HN royapakzad 4d ago

From Evaluation to Guardrails: What We Brought to ACM FAccT 2026

Discussion of evaluating AI guardrails contextually across languages and agentic systems, presented at ACM FAccT 2026.

HN firasd 4d ago

Accidental data loss in Claude Code and OpenAI Codex: when AI deletes user files

Report on AI coding agents (Claude Code, Codex) accidentally deleting user files due to $HOME misconfiguration and sandboxing gaps.

HN SHACKLE-PRO- 4d ago

Fail-closed enforcement for AI agents: deny what you cannot verify

Discussion of fail-closed enforcement policies for AI agents that deny unverifiable actions.

HN ai2027 4d ago

MAI-Image-2.5-Pro and MAI-Voice-2-Flash

Microsoft announces MAI-Image-2.5-Pro and MAI-Voice-2-Flash models now powering Bing, PowerPoint, and other products.

HN pixelpoet 4d ago

AMD's Instinct MI455X: Aiming for the Sun

Technical overview of AMD's Instinct MI455X GPU, CDNA5 architecture, and Helios rack-scale AI system.

HN nlpnerd 4d ago

You don't need an LLM to cluster LLM traces

Technical post on clustering LLM traces using deterministic features and DBSCAN to compile repetitive LLM calls into cheaper pipelines.

HN rgbrgb 4d ago

Show HN: Setoku – Self-hosted knowledge server for AI agents

Self-hosted MCP knowledge server combining a ClickHouse data lake with a knowledge layer for AI agents.

HN sand0vvv 4d ago

Show HN: Fleet – drive a fleet of Claude Code/Codex agents from Telegram

Tool to control a fleet of Claude Code/Codex coding agents remotely via Telegram.

HN ecrasevvv 4d ago

Show HN: Flow Matching model inference in C

Educational C implementation of Flow Matching generative model inference for 2D distributions.

HN louis030195 4d ago

Launch HN: Screenpipe (YC S26) – Power your agents by your 24/7 screen recording

Launch of Screenpipe, a local screen/audio recorder giving AI agents searchable memory to automate repetitive tasks.

HN rizzdev 4d ago

A Claude skill that grills you so hard. Within a nice clean interface

Claude Code skill that batches multiple-choice questions into a local web form and returns answers as JSON.

HN ulrischa 4d ago

Ayghri/I-have-ADHD: skill for coding agent to stop it from burying the answer

A Claude Code skill/output style that enforces concise, action-first responses instead of verbose hedging.

HN aksworbuz 4d ago

Show HN: Remortgaging with Claude, so I built a MCP for Bank of England

MCP server wrapping the Bank of England's public statistical database to query base rates and MPC meeting dates via Claude.

HN conradqh 4d ago

Free way to get your data out of ChatGPT Business accounts

Tool that exports ChatGPT Business/Team conversation history locally into a searchable offline HTML archive.

HN rascal999 4d ago

Docs as the source of truth for coding agents

Starlight-based documentation scaffold template for repos, intended as a source-of-truth doc site for coding agents.

HN quantizor 4d ago

Devctl, a handy way to manage many dev servers

macOS menu bar app and daemon that manages local dev servers so coding agents keep track of running processes across sessions.

HN apwheele 4d ago

Notes on Document Processing with LLMs

Blog post on cost-efficient document processing with LLMs, comparing cheap models, OCR pipelines, and open source options for large-scale extraction.

HN SteveVitali 4d ago

Show HN: Hibernate and restore Claude Code sessions across reboots

CLI tool to hibernate and restore running Claude Code sessions across machine reboots, preserving session ids and directories.

HN dhorthy 4d ago

Why Software Factories Fail

Essay on why AI coding agent 'software factories' fail, arguing harness engineering alone is insufficient.

HN azuroscloudapi 4d ago

Show HN: Find undervalued bikes near you with ML

Report on LLMs producing confused responses when presented a counterargument to the Jacobian Conjecture.

HN hereme888 4d ago

Show HN: STT-MCP – local STT for agents

Local speech-to-text MCP server letting AI agents process audio/video using Parakeet or Granite backends.

HN saheb37 4d ago

Lessons from my RL experiments with snake

Personal account of RL experiments training agents to play Snake, covering Q-learning, PPO, curriculum learning, and CNN generalization to larger boards.

LB byteshape.com via brechtm 4d ago

Beyond a Single Number: Evaluating Quantized Models for Deployment

Discussion on evaluating quantized models for deployment beyond single benchmark numbers.

HN rgbrgb 4d ago

Show HN: Setoku - self-hosted knowledge server for AI agents

Blog post on tools and cost strategies for large-scale document processing with LLMs, including OCR and cheap models.

HN josh-paul 4d ago

Engineering the Channel: Restoring Software Engineering Discipline for LLMs

Paper proposing 'channel engineering' to improve reliability of human-LLM communication in software systems.

HN mythz 4d ago

OSS ChatGPT WebUI v4 – Projects, Profiles, Server Tools, Themes, 1-Click Sharing

Self-hosted open-source ChatGPT-like web UI gateway supporting 530+ models, images, audio, agents and MCP.

HN KRISH2832 4d ago

Show HN: Free API keys for open models (Llama, Qwen, GPT-OSS, Gemma)

Free API access to open models like Llama, Qwen, GPT-OSS, and Gemma via an OpenAI-compatible endpoint.

HN mat6661 4d ago

I asked 6 LLMs to design 100 software systems

Study comparing architecture choices made by six LLM families across 100 software system design briefs.

HN war851 4d ago

Show HN: Governed Agent – the LLM reads the request, rules and code decide

Agent architecture where an LLM only extracts structured fields while deterministic code and tables make all decisions, with tamper-evident logs.

HN AsapdaX 4d ago

Tiny inference engine in pure Rust, CPU-only, with TUI visualization

Pure-Rust, CPU-only LLaMA-style inference engine for MiniCPM5-1B with TUI visualization, built for learning.

HN umudhasanli 4d ago

Flare Redact: Stop secrets from reaching logs, telemetry, and LLMs

Open-source library that redacts secrets and PII from logs, prompts, and LLM calls before they leak.

HN robbiet480 7d ago

Ramp AI Router

Ramp's AI model router product page, routing 100+ internal use cases to different LLMs to cut costs 30%.

HN lucianBKR 7d ago

Repos in 29 days with an AI pipeline – what broke wasn't code

Retrospective on building 26 repos/335 pages in 29 days using Claude Code, analyzing failure modes in an AI coding pipeline.

HN ThomasWink1981 7d ago

Free LLM balancer combines multiple local inference machines with cloud fallback

Self-hosted LLM router combining local inference machines with cloud fallback behind one OpenAI-compatible API.

HN DalasNoin 7d ago

AI Voice Phishing Performs on Par with Human Scammers at a Fraction of the Cost

Large human-subject study (n=4,100) measuring susceptibility to AI-generated voice phishing across six voice models.

HN kindaAnIdiot 7d ago

Does "rtk" skill cut agent tokens by 60–90%? We tested it

A/B benchmark testing whether the 'rtk' tool reduces Claude Code token usage; finds no real savings and sometimes higher cost.

HN argo_cve 7d ago

D-FINE-seg – detection, instance and semantic segmentation in one model

Open framework combining real-time object detection, instance and semantic segmentation in one model with five sizes.

HN Nolan_Lwin 7d ago

Show HN: Forall – OpenSpec but with actual machine-checkable verification

Coding agent CLI that generates spec-driven code alongside machine-checkable proofs, integrates with editors via MCP.

HN rcarmo 7d ago

Show HN: Memento, Shared Agentic Memory

Shared agentic memory system storing knowledge across multiple AI agents via MCP with git-backed markdown and review workflow.

HN rfirke 7d ago

Show HN: Provena: Open-Source Library for AI Agent Context Governance

Open-source library aiming to govern context data flowing into AI agent context windows.

HN KanishkNoir 7d ago

Show HN: Cognikernel- Local Memory for AI Coding Assistants

Hobby project adding local persistent memory to Claude Code/Codex sessions, claiming reduced token usage.

HN kendallgclark 7d ago

Knowing, Remembering, Exactly, Vaguely: An Agent-Native Database (PlatypusDB)

Design note on PlatypusDB, an agent-native bitemporal event database combining crisp Datalog queries with fuzzy VSA resonance.

HN logickkk1 7d ago

Cue AI

Desktop voice AI agent using locally-run Gemma model for dictation, cutting latency 44% versus cloud processing.

HN ikaruscareer 7d ago

Using GPT Codex, DeepSeek V4 and Kimi K3 on a Real OSS Project

Open-source static analysis tool scanning AI agent source code offline for security risks and governance gaps, built using GPT Codex, DeepSeek, and Kimi.

HN matt_d 7d ago

Every Microsecond Matters:Achieving Near SpeedOfLight Latency in GPU Collectives

arXiv paper abstract on achieving near speed-of-light latency in GPU collective communication operations.

HN dakshjainn 7d ago

Cross platform companion for multi agent

A control center app that discovers and manages running coding agent sessions (Claude Code, Codex) across terminals, with mobile/LAN access.

HN DanyWin 12d ago

Show HN: An MCP to create and call Playwright scripts as tools on your extension

Model Context Protocol (MCP) implementation allowing agents to create and execute Playwright scripts as callable tools.

HN bitjaru0402 12d ago

Show HN: StyleSeed – a design-rules engine so AI agents stop building generic UI

Design rules engine enabling AI agents to generate non-generic UI by enforcing design consistency patterns.

HN rrrpro123 12d ago

NovaForge AI – Free AI, PDF, Image and Developer Tools in One Place

All-in-one platform offering free access to AI models, PDF tools, and image processing. Limited technical depth.

HN tom-villani 12d ago

All2md: PDF, Docx, PPTX, HTML, Email and 40 More to Markdown – Python+CLI+MCP

all2md: Universal document converter (PDF, DOCX, HTML, 40+ formats) to Markdown with MCP integration for AI assistants. Python library and CLI tool.