HN Bender 16d ago

AI customers are coming around to the idea that small is beautiful

Analysis of market shift toward smaller, task-specific AI models over large general-purpose models like GPT.

HN plushywushy 16d ago

Theia, a fact-checker that flags sycophancy from ChatGPT/Claude/Gemini

Fact-checker browser extension for ChatGPT, Claude, Gemini that detects hallucinations and sycophancy using grounded sources and on-device checks.

HN alindnbrg 16d ago

AI's memory. On your machine, under your control

Local agentic runtime with persistent, user-controlled memory for AI agents; all data stored locally on disk.

HN matt_d 16d ago

AI Model Co-Design: Hardware-Friendly LLM Design

Technical analysis of hardware-LLM co-design trade-offs optimizing for accuracy, throughput, and latency simultaneously.

HN pedrodduarte 16d ago

Show HN: Kote – Capture and reuse engineering context from AI chats and Git

Kote: VS Code tool capturing AI chat context from Claude, Codex, OpenCode and Git for engineering knowledge reuse.

HN KZ_AI 16d ago

Dari AI – An all-in-one AI assistant for macOS that runs on-device

Dari AI: on-device macOS AI assistant with multimodal model, offline operation, no subscription. Local inference tool.

HN therepanic 16d ago

I love LLMs, I hate hype

Personal blog post about enthusiasm for LLMs, coding agents, and local AI setup. Incomplete/corrupted text.

HN amarble 16d ago

LLMs are still just low code / no code software

Opinion piece comparing LLM applicability to low-code programming, arguing both hit productivity plateaus with inherent limitations.

HN systima 16d ago

Claude Code sends 33k tokens before reading the prompt; OpenCode sends 7k

Empirical analysis comparing token usage: Claude Code sends 33k tokens vs OpenCode 7k before reading prompts. Optimization study.

HN adunk 16d ago

Mechanistic interpretability researchers applying causality theory to LLMs

Research applying causality theory to mechanistic interpretability of LLMs via arxiv paper.

HN adithyaharish 16d ago

Show HN: Block dangerous Git and shell commands from being executed by agents

Pre-execution hook blocking destructive shell/git commands in AI coding agents including Claude Code, Cursor, and Copilot.

HN speckx 16d ago

W11 Copilot tells you what's slowing down your PC, while using 1GB RAM itself

Windows 11 Copilot PC Insights feature analyzes system resource usage; Copilot itself consumes 1GB RAM. Performance critique.

HN innovatorved 16d ago

Show HN: SayItDev LLM and Speech capabilities with 0 dependencies and no models

macOS CLI leveraging Apple Intelligence for local speech, transcription, and chat with OpenAI-compatible API.

HN WispAIOverlay 16d ago

Show HN: Wisp – open-source private desktop AI overlay with MCP support

Open-source desktop AI overlay with MCP support enabling in-context prompting across applications with voice input and local execution options.

HN aethelyon 16d ago

GPT-5.6, Fable 5, and Grok 4.5 rebuild Basecamp from the same spec

Comparative evaluation of GPT-5.6, Fable 5, and Grok 4.5 models across 20 dimensions using a complex software problem as test case.

HN gnabgib 16d ago

Using AI to Let History Speak About Bank Runs

LLM-based system extracting bank run data from digitized historical newspapers to create comprehensive empirical dataset for banking crisis research.

HN lukefairbanks 16d ago

Show HN: Broll – an MCP server that gives coding agents a content studio

Broll is an MCP server enabling AI coding agents to generate, render, and publish media content with deterministic code-driven workflows and draft confirmation.

HN tanglearncode 16d ago

Show HN: A lightweight app to let LLM work for oncall

Desktop application providing domain knowledge context to LLM-based SRE agents for accurate incident handling.

HN arm32 16d ago

Honeyprompt: LLM-first honeypot that talks back

LLM-first honeypot deception framework supporting SSH, HTTP, TLS with multiple LLM providers as single binary.

HN amandafranc 16d ago

I connected my net worth tracker to Claude via MCP

Example of integrating net worth tracker with Claude using Model Context Protocol for AI-assisted analysis.

HN casbattle 16d ago

Show HN: AgentIndexed – a curated AI agent directory, no pay-to-play rankings

Directory/registry for discovering AI agents without pay-to-play model. Curation-focused resource for finding available agents.

HN yonidavidson 16d ago

Show HN: I gave my AI coding agents a group chat (it's just a Git repo)

Developer shows AI coding agents collaborating via Git repo as communication layer, demonstrating practical multi-agent coordination.

HN abhijeet_gupte 16d ago

Concerned about how companies will manage their cloud bill once agents dominate

Discussion of cost management challenges as autonomous AI agents proliferate and increase cloud compute bills.

HN vtoolpro 16d ago

Show HN: AI Photo Editor – Professional-Grade Image Editing with Text Prompts

AI-powered image editor enabling professional editing via natural language text prompts.

HN gkiss0001 16d ago

Show HN: Itara – Distributed system topology as an explicit, executable layer

Itara is a tool for managing distributed system topology as explicit executable layer, simplifying service boundary and protocol changes in complex architectures.

HN Saya24f 16d ago

Linux of AI open-source tools for reducing AI vendor lock-in

Linux of AI initiative providing open-source tools and infrastructure for building AI systems without vendor lock-in, offering inspectable and portable AI alternatives.

HN theletterf 16d ago

You can write docs in LLMese, but you don't have to

Exploration of machine-readable documentation formats for inter-agent communication, questioning whether LLMs need human-like rhetoric or terser token-efficient formats.

HN atmask 16d ago

Show HN: Containix – Run Nix flakes as Kubernetes pods without images

Containerd shim enabling direct execution of Nix flakes as Kubernetes pods without OCI image building.

HN ffwd 16d ago

LLMs might not learn concepts

Discussion post questioning whether LLMs learn concepts. Minimal content provided.

HN atomicnature 16d ago

Perfectly Hitting the Wrong Target: The Story of an AI Code Review Benchmark

Critical analysis of an AI code review benchmark, examining flaws in benchmark design and cautioning against blind reliance on seemingly authoritative metrics.

HN gmays 16d ago

Autoresearch, Claude and Constrained Optimization

Application of Claude LLM to autoresearch and constrained optimization problems.

HN Abdalsalaam 16d ago

Show HN: Agent Legibility Analyzer see if AI shopping agents can read your store

AgentMint research platform analyzing how AI shopping agents read and select products from stores, with controlled simulation results on product placement effects.

HN Aherontas 16d ago

I Use Multiple AI Agents in My Team's SDLC to Ship Production Features

Case study of using multiple AI agents in software development lifecycle to ship production features, covering agentic workflows and orchestration.

HN thekiraproject 16d ago

Your AI's logs can be edited. Its evidence shouldn't be

Tamper-evident two-layer evidence architecture for local AI pipelines with immutable logging and observability beyond standard span traces.

HN franksoftware 17d ago

OpenAvocado: Open-Source Adaptive Learning

Open-source adaptive learning system.

HN quantiota 17d ago

Multi-agent network: Own the node. Let the agents build together

Multi-agent network framework for collaborative agent development with node ownership model.

HN kordlessagain 17d ago

Litert.js, Google's High Performance Web AI Inference

LiteRT.js brings Google's on-device ML inference library to JavaScript for browser-based AI models with privacy and low latency.

HN Brajeshwar 17d ago

Microsoft joins Google in backing Go for AI agents — OpenAI and Anthropic lag

Go language adoption for AI agents backed by Microsoft and Google, contrasting with OpenAI/Anthropic focus on Python.

HN sshwarts 17d ago

Show HN: Skillscript – A declarative, sandboxed language for tool orchestration

Declarative DSL for defining reproducible agent workflows; enables version control and reduces token drift in local AI agents.

HN zaikunzhang 17d ago

AI Boosts Research Careers but Flattens Scientific Discovery

Analysis of 40M academic papers shows AI tool users publish more but research narrows to crowded areas, reducing scientific diversity.

HN gabe_karina 17d ago

The Amazon SES Alternative for AI Agents – MailKite

MailKite email service designed for AI agents with parsed JSON inbound messages and autonomous reply loops, alternative to Amazon SES.

HN Brajeshwar 17d ago

Do LLMs make code a "materialized view"?

Analysis of how LLMs change code's role from source of truth to materialized view, exploring implications for software development workflows.

HN sgury 17d ago

Show HN: LocalMask – local-first CLI that masks PII and secrets before LLMs

LocalMask is a local-first CLI tool that masks PII and secrets before sending data to LLMs, addressing enterprise data governance concerns.

HN caveman23 17d ago

Ollama vs. Llama.cpp – Quick Benchmark

Benchmark comparison of Ollama vs Llama.cpp for LLM serving on Tesla V100, measuring overhead and performance differences.

HN Brajeshwar 17d ago

The impressive AI demo is dead. Here's what actually reaches production

Discussion of AI demo limitations and what actually ships to production, engineering perspective on gap between demos and real systems.

HN MerlijnW70 17d ago

Mnema: A local, encrypted memory layer for AI agents

Mnema: local encrypted memory storage layer designed for AI agents, enabling persistent context.

HN joozio 17d ago

Scientists' Side Hustle? Using AI and Quantum Computing to Generate New Peptides

Quantum computing combined with generative AI for protein/peptide discovery by Technical University of Denmark and ORCA Computing.

HN siroj 17d ago

AI Agents Are About to Change Payments Operations

Headline about AI agents transforming payment operations without substantive details.

HN beardyw 17d ago

Chasing new skills, going back to basics: how software engineers adapting to AI

Profile of software engineer adapting skills amid AI-driven job market disruption.

HN clodex_lab 17d ago

Clodex – Open-source agentic IDE with governed execution and graph memory

Open-source agentic IDE combining tasks, code, terminal, browser, Git. Model output as untrusted input with governed execution.