Home / Docs / Connect a model provider
Models

Connect a model provider

A provider is a service that runs AI models. You give the app a key for that service, the app asks which models it offers, and you pick one for each agent. You can connect as many providers as you like.

Video: Connect any model (1:52)

Add a provider with an API key

  1. Click the gear at the top right to open Settings.
The gear button in the title bar
  1. Settings.
  1. Under Model providers, press Add provider.
  2. Pick your provider from the list.
The list of ready-made providers
  1. Add provider.
  2. Click a provider. OpenRouter is a good first choice: one key reaches hundreds of models.
  1. Get an API key from the provider (see the table below) and paste it into API key.
  2. Press Test and fetch models.
A provider with its key and models
  1. Paste the key here. It is shown as dots.
  2. Test and fetch models checks the key and loads the list of models.
  3. The models that will appear in the picker. Click the × on one to hide it.

When it works, the app says how many models it found:

The message after a successful test
  1. “Connected. 13 models available.”

Keys are stored only on this computer and sent only to the provider they belong to.

Where to get a key

ProviderGet a key atGood to know
Claude (Anthropic)console.anthropic.comClaude models, direct from Anthropic.
OpenAIplatform.openai.comGPT models, plus image, voice and transcription models.
OpenRouteropenrouter.aiOne key for models from many companies.
Google Geminiaistudio.google.comGemini models.
Groqconsole.groq.comFast hosted open models; also fast transcription.
DeepSeekplatform.deepseek.com
Mistralconsole.mistral.ai
xAIconsole.x.ai
Together AItogether.aiOpen models and FLUX images.
Fireworks AIfireworks.ai
Cerebrascloud.cerebras.ai
Z.ai (GLM)z.aiOn the GLM Coding Plan, change the address as the hint in the app says.
Moonshot AI (Kimi)platform.kimi.ai
Qwen (Alibaba Cloud)Alibaba Cloud Model StudioNewer accounts use their own address; the hint in the app shows its shape.
Cloudflare Workers AIdash.cloudflare.comReplace YOUR-ACCOUNT-ID in the address with your account id.

Each provider bills you directly for what you use. VibeCode.surf adds nothing on top.

Choose a model for an agent

Every agent has its own model. Click the model button under the message box.

The model picker
  1. The model button.
  2. Search by model or provider name.
  3. Click a model to use it. The numbers are US dollars per million tokens, in and out.

New agents start on the model of the agent they were opened from.

Use a local model

Local models run on your own computer. They need no key and cost nothing per request. Install Ollama or LM Studio first and download a model there.

  1. Start Ollama, or start the local server in LM Studio.
  2. In Settings, press Add provider and choose Ollama (local) or LM Studio (local).
  3. Leave the key empty and press Test and fetch models.
Ollama connected
  1. The address of the local server. Change it only if you moved the server to another port.
  2. Test and fetch models lists the models you have downloaded.

Local models usually have a smaller memory for the conversation. If a local model loses track, lower Condense chats past in Settings (see Usage and cost).

Add a custom provider

Any server that speaks the OpenAI chat format works: vLLM, llama.cpp, LiteLLM, a company gateway, or a provider that is not on the list.

  1. Press Add provider and choose Custom (OpenAI-compatible).
  2. Fill in the form as shown below.
  3. Press Test and fetch models.
A custom provider
  1. A name you will recognise.
  2. The base URL from the provider's documentation. It usually ends in /v1.
  3. The key, if the server needs one.
  4. Some servers do not list their models. Type the model id here.
  5. Press Add. The model appears in the picker.

If it does not connect

What the app saysWhat it meansWhat to do
The provider refused the API key (401 or 403)The key is wrong, was deleted, or lacks permission.Copy the key again. Check that it has no space at the end.
A message about credit, quota or billing (402 or 429)The account has no credit, or you are sending requests too fast.Add credit at the provider, or wait and try again.
Connected, but the provider listed no modelsThe server has no model list.Add a model id by hand.
Could not connectThe address is wrong, or a local server is not running.Check the base URL. For a local model, start Ollama or LM Studio.

For image, voice and video services, see Images, voice, video and dictation.