Neuro Lurker

Neuro Lurker is a psychological‑horror presence powered by a local AI (Ollama). There is no fixed script — a language model decides what the creature does: it stalks, leaves anomalies, builds, makes sounds, reads your chat, hears your voice, and escalates over time. When it finally hunts, it chases you, smashes through glass and doors, screams, and one‑shots you. It is immortal — you can't kill it, only survive it. Everything runs on your own machine, offline and free. The AI runs only on the host; friends just need the mod.

Features

• 🧠 LLM‑driven, un‑scripted (Ollama/qwen): builds, digs, breaks/places blocks, plays sounds, spawns mobs, douses lights, leaves anomalies, summons doubles — its own choice every turn.
• ♾️ Thinks continuously at a tunable pace, not just on triggers.
• 🌑 Escalating Director: dormant → traces → night sightings → cave sounds → daytime watching → window peeking → villager loss → hunting. Taunting it (chat/voice) makes it angrier and escalates faster.
• 💬 Reads your chat & hears your voice (RU/EN) — but answers only ~1 in 4 times (silence is scarier).
• 🩸 Immortal hunter: can't be killed; breaks glass/doors/trapdoors/fences/leaves to reach you, screams loudly, and one‑shots on contact.
• 🪦 20+ anomalies: cairns, skull‑on‑post, mushroom rings, graves, soul fire, bone piles, hanging lanterns, dug pits, sculk, wither roses, iron cages, blood trails, glowing signs with your name, a hanged figure, and a copy of your base — with a copy of YOU inside that vanishes as you approach. qwen can also invent its own.
• 👥 Doubles: figures wearing a real player's skin appear and vanish as you get close.
• 🎞️ Custom animated model (GeckoLib, bundled): idle sway, a lurching run, glowing eyes in the dark.
• 🗣️ Whisper voice (optional) via Simple Voice Chat.
• 👨‍👩‍👧 Multiplayer aware: hears every player, personal memory per player.
• ⚙️ All in‑game, no command line: “👁 Lurker” button on title & pause screens — live status, event journal, model auto‑detect, and a one‑click Setup that downloads Ollama, the model and whisper for you.

Requirements

Minecraft 26.1.2 + Fabric Loader + Fabric API · Ollama (host) with qwen2.5:7b (or 14b), downloadable in‑game · GeckoLib bundled.

Installation

1. Install Fabric for MC 26.1.2 — https://fabricmc.net/use/installer/
2. Put lurker.jar + Fabric API (https://modrinth.com/mod/fabric-api) into mods (%appdata%\.minecraft\mods / ~/Library/Application Support/minecraft/mods).
3. Launch the fabric‑loader 26.1.2 profile.

Set up the AI (host) — no command line

Install & run Ollama (https://ollama.com/download) → in game “👁 Lurker” → ⚙ Setup → Download model → I am host = YES → Save.

Optional voice

Simple Voice Chat (all) lets it speak; whisper (host) lets it hear — ⚙ Setup → Install whisper downloads it for you (or point the “whisper folder” at whisper-server + ggml-small.bin).

Commands

/lurker spawn · hunt · doubles · anomaly · status · advance · day [n] · config (continuous on|off, interval <sec>, whisper <path>, host, model, url)

System requirements

 

👥 Guests (just the mod)

Same as normal modded Fabric 26.1.2:

• 4‑core CPU · 8 GB RAM · any GPU that runs Minecraft · ~2 GB free disk.
• They don't run the AI, so the mod adds almost no load.

🖥️ Host — Minimum (use qwen2.5:7b)

• CPU: modern 6‑core (Ryzen 5 / Core i5)
• RAM: 16 GB
• GPU: 6 GB VRAM (GTX 1660 / RTX 2060) — or Apple Silicon with 16 GB unified memory
• Disk: ~15 GB free (SSD)
• OS: Windows 10/11 64‑bit, or macOS (Apple Silicon)
• ⚠️ CPU‑only works, but the creature reacts slowly (responses take several seconds).

🚀 Host — Recommended (qwen2.5:14b, continuous mode, smooth)

• CPU: 8‑core+ (Ryzen 7 / Core i7)
• RAM: 32 GB
• GPU: 10–12 GB VRAM (RTX 3060 12 GB / 4060 Ti or better) — or Apple Silicon with 24 GB+ unified memory
• Disk: ~25 GB free SSD
• OS: Windows 11, or macOS (Apple Silicon)

Weak / no GPU?

• Use qwen2.5:7b instead of 14b.
• Turn off continuous mode (/lurker config continuous off) or raise /lurker config interval 60.
• Speech (whisper) is light: ~1 GB RAM and runs fine on CPU.

  📥 Network & downloads — full disclosure

  The mod does NOT download or execute anything on its own. It ships no bundled binaries and makes zero network requests at launch or during normal gameplay. All downloads are 100% user-initiated, optional, and host-only — triggered by clicking a button on the in-game Setup screen.

  If a user prefers, they can install everything manually and the mod never touches the network. The optional helper simply saves non-technical players from using a terminal.

  What the Setup buttons download, and from where (official first-party sources only):

  Component	Source (official)	Purpose
  Ollama installer	https://ollama.com/download/...	The local AI runtime
  AI model (e.g. qwen)	via the user's local Ollama (ollama.com registry)	The language model
  whisper (optional voice)	https://github.com/ggml-org/whisper.cpp/releases	Local speech-to-text
  whisper model	https://huggingface.co/ggerganov/whisper.cpp	STT model file

  Safeguards:

  • Nothing is run silently. The Ollama installer is opened for the user to run manually — the mod never executes an installer in the background.
  • Downloads only occur after an explicit button press by the host, with on-screen progress.
  • No data is ever uploaded; the AI runs entirely on localhost. No analytics, no telemetry.
  • All source code is open and unobfuscated.