Bring Your Own Key

With your own API key, your agents reach a model on your own provider account — OpenAI, Azure OpenAI, Azure AI Inference or DeepSeek — instead of a Qlar model. The provider bills you directly, and no Qlar credit is charged for that model's tokens.

Each key you add is one model: an alias your team picks from model lists, the provider and model it reaches, and the key to reach it with.

Who can do this? Owners and Admins, in an organization that Qlar has allowed to bring its own AI models. If the Sources tab or Add API key doesn't appear, ask Qlar support to turn it on for your organization.

Before you start

• An account with a supported provider and an API key for it:
  • OpenAI — the model name, for example gpt-4o-mini.
  • Azure OpenAI — the endpoint and the deployment name.
  • Azure AI Inference — the endpoint, and optionally a Response API endpoint.
  • DeepSeek — the model name; the endpoint defaults to https://api.deepseek.com.
• What the model accepts and produces (images, audio, documents, …), from the provider's own documentation. Qlar can't detect it.
• Owner or Admin role in the organization.

Want to use your Claude or ChatGPT plan instead, with the key never leaving your own computer? See Connect a Gateway.

Add a key

Open the Organization page → Configure on your organization → AI Models tile → Sources tab → Add API key (or Add API key in the Your API keys group on the Models tab).

1. Alias — the name your team sees in model lists, for example my-gpt-4o. It must be unique and can't be the name of a Qlar model. It can't be changed later.
2. Provider — OpenAI, Azure OpenAI, Azure AI Inference or DeepSeek.
3. Invocation method — Chat Completion or Response API. DeepSeek supports Chat Completion only.
4. Native model name (for example gpt-4o-mini) or, for Azure, the Deployment name, plus the Endpoint for Azure providers.
5. API key — pasted once. It is stored encrypted; afterwards only a masked part of it is shown.
6. Optionally a Short description and an Estimated credit ratio (leave 0 if unknown).
7. Capabilities — switch on what the model accepts (Image, Audio, Video, Document) and produces (Image, Audio, Video). With Document on, pick at least one document type.
8. Click Create.

The model appears under Models in Your API keys, switched on.

Use it

• In an agent: as the conversation model, in Allowed Conversation AI Models, or as the Answering AI model of a Natural conversation voice call. Model lists show it under API key.
• As a feature default: on the Defaults tab, choose it for Conversation, Summarization, Personalization, Reranker, KB Training or Conversation Analysis.
• Billing: no Qlar credit for the model's tokens; your provider bills you. Tools the answer uses — web search, knowledge base, plugins — are billed as usual.

Change or remove a key

Troubleshooting

Related Pages

• AI Models (organization)
• Connect a Gateway
• AI Models screen reference
• Model & Context