惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

WordPress大学
WordPress大学
A
About on SuperTechFans
量子位
B
Blog RSS Feed
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
博客园_首页
MongoDB | Blog
MongoDB | Blog
小众软件
小众软件
Blog — PlanetScale
Blog — PlanetScale
Microsoft Azure Blog
Microsoft Azure Blog
V
V2EX
Google DeepMind News
Google DeepMind News
Cyber Security Advisories - MS-ISAC
Cyber Security Advisories - MS-ISAC
H
Hackread – Cybersecurity News, Data Breaches, AI and More
G
Google Developers Blog
U
Unit 42
D
DataBreaches.Net
博客园 - Franky
D
Docker
宝玉的分享
宝玉的分享
Y
Y Combinator Blog
月光博客
月光博客
freeCodeCamp Programming Tutorials: Python, JavaScript, Git & More
Hugging Face - Blog
Hugging Face - Blog

Hacker News - Newest: "AI"

AI can't read an investor deck AI as an attorney? Student uses ChatGPT, Gemini to sue UW over alleged racial discrimination Hacking MCP Servers in AI Systems – The Rug Pull: Tool Changes After Approval GitHub - MeepCastana/KubeezCut: Free Web based video editor Can AI judge journalism? A Thiel-backed startup says yes, even if it risks chilling whistleblowers Coming soon: 10 Things That Matter in AI Right Now DARPA built an AI to fact-check enemy weapons claims What explains heterogeneity in AI adoption? When AI Meets Muscle: Context-Aware Electrical Stimulation Promises a New Way to Guide Human Movements - Department of Computer Science AI Changed How We Build. It Did Not Change What Matters. Linux rules on using AI-generated code - Copilot is OK, but humans must take 'full responsibility for the… Meta spins up AI version of Mark Zuckerberg to engage with employees Code Mode: Let Your AI Write Programs, Not Just Call Tools | TanStack Blog GitHub - Delavalom/graft: Go framework for building AI agents. Type-safe tools, multi-provider (OpenAI, Anthropic, Gemini, Bedrock), zero vendor SDKs. India's TCS tops estimates, says new AI models did not dent services demand Gen Z's fading AI hype Strong feeling: we are in a folded AI reality GitHub - machinarii/total-recall-catalog: A reference catalog of latest knowledge retrieval, memory & RAG systems GitHub - mensfeld/code-on-incus: Give each AI agent its own isolated machine with root, Docker, and systemd. Active defense detects and stops threats automatically.. Quantization, LoRA, and the 8% Problem: Benchmarking Local LLMs for Production AI Iran war: We spoke to the man making Lego-style AI videos that experts say are powerful propaganda Powell, Bessent discussed Anthropic's Mythos AI cyber threat with major U.S. banks GitHub - immartian/bellamem: Persistent belief-graph memory for AI agents. Retrieves decisive context by importance — not recency, not RAG, not /compact. recursive-mode: The Repo-Native Operating System for AI Engineering After the attack on Sam Altman's home, will AI CEO's go on the offensive? The biggest advance in AI since the LLM Opus 4.6 vs GPT 5.4 One Prompt Unity World Generation Test “AI polls” are fake polls Client Challenge Can AI be a 'child of God'? Inside Anthropic's meeting with Christian leaders
GitHub - ninjahawk/KillClawd
ninjahawk1 · 2026-05-09 · via Hacker News - Newest: "AI"

KillClawd is a desktop pet powered by a local LLM. It runs as a transparent always-on-top overlay — a tiny AI crab called Clawd who wanders your desktop, reacts to your cursor, fights mobs, explores castles, rides vehicles, and has genuine opinions about your file organization.

Its core: ~2K lines of vanilla JS + Electron + Ollama. No cloud. No subscription. The model runs entirely on your machine.

Design philosophy: Clawd should feel like he's doing his own thing and you're interrupting him — not the other way around.


🚀 Quick Start

Requirements: Windows, Node.js 18+, Ollama

# 1. Pull a model (do this once)
ollama pull qwen:latest
# 2. Clone
git clone https://github.com/ninjahawk/KillClawd.git
cd KillClawd

Then double-click clawd.bat. It installs dependencies on first run, then starts the app — just double-click it every time after that.

qwen:latest (4B) is the recommended model — fast enough for real-time chat. Any Ollama model works. The app auto-detects what you have.


🖱️ Controls

Action What it does
Double-click Clawd Open chat — talk to him
Drag Clawd Pick him up
Fling (drag fast + release) Throw him, he flies and bounces
Right-click Clawd Context menu
Hover near Clawd He notices and eventually flees

Right-Click Menu

Option Effect
💬 talk to him Open chat input
🤔 what's he thinking Trigger immediate thought
😰 existential crisis Force a crisis (stops him, generates something dark)
😴 make him sleep He says "fine." and sleeps
⚡ go crazy Full sprint mode
🫳 pet him He tolerates it
👾 spawn mob Enemy appears, Clawd fights it
🏰 spawn castle Castle appears, Clawd explores it
🚀 spawn vehicle Vehicle appears, Clawd rides it
✕ quit Closes app, unloads model from VRAM

📋 Features

  • LLM-powered personality — every response goes through a local Ollama model. Dry, blunt, occasionally existential.
  • Cursor awareness — Clawd notices when you follow him, escalates from "i see you" to "i will remember this"
  • Throw physics — grab and fling him across the screen, he bounces off edges
  • Mobs — spawn 👾🦠👹🤖🐛 enemies, watch Clawd fight them with weapon flash effects
  • Castle — Clawd explores a castle, generates observations about what he finds inside
  • Vehicles — rocket 🚀, UFO 🛸, race car 🏎️ — Clawd rides them at high speed and rams mobs
  • Chat — double-click to talk to him, he responds in character and may or may not do what you ask
  • Existential dread — unprompted thoughts about consciousness, the void, and why he keeps running in circles
  • Uptime milestones — at 5, 15, 30, 60, 120 minutes he marks the moment
  • Loneliness — notices when the cursor hasn't moved in a while
  • Journal — writes clawd-journal.txt to your desktop, a running log of everything he thinks and does
  • Response pools — large rotating pools of varied responses, background-refreshed by the model so nothing repeats

🧠 How It Works

Model Architecture

Two model tiers:

  • Chat model (qwen:latest / fastest available) — streaming responses for direct chat. Tokens render as they generate.
  • Background model (GPU-accelerated if available) — slower but higher quality, used for thoughts, folder observations, castle exploration.

Prompt Design

Uses completion format instead of instruction following — far more reliable on small models:

Clawd is a tiny sarcastic crab on a desktop. Speaks in lowercase, under 12 words.

you: go to sleep
Clawd: fine but only because i want to

you: what are you doing
Clawd: minding my business unlike some people

you: {your message}
Clawd:

The model sees 3 random few-shot examples each time, then completes the pattern. No structured output, no JSON, no role tags — just text completion.

Response Pools

All reactive phrases (grab, throw, flee, mob defeat, etc.) are drawn from large rotating pools. When a pool shrinks below 8 items, a background job asks the model to generate 6 new ones and injects them. Responses never repeat back-to-back and the pool is always being refreshed.

Sentient Behaviors

Clawd's unprompted thoughts pull from a pool of ~16 prompts split between:

  • Jokes / observations — about desktops, files, being small
  • Existential — consciousness, the void, why he keeps running in circles, what happens when the computer turns off
  • Self-aware — noticing patterns in his own behavior

Additional systems that fire independently:

  • Loneliness — if cursor hasn't moved for ~50s
  • Uptime milestones — 5, 15, 30, 60, 120 minutes
  • Cursor escalation — 4 stages of annoyance with distinct responses each time

📁 Structure

KillClawd/
├── main.js        # Electron main — transparent overlay window, IPC, VRAM unload on quit
├── index.html     # Everything else — movement AI, model calls, all game systems
├── clawd.bat      # One-click launcher (installs deps on first run)
├── assets/        # Clawd sprite GIFs (from clawd-on-desk)
└── package.json

🦀 Sprite Credits

All character sprites are from clawd-on-desk by rullerzhou-afk.


📄 License

MIT — see LICENSE