BYOK & Local Models
Use your own OpenAI, Anthropic, or Gemini API key, or connect your own local AI server.
Use your own AI provider instead of the platform subscription. Either paste an API key for a cloud provider (OpenAI, Anthropic, or Google Gemini), or point the editor at an OpenAI-compatible local server running on your machine.
Before you start
You'll need one of:
- An API key from OpenAI, Anthropic, or Google Gemini — to use a cloud provider.
- A running OpenAI-compatible local server (e.g. Ollama) — to use a local model.
Turn it on
- Open Settings → Configure AI.
- Find the BYOK/Local card and flip its toggle to ON.
- Pick a Provider — OpenAI, Anthropic, Google Gemini, or Local Server — and set it up as described below.
Once configured, choose which model to use from the model picker in the chat panel.
Use your own API key
For a cloud provider, you supply an API key and the editor calls that provider directly from your browser.
1. Pick a provider — set Provider to OpenAI, Anthropic, or Google Gemini.
2. Paste your API key — get one from your provider's console:
- OpenAI — platform.openai.com/api-keys
- Anthropic — console.anthropic.com
- Google Gemini — aistudio.google.com/apikey
3. Test Connection — the editor sends a tiny prompt to the provider's default model to confirm the key works.
Your key stays in this browser — re-enter it after a refresh
Requests go straight to the provider — your API key and prompts never pass through Vibe Games servers. A browser can't store a key securely, so we keep it in memory only: it's wiped on refresh or when you close the tab. When that happens, a key button in the chat input prompts you to paste it again.
Local Server
Run a model on your own machine with Ollama and point the editor at it. Any OpenAI-compatible server works (LM Studio, llama.cpp, vLLM too), but the steps below use Ollama.
Local models are not recommended for most users
On consumer hardware, locally-run models are typically much slower and noticeably less capable than the cloud providers, so results in the editor will likely be poor. The local option is mainly for experimentation and offline use — use a cloud provider for the best experience.
1. Download a model:
ollama pull gemma42. Serve it with the editor allowed — Ollama blocks browser requests from other origins, so start it with OLLAMA_ORIGINS set to the editor's URL:
OLLAMA_ORIGINS='https://vibe-games.ai' ollama serve$env:OLLAMA_ORIGINS='https://vibe-games.ai'; ollama serve3. Connect the editor — set Provider to Local Server (the URL is pre-filled with http://localhost:11434), click Test Connection, then pick your model from the chat panel's model picker.
Supported models
Pick a model from the model picker in the chat panel. Each cloud provider exposes a fixed set, with the default selected automatically when you switch to it; a local server lists whatever models it reports.
| Provider | Models |
|---|---|
| OpenAI | GPT-5.5 (default), GPT-5.4, GPT-5.4 mini, GPT-5.4 nano |
| Anthropic | Claude Sonnet 4.6 (default), Claude Opus 4.8, Claude Haiku 4.5 |
| Google Gemini | Gemini 3.5 Flash (default), Gemini 3.1 Pro (Preview), Gemini 3.1 Flash-Lite |
| Local Server | Whatever models your server reports |