Docs / Get AI provider keys

Get AI provider keys

Step-by-step: create the key or role at OpenAI, Anthropic, Gemini, xAI, Perplexity, Azure and Bedrock.

12 min read

exelens is bring-your-own-key. You create the secret at the provider, paste it once in Add an AI key, we run a tiny test call, then we encrypt it and never show it again. You pay the provider; we do not add an AI line on our invoice.

Anthropic

Console: API keys. Docs: getting started.

  1. 1

    Open API keys

    Sign in at console.anthropic.com → Settings → API keys. You need a workspace with API access and billing or credits.

  2. 2

    Create a key

    Create Key. Name it exelens. Copy the value once — it starts with sk-ant-. Paste it into exelens as API key.

OpenAI

Dashboard: API keys. Docs: authentication, where to find the key.

  1. 1

    Create a project key

    platform.openai.com → API keys → Create new secret key. Restrict it to the project that has billing. Copy the sk-... value immediately.

  2. 2

    Confirm models are enabled

    The key must be allowed to call the chat models you will pick in Build. After you save it in exelens, press Update models if a new model does not appear.

Google Gemini

Studio: Get API key. Docs: Gemini API keys.

  1. 1

    Create a key in AI Studio

    Open AI Studio → Get API key → create in a Google Cloud project that has the Gemini API enabled. Paste the key into exelens. This is the Gemini API key, not a Vertex service-account JSON (that path is for warehouses, not this picker).

xAI

Console: console.x.ai. Docs: xAI API.

  1. 1

    Create an API key

    Sign in → API keys → create. Copy the secret and paste it as API key. Billing must be active on the xAI team that owns the key.

Perplexity

Settings: API. Docs: getting started.

  1. 1

    Create an API key

    Sign in → Settings → API → generate a key. Copy the value once — it starts with pplx-. Paste it into exelens as API key. Billing must be active for Sonar models.

Azure OpenAI

Docs: create a resource and deploy a model, retrieve key and endpoint.

  1. 1

    Create the resource and a deployment

    Azure portal or AI Foundry: create an Azure OpenAI (or AI Foundry) resource, then Deploy a chat model. The deployment name is what you type in exelens — not the model family name.

  2. 2

    Copy endpoint and key

    Resource → Keys and Endpoint. Endpoint looks like https://your-resource.openai.azure.com. Use KEY 1 as the API key. You need all three fields: endpoint, deployment name, API key.

Field in exelensWhere to find it
EndpointKeys and Endpoint on the Azure OpenAI resource
Deployment nameThe name you gave the model deployment
API keyKEY 1 or KEY 2 from the same blade

Amazon Bedrock

No API key is stored. exelens assumes a role in your AWS account. Official docs: model access, third-party role with External ID, Bedrock getting started.

  1. 1

    Turn on the models in Bedrock

    In the target region (for example us-east-1), open Amazon Bedrock → Model access and enable the Claude (or other) models you will use.

  2. 2

    Create a role exelens can assume

    Trust the exelens AWS account and require the External ID shown in Settings when you pick Bedrock. Allow bedrock:InvokeModel, bedrock:InvokeModelWithResponseStream and bedrock:ListFoundationModels on those model ARNs.

  3. 3

    Paste role, External ID and region

    Role ARN looks like arn:aws:iam::123456789012:role/exelens-bedrock. Use the same External ID you put in the trust policy. Region must match where you enabled the models.

OpenAI-compatible endpoint

For a gateway or self-hosted server that speaks the OpenAI Chat Completions API. You need the base URL (for example https://llm.example.com/v1) and an API key the gateway accepts. The host must be reachable from exelens and is subject to the same SSRF rules as other customer hosts.

After the key is saved

  • exelens runs one small request. If it fails, the key is not stored.
  • Set a monthly budget (and optional per-person cap) on the key.
  • Use Update models when the provider ships a new model. If a key dies later, rotate it at the provider and add the new one — chats that used the old key need a model picked again.