LLM providers
Choose a cloud or local model for in-app agents. Keys stay on your host.
| Provider | Needs API key | Default model | Guide |
|---|---|---|---|
| Venice | Yes | grok-4-5 | venice |
| OpenAI | Yes | gpt-4.1 | openai |
| Anthropic | Yes | claude-sonnet-5 | anthropic |
| Google Gemini | Yes | gemini-3.6-flash | gemini |
| OpenRouter | Yes | anthropic/claude-sonnet-5 | openrouter |
| DeepSeek | Yes | deepseek-chat | deepseek |
| NVIDIA | Yes | meta/llama-3.3-70b-instruct | nvidia |
| Ollama (offline) | No | qwen2.5:latest | ollama |
| Custom (OpenAI-compatible) | Yes | — | custom |
In-app agents (the admin chat) need an LLM provider under Settings. External MCP clients bring their own model — they only need a PlayOn access token (MCP overview).
Min and recommended models for a LAN night: Getting started with AI.
Cloud vs offline
| Situation | Start with |
|---|---|
| Normal home / party with internet | Venice (grok-4-5), NVIDIA, or any cloud preset you already pay for |
| Air-gapped or no cloud spend | Ollama (offline) + qwen2.5 7B on the party box or another LAN host |
| You already have OpenAI / Anthropic / Gemini keys | Use that preset — same Settings form; pick the recommended model |
| Proxy or self-hosted OpenAI-compatible API | Custom |
Keys are stored encrypted on your PlayOn host.
How to connect any preset
- Create an account / API key with the provider (see the guide for that provider).
- PlayOn → Settings → In-app agents.
- Choose the provider from the list.
- Paste the key (if required), confirm model, save.
- Open Map / chat and send a short message to verify tool calling works.
Prefer models that support function / tool calling. High-impact actions still require host Confirm in the UI unless you use MCP auto-approve on a trusted token.