HN pallakatos 25d ago

Kars – treat every AI agent as untrusted code, on Kubernetes

Kars: Kubernetes tool treating AI agents as untrusted code with security sandboxing.

HN 3littlefish 25d ago

Meko Is Open for Self Sign-Up

Meko is a memory system for AI agents that maintains persistent context across different coding tools and models.

HN lpratheesh 25d ago

Context-clipper – a priority min-heap for LLM context windows, zero deps

Context-clipper: zero-dependency priority min-heap data structure for managing LLM context window allocation.

HN akiarie 25d ago

Opinion: Agents > Unix Shell

Opinion piece arguing AI agents are more useful as shell replacements than as code programmers, requiring less auditing overhead.

HN CroviaTrust 25d ago

Show HN: Causari – Content-addressable ledger for AI agent code edits

Curated infrastructure resource collection for production AI agents covering runtimes, security, observability, and evaluation tools.

HN gmays 25d ago

Learning to Replicate Expert Judgment in Financial Tasks

ML research on replicating expert judgment in finance using machine learning techniques.

HN LabsLucas 25d ago

AMD Ryzen AI Halo – $4k AI Dev Kit

AMD Ryzen AI Halo $4k dev kit with Zen 5 processor, 128GB RAM, NPU for streamlined AI development with ROCm.

HN BrKeaton 25d ago

Self-Service Ransomware as Security Against Local AI Tools

Report summary on AI engineering trends and acceleration patterns in 2026.

HN anirudhak47 25d ago

Show HN: AI harness for C/C++ with GDB, sanitizers, perf and compile tools

File encryption tool as security mechanism against local AI agent threats using Supabase, Next.js and Tauri.

HN shenli3514 25d ago

Agent Infra: curated resources for production AI agent infrastructure

C/C++ coding harness for AI models with GDB, sanitizers, perf integration supporting multiple LLM providers.

HN rchaz 25d ago

Solo – scaffold a production SaaS, then run its operations autonomously

Framework to autonomously scaffold and operate SaaS applications using AI agents.

HN adletbalzhanov 25d ago

Does Speaking to Agents Like Cavemen Save 65% of Tokens? We Test

Experimental test comparing token efficiency of simplified prompting (caveman speak) vs standard prompting for AI agents.

HN kolesnikov-arch 25d ago

Show HN: I pre-registered the rules,then measured AI agents shipping wrong fixes

Controlled benchmark comparing gated vs ungated AI patch generation on SWE-bench Lite with reproducible methodology and scripts.

HN varad-khoriya 25d ago

Show HN: Loopers – Open-source fail-closed firewall for AI agent runtimes

Open-source fail-closed firewall for constraining AI agent runtimes and preventing unauthorized actions.

HN frnic0las 25d ago

A 13-agent pipeline that ships full-stack web apps from a Markdown spec

AI agent pipeline with 13 specialized agents that generates full-stack web applications from Markdown specifications. Multi-agent system for code generation.

HN latand6 25d ago

Show HN: Orchestrate parallel Claude Code and Codex agents on a live map

Tool for orchestrating multiple parallel AI coding agents (Claude, Codex) with visual map interface to track agent status, hierarchy, and progress.

HN AgentCrush 25d ago

Show HN: We track 1,397 AI agents. 43 famous ones (402k stars) have gone dark

Dataset tracking 1,397 AI agents; analysis showing 43 popular agents with 402k stars have become inactive. Original data on AI agent ecosystem health.

HN k1r111 25d ago

We asked AI agents to fix real security bugs. This is what happened

Empirical study of AI agents' ability to fix real security vulnerabilities. Controlled experiment measuring agent effectiveness on actual bugs.

HN super00826 25d ago

BenchLocal – AI-powered local business search

Desktop app for running, comparing, and benchmarking local/remote LLMs with support for AI agent integration and benchmark workflows.

HN ljlolel 25d ago

Show HN: BurstyRouter: local LLM first then burst to end-to-end encrypted LLM

Developer tool routing local LLM queries first, then bursting to encrypted end-to-end LLM services. Open source project with practical privacy design.

HN ilreb 25d ago

Otari: The Open-Source LLM Control Plane

Open-source control plane for managing LLM deployments. Infrastructure tool for LLM operations and orchestration.

HN rruxandra_l 25d ago

Regression to the Mean: on LLMs and the quiet death of the new

Opinion piece arguing LLMs may reduce creative output by returning statistically average results. Conceptual critique without data or evidence.

HN smashini 25d ago

Show HN: Scan your AI agents for dangerous capabilities

Framework-agnostic checkpoint system for AI agents with deny-by-default permissions, role-based access control, and cryptographic audit trails.

HN micstradev 25d ago

Show HN: A strategy game about the AI race where you can't verify alignment

Browser-based strategy game simulating AI race between US and China with hidden alignment state to explore verification challenges.

HN dpc10 25d ago

LLMs: A cause of, and solution to, UI drift

LLMs as tools to reduce UI/design system drift in software engineering. Discusses both problems and solutions using LLM-supported approaches.

HN riyajoshi 25d ago

Whitepaper: The judgment problem in AI-generated API test suites

Technical whitepaper on improving AI-generated API test suite quality through model orchestration, fine-tuned judgment, and adaptive coverage systems.

HN GemsGame 25d ago

Probably the best AI detector tool is ready

AI-generated text detection tool using linguistic pattern analysis. Claims 99.5% accuracy across multiple LLM models.

HN dimitrisflwr 25d ago

FlowerBench: Benchmarking AI Agents on Real Enterprise Work

FlowerBench is a benchmark for evaluating AI agents on real, long-horizon enterprise workflows with access to proprietary tools and context.

HN melvinroest 25d ago

Ask HN: Are there good security benchmarks for LLMs?

Discussion asking about security benchmarks for LLMs, mentions eyeballvull and need for agent-based scanning benchmarks.

HN nova-gaia 25d ago

Show HN: Open-source CLI for SKILL.md registry and parser lookup

Gaia is an open-source registry for verifying AI agent capabilities through public ledger of code-execution runs, license checks, and security audits.

HN kristofers32 25d ago

SvelteChatKit: Provider agnostic AI chat UI for OpenAI, Dify, n8n, and others

SvelteChatKit is a provider-agnostic AI chat UI for SvelteKit supporting OpenAI, Dify, n8n and custom LLM backends via configurable interface.

HN kartik_malik 25d ago

Show HN: Product Hunters liked my idea - an AI journal with memory

AI-powered personal journal application that auto-categorizes, tags, and organizes entries from text, voice, or links with export capability.

HN takakaze 25d ago

Sakana Translate: Sakana Chat Now Supports Translation

Sakana AI adds translation feature to chat service using Namazu model, supporting Japanese, English, and Chinese with edit and QA modes.

HN Athena-maref 25d ago

We audited 100 open-source Agent projects – 73% have permission overreach

Security audit of 100 open-source agent projects finds 73% have permission overreach vulnerabilities. Original research on AI agent security risks.

HN ardatasci 25d ago

Show HN: PAI, a Linux-y personal-assistant for Mac

PAI is a Linux-style personal assistant for Mac that runs LLM in terminal-like interface. Manages calendar, email, and communications autonomously.

HN throwaw12 25d ago

Tell HN: Error: Claude-fable-5 is temporarily unavailable

User reports Claude model unavailability errors during bash tool calls with auto-retry loops. Support issue with Claude models.

HN alsterg 25d ago

Show HN: An always-fresh memory that learns your repo, so agents stop re-reading

Live-Memory is Claude Code MCP plugin providing persistent codebase memory using large-context models to distill repo structure, enabling agents to avoid re-reading code.

HN akashtndn 25d ago

Show HN: Explain this – agent skills to create HTML explainer for topics or code

Agent skills plugin for creating interactive HTML explainers of topics and codebases using coding agents. Open-source skill for Claude Code.

HN ibrahim_abdu777 25d ago

Bounding the Blast Radius: A Survey of Prompt-Injection Defenses for LLM Agents

Survey of prompt-injection defense mechanisms for LLM agents. Research on securing autonomous agent systems against adversarial inputs.

HN _pdp_ 25d ago

What Happens When AI Agents Can Pay Their Own Bills

Research on self-sovereign agents (SSAs): AI systems capable of autonomous economic sustainability and self-replication through improved decision-making and revenue generation.

HN teleforce 25d ago

Generative AI without guardrail can harm learning:Evidence from high school math

Research article on negative effects of unguarded generative AI on high school math learning. Educational impact study with paywalled content.

HN mieubrisse 25d ago

Claude Code deletes conversations after 30 days

Claude Code deletes conversations after 30 days by default. Technical note on data retention settings and potential workaround in settings.json.

HN throwaway_887 25d ago

Ask HN: What is your go-to prompt to prove AI can be wrong?

Hacker News discussion on prompts to demonstrate AI failures and hallucinations. Community crowdsourced examples.

HN harshvkarn 25d ago

Show HN: Kiwi – Run agentic dev loops in the cloud, keep keys on your laptop

Kiwi enables running autonomous developer agent workflows in cloud while keeping secrets local via reverse tunnel. Enterprise-ready secure execution engine for agentic loops.

HN kokakiwi 25d ago

Compressor V2: three compression layers for a 50% LLM agent cost cut

Compressor V2 reduces LLM agent inference costs by 50% through three compression layers, addressing economic pressures in long-running coding agents that consume millions of tokens per task.

HN vaquarkhan 25d ago

HN: An automated CI/CD testing harness for Model Context Protocol servers

MCP-test-harness 2.0.0: CI/CD testing tool for Model Context Protocol servers with 690+ tests and 100% code coverage, available on PyPI and GitHub Container Registry.

HN loomadesignai 25d ago

AI-powered product photography suite designed specifically for Amazon sellers

LoomaDesign is an AI-powered product photography service for Amazon sellers using image generation to create marketing assets.

HN Rafael_Casuso 25d ago

Show HN: Excalibur. The open-source AI coding agent for product engineers

Excalibur is an open-source AI coding agent for product engineers covering full product lifecycle (discover, build, verify, ship) with immutable event logs and replay capabilities.

HN handfuloflight 25d ago

Interdict: Stop agents from destroying production databases

Interdict is a safety layer for AI agents accessing Postgres, preventing destructive SQL statements through permission-aware validation and rollback capabilities.

HN kiyanwang 25d ago

When AI Costs More Than the Engineer

Analysis of Anthropic's compute spending ($2.3x payroll) versus industry averages, examining how AI lab economics differ from traditional software companies.