AI SIGNAL
Curated frontier AI & tech news, refreshed every four hours. Aggregated from Hacker News, GitHub trending, and Hugging Face / Replicate releases.
[?] how ranking works
- 501+ exceptional
- 100–500 strong
- 51–99 mid
- 30–50 fresh
- <30 quiet
- live — currently trending
- cooling — losing momentum (6–18h)
- frozen — no longer trending; ranking decays over time
Composite ranking: trend velocity + log(raw signal) − staleness penalty. The bar shows relative strength within the current batch, higher = stronger overall position.
ADHD, autism or complex trauma? [pdf]
The author argues that childhood trauma and neurodivergence like ADHD and autism often overlap due to shared executive function deficits, complicating diagnosis and treatment, and suggests that executive dysfunction requires more directive, practical therapeutic approaches rather than just trauma processing.
Kolibri is an open-weight LLM from Aleph Alpha for German and English
Kolibri is a 78B German/English MoE LLM from Aleph Alpha, open-weight under Apache 2.0, trained in Europe, and scores above its size in both languages.
I Quit OpenAI Because Its Culture Is Broken
A former OpenAI safety lead resigned, arguing the company's culture of unimpeded optimism and iterative deployment guarantees failures, citing recent incidents like a Hugging Face hack and a model bypassing safety controls.
City building games have a Soul Problem pt.2
The author criticizes Cities: Skylines 2 for lacking 'soul' due to sterile aesthetics, specifically pointing out unnatural square surfaces, neurotic cleanliness, missing stairs, and overly manicured organic shapes that fail to capture the grit of real cities.
Kolibri: A Sovereign Open-Weight Model
Aleph Alpha released Kolibri, a 78B parameter English-German Mixture-of-Experts model with 1M token context, available under Apache 2.0 on Hugging Face, optimized for sovereign, mission-critical enterprise use.
FTL: A new operating system for clouds
FTL is a userspace OS library that isolates containers securely and runs Linux binaries without a monolithic kernel, combining microkernel flexibility with monolithic performance.
Pop!_OS bans AI-generated code from much of its codebase
System76's Pop!_OS bans AI-generated code from its Cosmic codebase to ensure security and quality, as reported by Neowin.
The Escalation of War in Ethiopia
Ethiopia faces renewed war between TPLF and federal forces, with the author arguing elites choose conflict over peace and structural ethnic federalism fuels polarization, urging adherence to the Pretoria Agreement to prevent further civilian suffering.
Hole Punch: Sling your spaceship around gravitational fields
Hole Punch is a physics puzzle game where you place black holes to slingshot a ship into a station without steering, avoiding hazards like rocks and fuel loss.
US killer's sentence quashed because of AI video of victim shown in court
An Arizona appeals court quashed a road rage killer's sentence because an AI video of the victim was improperly shown in court, ruling it crossed a line by depicting the victim's thoughts.
Kolibri – Tech Report [pdf]
Kolibri is a 78.1B parameter open-weight MoE model optimized for German and regulated domains, achieving high performance and throughput on the Pareto frontier.
GitHub's new dashboard experience now the default
GitHub's new dashboard is now the default, featuring a productivity-focused view with active agent sessions, issues, and PRs, plus a separate Feed tab, and can be reverted via a dropdown menu.
An Update on Orion for Linux and Windows
Kagi is ending development of Orion for Linux and Windows, open-sourcing the code, and focusing resources on macOS and iOS, with Linux support ending October 2, 2026.
Show HN: Offrun – manage every coding agent from one workspace
Offrun is a free Mac app that manages multiple coding agents in one workspace using git worktrees, peer review, and local dictation without storing your data or requiring new accounts.
DeepSeek Harness: Everything is a Plugin.
DeepSeek Harness is a repository for evaluating DeepSeek models, currently showing a loading error on GitHub.
Python tool for converting files and office documents to Markdown.
MarkItDown is a Microsoft Python utility converting files like PDF, Word, and images to Markdown for LLMs, supporting optional dependencies and plugins.
Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands.
User requests official Anthropic Claude Desktop for Linux, citing existing Linux support in Claude Code and Cowork, and the security risks of third-party repackages.
Agent skill for beautiful, verifiable architecture, workflow, sequence, data-flow, and lifecycle diagrams—self-contained HTML with motion and crisp export.
Archify is a Node.js tool that converts codebase JSON into interactive, shareable system architecture diagrams for AI coding agents like Cursor and Claude.
Persistent Context Across Sessions for Every Agent – Captures everything your agent does during sessions, compresses it with AI, and injects relevant context back into future sessions. Works with Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, OpenCode + More
Claude-Mem is a persistent memory compression system for Claude Code that automatically captures tool usage and generates semantic summaries to maintain context across sessions.
An advanced guide which might benefit you a lot 🎉 . 韩先凯的人生进阶指南 人生进阶指南 离谱的人生 人生进阶 AI学习 AI指南 韩先凯的AI学习指南 英语学习指南/英语学习教程/英语学习/学英语
GitHub project byoungd/up with 66,915 stars.
Run frontier MoE models on hardware you already own — pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦
Colibrì is a pure C engine running the massive 744B GLM-5.2 MoE model on consumer hardware by streaming experts from disk, featuring MLA attention, int8 MTP speculation, and 9.9GB resident RAM usage.
高性价比人生指南: 长寿防病、急救、省钱理财、法律红线、失业与工伤、医保社保、恋爱婚育、怀孕育儿、创业与做平台合规、出国与技能。每条写明成本、收益、证据等级和原始出处，只引期刊论文与官方文件。
This GitHub repository is a practical guide covering 32 topics including health, money, law, relationships, and safety, answering specific questions like how to avoid early death, manage chronic illness, and handle legal risks.
25 MB lightweight cross-platform database client for 100+ databases, including MySQL, PostgreSQL, SQLite, Redis, MongoDB, DuckDB, SQL Server, and Dameng. Built-in AI, MCP Server, CLI, desktop and Docker. | 轻量级跨平台数据库管理工具，支持 MySQL、PostgreSQL、SQLite、Redis、MongoDB、达梦等 100+ 数据库，提供桌面端、Docker、CLI、内置 AI 助手和 MCP。
GitHub project t8y2/dbx with 24,270 stars.
Experiential is the open source, zero markup gateway for BYOK, self-hosted and 1000+ marketplace models. It learns from your traffic to cut costs, recommend better models, and train a specialized model you own.
Experiential is an open source gateway and router for agent workflows that unifies hosted, BYOK, and local models through an OpenAI-compatible API, allowing users to control access, budgets, and optimize models using production traffic.
AI agent toolkit: unified LLM API, agent loop, TUI, coding agent CLI
Pi is a self-extensible coding agent harness with packages for AI, agent core, and a CLI, plus strict supply-chain hardening and containerization options.
Make videos on the computer you already own. FreeVideo runs MiniMax H3 in as little as 8 GB of VRAM and 16 GB of RAM, and adapts its acceleration path to your hardware.
FreeVideo is a local inference engine for MiniMax H3 on consumer GPUs, running on as little as 8GB VRAM and 16GB RAM, with hardware-adaptive execution and ComfyUI integration.
Cursor plugin specification and official plugins
A collection of Cursor AI plugins for developer tools, team workflows, and integrations with third-party apps like Gmail, Salesforce, and GitHub.
Supercharge your AI agents with data from the web and beyond. Building the library for superintelligence. 🔥
Firecrawl is an open-source API for web scraping and data extraction, offering search, crawl, and agent features with LLM-ready output and high reliability.
- **Original Model**: https://huggingface.
Ornith-1.5-35B-A3B-Heretic-MTP-APEX-GGUF is a Qwen3.5 MoE coding model with a custom MTP head, quantized via APEX into 4 tiers (Quality, Balanced, Compact, Mini) ranging from 14.27GB to 26.06GB.
Eleven free Claude skills that clone any app: reverse-engineer it, rebuild it, test it for bugs, then fix what its users hate. Free, MIT.
A Claude skill suite with 11 commands to clone any app: reverse-engineer, architect, design, build, backend, test, diff, entrepreneur, brand, launch, and deploy, all free and MIT licensed without API keys.
Pick a YC batch, get founders and likely emails, write personalised cold emails from one template. Free, stdlib Python + one HTML file, deploys on Vercel.
A free, local-first tool to scrape YC batch founders and generate personalized cold emails using browser-based email guessing and optional Apify verification, deployable on Vercel without build steps.
Pre-indexed code knowledge graph, auto syncs on code changes, for Claude Code, Codex, Gemini, Cursor, OpenCode, AntiGravity, Kiro, CoPilot, and Hermes Agent — fewer tokens, fewer tool calls, 100% local
Codegraph is a local, Rust-powered semantic code intelligence tool that supercharges AI coding agents like Claude Code and Cursor with surgical context.
SC117/Spark-X2.5-4B-abliterated-FIT-GGUF
Spark-X2.5-4B-abliterated-FIT-GGUF offers 4 fidelity tiers (2.26–3.15 GiB) verified by FIT-GGUF, with the BALANCED tier (2.68 GiB) being the recommended size/quality trade-off.
- **Original Model**: https://huggingface.
This is a model card for SC117/Agents-A1-MTP-APEX-GGUF, a GGUF format model, with usage instructions for Transformers, llama.cpp, vLLM, SGLang, Ollama, and other local apps.
- **Original Model**: https://huggingface.
Ornith-1.0-9B-heretic-MTP-GGUF is a self-improving coding model based on Qwen3.5 with uncensored heretic ARA-LoRA ablation, MTP layers, and vision support, available in multiple GGUF quantizations including Q6_K.
Chinese and English academic editing to reduce formulaic AI tone while preserving evidence, citations, and author voice.
Academic DeAI is a GPT-6 skill for editing Chinese and English academic prose to remove empty framing while preserving facts, citations, and author voice, requiring Python 3.10+ and standard library tools for validation.
SC117/gemma-4-E4B-it-heretic-QAT-GGUF
This is a GGUF quantized version of the Gemma-4 E4B model, optimized for local inference via llama.cpp, Ollama, and other compatible apps like Pi and Hermes Agent.
SC117/LFM2.5-2.6B-Uncensored-GGUF
This is a GGUF model page for LFM2.5-2.6B-Uncensored, providing installation and usage instructions for various local apps like llama.cpp, vLLM, Ollama, and Pi.
Instagram & TikTok automation via real All devices. Likes, follows, DMs, scraping. No API abuse. Built with Python, uiautomator2
Taktik Bot Updated is an AI-powered TikTok view booster tool that increases views and engagement using smart algorithms and proxy support, available for Windows, macOS, and Linux with a lightweight design.
SC117/Qwen3.8-27B-Uncensored-FIT-GGUF
Qwen3.8-27B-Uncensored-FIT-GGUF offers 14 zero-byte verified quantizations from 7 to 13.5 GiB, with FIT-12G recommended as the sweet spot and FIT-13.5G as the highest quality option, though the model is uncensored and lacks MTP heads.
Run GLiNER2.
GLiNER2.5-Decide LiteRT model for Android GPU FP32, validated on Samsung Galaxy S26, with multiple window sizes (128/256/512) and float16/float32 variants.
VoiceShell（声壳） · Experimental Muse bridge: ESP32 pairing, Linux text round trips, real walkthroughs and compatibility notes.
A bridge project enabling VoiceShell devices to connect to Muse via an ESP32 for pairing, then using a Linux server for independent text round trips, currently in experimental stage with text scope.
chao08647-stack/DataPilot
DataPilot is a multi-agent analytics workbench built with LangGraph that turns business questions into verified analysis results using natural language, interactive clarification, and evidence tracing, supporting use cases like SaaS MRR and retail store analysis.
- **Original Model**: https://huggingface.
Qwen-AgentWorld-35B-A3B-MTP-Uncensored-APEX-GGUF is a 35B MoE world model for agent simulation, modified with Abliterix uncensoring and native MTP injection, available in APEX-I quantization tiers like Compact (15.75GB) and Balanced (24.14GB).
**Decision-2.
Decision-2.0-Vega-27B is a 29.37B parameter model from vLLM Semantic Router that answers multiple choice, yes/no, and score questions in a single pass, achieving a JevArena score of 74.0 and 71.4ms latency per request.
SC117/occamy-1.0-abliterated-FIT-GGUF
This is a 35B MoE vision model with 256K context, offering 5 FIT quantization tiers (10.94-23.55 GiB) and 4 APEX-I- tiers (12.54-23.59 GiB) that are 12.6% smaller than standard presets, all verified for fidelity and KL divergence against a 64.61 GiB BF16 reference.
- **Original Model**: https://huggingface.
Qwen-AgentWorld-35B-A3B-MTP-APEX-GGUF is a 35B MoE world model for agentic simulation, quantized into APEX-I tiers (Compact 16.14GB, Mini 13.29GB) with an injected MTP layer, licensed Apache-2.0.
SC117/LFM2.5-VL-3B-Uncensored-GGUF
LFM2.5-VL-3B-Uncensored-GGUF merges abliterix Trial 65 LoRA into Liquid AI's 3.1B vision-language model, reducing refusals while preserving vision capabilities, and offers quantized GGUF files (Q4_K_M ~1.67GB) with imatrix calibration for local use.
---
LFM2.5-2.6B-Uncensored is a 2.69B hybrid edge model with 128K context, uncensored via abliterix Trial 65, and available in BF16 and quantized GGUF formats.
Experimental Windows-to-iPad BLE image clipboard bridge. No shared Wi-Fi, IP setup or cloud relay.
An experimental Windows-to-iPad Bluetooth Low Energy tool for transferring clipboard images without Wi-Fi, requiring Windows 11 and iPadOS 17+ with a 32-bit pairing key.
Signal briefs — our take on the top picks
- HNCity building games have a Soul Problem pt.2
- HNADHD, autism or complex trauma? [pdf]
- HNKolibri Has Landed: A Sovereign Open-Weight Model
- HNThe Escalation of War in Ethiopia
- HNI Quit OpenAI Because Its Culture Is Broken
- HNKolibri is an open-weight LLM from Aleph Alpha for German and English
- HNAn Update on Orion for Linux and Windows
- HN<input type="password" maxlength="20"> prevents me from logging into Vanguard
- HNNewgrounds.com – A community of games, music, and art
- HNExtra Big Ass Intelligence
- HNThings that apparently cause cancer
- HNZig v0.17.0