How to use your own AI provider keys in your web app

In TeleportHQ, the AI steps in your web app's workflows run on your own AI provider account, and the AI chatbot for visitors runs on your OpenAI key. Add your API key in the editor under Project settings → Credentials → AI Providers. TeleportHQ saves the key as a project secret, and the provider bills you directly for that usage.

Before you start#

• Get an API key from your AI provider. For example, an OpenAI key from https://platform.openai.com/api-keys.
• Know what these keys are for. They power AI inside your live web app. Building your site with the TeleportHQ AI Assistant or generating pages uses TeleportHQ AI tokens instead. See What are AI tokens, and how many do I get?
• Workflows need a web app. AI steps live in workflows, and adding a workflow to a website turns it into a web app. See What is a web app in TeleportHQ?

TeleportHQ supports these providers: OpenAI, Anthropic, Google, Cohere, Mistral AI, Meta and Perplexity.

Steps#

1. Open your project in the TeleportHQ editor.
2. At the bottom of the left sidebar, click the Project settings cog.
3. Under Credentials, click AI Providers. The provider list opens next to the panel, with a status on each row: Configured, Key missing or Not set up.

4. Click the provider you want, for example Anthropic, to expand it. The help text tells you where to find the key.
5. Paste your key into the key field, or pick an existing project secret.
6. Click Save Anthropic key (the button names the provider). The row closes and shows Configured.

7. To close Project settings, click the cog again.

You can also get here from an AI step in a workflow (Manage AI providers… under the key field) or from the Issues window. When a workflow needs a key you haven't added, the yellow issues chip at the top left of the editor (for example 1 issue) counts it: click it and add the key under Missing Credentials, or click Manage AI providers at the bottom of the window.

Which models you can pick#

Each provider row shows its default model while it's Not set up. In an AI step of a workflow, you pick the model from the provider's list. TeleportHQ offers these models, in the order the list shows them:

Models their providers have shut down are no longer in the lists, for example OpenAI's o4-mini, GPT-4 Turbo and GPT-3.5 Turbo, Meta's Llama models and Perplexity's Sonar Reasoning. Gemini 2.x and Gemini 3.0 models are gone too. An AI step saved with a shut-down model still works on your published site: TeleportHQ publishes it with the provider's replacement. The AI chatbot for visitors uses OpenAI models only; see How to add an AI chatbot to your web app.

Check that it worked#

• The provider row in AI Providers shows Configured, with "Key stored as" and the secret's name.
• To test the key before publishing, open a workflow with an AI step and test that step. See How to test a workflow step.

Troubleshooting#

The provider shows "Key missing" The row says "The secret (name) no longer holds a value — re-enter the key". The project secret was emptied or deleted. Expand the provider, paste the key again and click Save … key.

My Meta key doesn't work For Meta, TeleportHQ runs the models through Together AI, so it expects a Together AI key, not a key from Meta. The field is called Together AI API Key: "Your Together AI API key — Meta's models are served through Together's API." Create the key at https://api.together.ai/settings/api-keys.

There's no Temperature field in my AI step Some models don't accept a temperature setting. For those models, TeleportHQ hides the Temperature field in the step's settings and explains why.

Good to know#

• To change a key later, open AI Providers again, expand the provider and click Replace key.
• Keys are stored with your other project secrets: "Saved to your project's secrets in the editor — the AI assistant never sees the values you enter here." See How to keep passwords and API keys safe (Secrets).
• The AI chatbot for visitors uses the same OpenAI key: saving it in one place covers both.
• AI steps on your published web app call the provider with your key, so they don't use TeleportHQ AI tokens. Training the AI chatbot's knowledge does use TeleportHQ AI tokens.
• An AI step saved with a model its provider has shut down keeps working on your published TeleportHQ site. When you publish, TeleportHQ puts the provider's replacement in that step on the published site, and the step calls the replacement with your key. A site you published earlier keeps calling the shut-down model until you publish it again.
• In the editor, that AI step keeps the model you picked until you pick another. Its model list shows the replacement as chosen, with a note such as "OpenAI has shut down o4-mini" and "Your site uses GPT-5.6 Terra instead. Pick a model above to choose another."
• The Issues window lists each of those steps under AI models that were shut down, and the yellow issues chip at the top left of the editor counts them. A card names the step and says, for example, "Answer questions uses o4-mini, which OpenAI has shut down; your site uses GPT-5.6 Terra instead. Pick a model in the step to choose another." (Answer questions is the workflow's name). Click Open step and pick a model, the replacement or another one, to clear it from the list.
• The replacements TeleportHQ publishes: GPT-5.6 Terra for GPT-3.5 Turbo and o4-mini; GPT-5.6 Sol for GPT-4, GPT-4 Turbo and o1; Claude Opus 4.8 for Claude Opus 3; Claude Sonnet 5.5 for Claude Sonnet 3.5; Mistral Small for Open Mixtral 8x7B; Muse Glimmer 30B for Llama 4 Maverick, Llama 4 Scout, Llama 3.3 70B Turbo and the Llama 3.1 Turbo models; Sonar Reasoning Pro for Sonar Reasoning.