phi4minecraftbot

What the mod does

• Listens to all client-side chat; any message containing !phi4 triggers a request to your local Ollama phi4 model.
• Immediately posts [phi4] phi4 is thinking... to server chat, then streams the model reply into chat.
• First streamed chunk is prefixed with the triggering player name (e.g., Player: text); later chunks send just the text.
• Responses are buffered until a newline or sentence-ending punctuation, so chat updates arrive in readable pieces.
• Runs only on the client; original !phi4 messages remain visible.

How to use it

• Keep Ollama running locally on http://localhost:11434 with the phi4 model available.
• Say !phi4 followed by any text (even empty) in chat; the mod will forward the prompt to Ollama and stream back the reply for everyone to see.
• Only one request per player runs at once; extra triggers from the same player show a private notice [phi4] Request already in progress; please wait..
• If the Ollama call fails or times out (60s), you see a private system chat message [phi4] An error has occured!.

Notable implementation details

• Uses the streaming /api/generate endpoint with body { "model": "phi4", "prompt": <question>, "stream": true }.
• Sentence buffering trims leading/trailing whitespace around each emitted chunk and discards newline separators between chunks.

Limitations and future ideas

• Requires a reachable local Ollama service; no fallback to remote endpoints.
• Placeholder lines are not removed when the response finishes.
• Requests from players without an identifiable sender share a single queue slot, so only one of those can run concurrently.