Local AI Ollama and Player2

Local & easy AI — Ollama and Player2

Blockpal works out of the box with a free built-in AI, and with a HuggingFace (or any OpenAI-compatible) key. As of 3.21.0 there are two more ways to power your companion that need no HuggingFace key — great for running your own models, or for the absolute easiest setup.

Changed in 3.25.0 — there is no priority order any more. A server picks one AI connection and that's the one that runs: MCP, your own API key, Player2, Ollama, the free service, or none. Turning Player2 or Ollama on switches the others off. See Settings → One AI connection at a time.

/ai connection                      # what's running now
/ai connection set ollama           # switch (ops)

Upgrading servers keep whichever provider the old priority order would have chosen, so nothing changes when you update.

Easier than either of these: if you already pay for Claude, ChatGPT, Grok or Gemini, connect it straight to your world with the MCP server — no key in the game, no model to download.

Want a big cloud provider instead? ChatGPT, Claude, Gemini and Grok are all OpenAI-compatible too — pick one in one click with the AI provider presets on the AI & API tab (or /ai admin provider <name>). Those use an API key, so they sit at priority 1 above. See Settings → AI provider presets.

In the settings menu: open /ai menu → AI & API tab. The AI connection picker at the top chooses which one runs; the endpoint and model boxes for Player2 and Ollama sit further down under "Local & easy AI endpoints", and a live "▶ Bots will use: …" line names exactly what will answer. The /ai admin … text commands below still work (and are the way to configure on Bedrock/vanilla clients with no GUI) — /ai admin ollama on now simply switches the connection to Ollama.

Ollama — run your own custom local models

Ollama runs language models on your own machine and exposes an OpenAI-compatible endpoint. Point Blockpal at it and you can run any model you have pulled, with no key and no internet.

# once, on the machine running the server:
ollama pull llama3.2

# in-game (ops):
/ai admin ollama on
/ai admin ollama model llama3.2
/ai admin ollama            # show status

• Default endpoint: http://localhost:11434/v1/chat/completions (change with /ai admin ollama url <url> — e.g. LM Studio, or a remote box).
• Model pool for the Growth village game: add several models with /ai admin ollama models add <id> and each villager is handed a different one, so they "think differently".

Any keyless local OpenAI-compatible server (LM Studio, llama.cpp's server, …) works the same way — just set the URL.

Player2 — the easiest AI (local or online)

Player2 is an OpenAI-compatible AI made for games. It works two ways:

• Local (no key to type): install the free Player2 app, sign into it, and it serves AI on localhost:4315 — no key to paste, no model download. You don't hand Blockpal a token: the app's API is still signed-in-and-authenticated, and Blockpal does Player2's built-in web-login handshake for you to fetch a short-lived key behind the scenes. Just make sure the app is running and logged in before you play.
• Online (cloud): set a PLAYER2_KEY and Blockpal uses Player2's hosted API with the strong gpt-oss-120b model by default.

/ai admin player2 on
/ai admin player2            # show status (local vs online, model, key)
/ai admin player2 url <url>  # repoint the online endpoint if needed

The PLAYER2_KEY (online)

• Blockpal reads the key from the PLAYER2_KEY environment variable (or the -Dplayer2.key JVM property). It is used but never written to disk, and is never baked into the mod jar — the same secure pattern as BLOCKPAL_API_TOKEN.
• On a server you host, set PLAYER2_KEY in the environment before launching.
• A key stored via config is obfuscated at rest (still, prefer the env var).
• Do not hardcode a real key into a public build — the jar is decompilable, so anyone who downloads it would get your key.

Endpoint note. player2.game also publishes an /api/v1/mcp route, which is an MCP server — a different protocol from the OpenAI-style chat-completions that Blockpal speaks. Blockpal's online default is Player2's chat-completions endpoint (https://api.player2.game/v1/chat/completions); use /ai admin player2 url <url> if Player2's chat endpoint ever changes.

"Player2 isn't working"

• Local: the Player2 app must be running and signed in. If it isn't, Blockpal can't get a login key and the bot reports that Player2 didn't authorise the request — open the app, log in, and try again. (Enabling Player2 pre-fetches the login key, so the first request is already authenticated once the app is up.)
• A HuggingFace key wins over Player2. Resolution order is key → Player2 → Ollama → free, so if you (or the server) have an API key set, bots use that and skip Player2. Clear the key (/ai mykey clear, or the admin token) if you specifically want Player2.
• Online: set a valid PLAYER2_KEY — the cloud API rejects unauthenticated requests with a 401.

Related

• Growth — the AI village game
• Per-Player Keys & Models
• Settings · Troubleshooting