the same format, but signs in with a short-lived token instead of a fixed key, so its credential in Salesforce is set up for OAuth rather than a pasted key
Google Gemini, directly	Has an OpenAI-compatible endpoint. Not tested by us; if a field differs, it is the same small code change as below
A provider with its own format	A small code change. See A provider with a different format

OpenRouter is the one we test. Other OpenAI-compatible providers should work the same way for chat, tool calls, token counts, cost tracking and logs; What to check on a new provider lists what is most likely to differ.

Connect your own provider

You need the provider’s base URL, its chat completions path, and an API key. Three examples:

Provider	URL	Chat Completions Path	Key is sent as
OpenAI	https://api.openai.com	/v1/chat/completions	Authorization: Bearer <key>
Azure OpenAI	https://<your-resource>.openai.azure.com	/openai/v1/chat/completions	api-key: <key>
OpenRouter	https://openrouter.ai	/api/v1/chat/completions	Authorization: Bearer <key>

1. Store the key

In Setup, open Named Credentials and go to the External Credentials tab. Select New.

• Label and Name: for example My Provider Credential
• Authentication Protocol: Custom

Save it. Under Principals, select New, name the principal, and add an authentication parameter:

• Name: ApiKey
• Value: your API key

Under Custom Headers, select New and add the header your provider expects. Use the names you chose above in place of My_Provider_Credential:

• For Authorization: Bearer <key>: name Authorization, value {!'Bearer ' & $Credential.My_Provider_Credential.ApiKey}
• For api-key: <key>: name api-key, value {!$Credential.My_Provider_Credential.ApiKey}

2. Create the named credential

On the Named Credentials tab, select New.

• Label and Name: for example MyProvider. You enter this name on the LLM record
• URL: the provider’s base URL
• External Credential: the one you just created
• Turn on Allow Formulas in HTTP Header, and turn off Generate Authorization Header

3. Let your users use it

Open the ApexAgent User permission set, go to External Credential Principal Access, and add the principal you created. Do the same on ApexAgent Admin. Without this, the agent fails with a credential error for that user.

4. Create the LLM record

In the Setup Portal, open LLMs and select New. Use a different model explains every field.

Field	What to enter
Model API Name	The model’s name exactly as the provider gives it. On Azure, your deployment name
Named Credential	The name from step 2
Chat Completions Path	The provider’s path, from the table above
Family	The model’s maker, such as OpenAI, Anthropic or Google
Supports Reasoning?	Tick if the model can reason before answering
Token costs	The provider’s prices per million tokens. These drive the cost figures in the logs and dashboard

5. Use it

Open your Agent record, set LLM to the new record, and send a message.

If it fails, open Agent Error Logs. The provider’s own error message is saved there, and it usually names the setting it did not accept.

What to check on a new provider

Two settings on the Agent record depend on the model, on any provider:

• Reasoning Effort is sent as reasoning_effort. Set it to disabled for a model that does not support reasoning.
• Max Tokens is sent as max_completion_tokens.

The logs keep the reasoning token count for every provider that reports it. The model’s reasoning text itself is not saved.

A provider with a different format

If your provider or gateway does not use the OpenAI format, or names a field differently, the change is small and stays in two classes:

• ApexAgentLLMClient: builds the request and reads the response
• ApexAgentApiTypes: the request and response shapes

Give those two files and your provider’s API reference to a coding agent and ask it to adapt them. Nothing else in ApexAgent needs to change: tools, logs, the chat window and the MCP server are unaffected.