Connect your own AI model
Add a key for OpenAI, Anthropic, Azure AI Foundry, Kimi or any OpenAI-compatible endpoint, and test it.
In short
The assistant, report summaries and resume auto-fill run on a model, and you can bring your own key, so the vendor, the model and the bill are yours. Adding a key is free on every plan. Once it is tested, your own key answers before the platform's AI.
Before you start
- Who: anyone signed in can open Settings › AI and add a key, even before AI is in their plan. On Business and Enterprise an organization admin can also save one key for the whole organization.
- Plan: adding a key is free on every plan. The provider bills you; {{brand}} does not.
- First: an API key from your provider. For Azure AI Foundry or an OpenAI-compatible endpoint, also the endpoint as a full https:// address.
Steps
- Open Settings › AI and press "Use your own key instead" ("Add AI key" when nothing answers yet, "Add another key" once you have one).
- Choose the provider and give the key a name you will recognize, such as My OpenAI.
- Paste the API key. Type a model name, or leave it blank to use the default. For a saved key you can also name a "Model for light work", used for summaries and short answers; blank uses the same model.
- For Azure AI Foundry or an OpenAI-compatible endpoint, enter the endpoint as a full https:// address.
- Under "Where should this key live?", choose "Save it" or "Just this session". See /help/ai/saved-and-session-keys.
- Press "Test key" and wait for "The key works." If it says "The key did not work", read the reason and fix it.
- Press "Add provider", or "Use for this session".
Result: the key shows as a card in Settings › AI, and the page says which key answers where you are.
Providers
OpenAI-compatible covers Ollama, vLLM, LM Studio, OpenRouter or a server in your own network. The model is a free-text box, so any model the provider offers works. A server on a private network is reachable only where the deployment allows it.
Good to know
- Anyone signed in can open Settings › AI, even before the AI module is in their plan.
- Each person connects their own key. On Business and Enterprise, an organization admin can also save one key for the whole organization in Organization settings → AI.
- By default your own key answers first - the session key before a saved one - then your organization's key, then the platform's AI.
- An organization admin may choose that only the organization's key answers, or only the platform's AI, so the organization's data never reaches members' own AI accounts. Then your own key is not used in that organization.
- Jobs' candidate features - CV and answer summaries, follow-up topics and transcription - always use the platform's AI, because their safety checks are measured on its models.
- An organization that keeps AI in the European Union is answered by the platform's deployment there and no other key.
A key is a billing secret Paste it only into the API key box, never into a question, a request header, a card or a note. Every question, summary and test costs tokens on your account. An endpoint pointed at the wrong server sends your data there.
If something goes wrong
Your organization's own model
On Enterprise and in the trial, an organization can answer every AI feature with a model of its own: an OpenAI-compatible server it runs, an Azure OpenAI resource or an Azure AI Foundry endpoint.
- Organization settings › AI › "Your own model" › "Set up your own model".
- Choose where it runs, then give its address and the model or deployment name.
- Paste its key and press "Test the call". Save only once the test passes; the key is never shown again.
- Then every AI feature in the organization uses it: no member's own key and never {{brand}}'s AI. If it does not answer, AI says so - nothing is sent anywhere else.
- Features that run only on {{brand}}'s measured models - the candidate summaries of Jobs - are off while your own model answers.
- Its answers are your bill, not the plan's AI budget. A model on a private network needs your installation to allow private AI endpoints.
Related
- /help/ai/saved-and-session-keys
- /help/ai/asking-the-assistant
- /help/ai/answers-left