Getting started with AI
Pick local Ollama or a cloud preset so the in-app agent can create, start, and stop a game server.
In-app chat needs a model under Settings → In-app agents. External MCP clients (Cursor, Claude Code, Codex, OpenClaw, Hermes) bring their own model — skip this page and see External agents (MCP).
Home 0.2.4 slimmed the in-app tool set, so a cheaper cloud model or a local 7B can install a server. You do not need a huge model for a LAN night.
What to pick
A model is good enough when it can create → start → check health → stop a game server. That is the bar for a party box.
| Your situation | Start with |
|---|---|
| Linux mini-PC / i9, no cloud spend | Ollama + qwen2.5 7B (qwen2.5:latest) |
| Internet, want the Home default | Venice + grok-4-5 |
| You already pay OpenAI / Anthropic / Gemini / … | That cloud preset — min and recommended below |
| Proxy or self-hosted OpenAI-compatible API | Custom |
Local path that is true today: this runs on a Linux mini-PC / i9 with Ollama and qwen2.5 7B. Not 3B. Not llama3.2. An RX 580 / Polaris box is the wrong machine.
Local (Ollama)
Recommended: qwen2.5 7B (qwen2.5:latest, ~4.68GB). Proved create / start / health / stop on a CPU box (a few minutes on an i9).
qwen2.5 3B can call tools in a pinch. Do not use it as the LAN-night path — it is messy and does not reliably finish create → start → health → stop.
Setup: Ollama (offline).
Cloud presets
Use these ids in Settings → In-app agents. Prefer the recommended column unless you are saving cost.
| Provider | Minimum | Recommended |
|---|---|---|
| Venice | gpt-4o-mini | grok-4-5 (Home default) |
| OpenAI | gpt-4.1-mini | gpt-4.1 |
| Anthropic | claude-haiku-4-5 | claude-sonnet-5 |
| Google Gemini | gemini-3.6-flash | gemini-3.6-flash |
| OpenRouter | google/gemini-2.5-flash | anthropic/claude-sonnet-5 |
| DeepSeek | deepseek-chat | deepseek-chat |
| NVIDIA | meta/llama-3.1-8b-instruct | meta/llama-3.3-70b-instruct |
Venice extras
Also works: llama-3.3-70b, mistral-small-3-2-24b, qwen 27B, deepseek-v4-flash, kimi-k2-5.
Skip Gemma 27B and GLM flash.
Gemini vs OpenRouter
On the Google Gemini preset, set gemini-3.6-flash. Native gemini-2.5-flash / gemini-2.5-pro 404 for new Google AI keys. OpenRouter still serves google/gemini-2.5-flash.
DeepSeek reasoner is untested — stay on deepseek-chat.
Keys stay encrypted on your PlayOn host. Per-provider setup (how to get a key and paste it): LLM providers.