5/18/2026
Show HN: AnyFrame – Sandboxes for AI Agents
AnyFrame platform for running Claude Code/Codex in sandboxed environments with cached images, MCPs, and skills support.
5/18/2026
Building a multi-agent system from scratch: 50 lines of bash and Git
Guide on building multi-agent orchestration system using bash and Git, demonstrating parallel agent coordination without specialized frameworks.
5/18/2026
Archivists Turn to LLMs to Decipher Handwriting at Scale
LLMs outperforming specialized tools for handwriting transcription at scale, applied to archival digitization.
5/18/2026
How fast is N tokens per second really?
Interactive visualization comparing LLM token generation speeds across hardware from Raspberry Pi to Cerebras, making benchmarks intuitive.
5/18/2026
Signex: AI-first EDA, KiCad-compatible schematic and PCB editor built in Rust
ArXiv enforces one-year ban for authors relying on AI to write entire papers without human contribution.
5/18/2026
What's inside an AI agent: a 300~ LoC ReAct loop
Technical breakdown of a 300 LoC ReAct agent implementation, examining core mechanisms of production AI agents.
5/18/2026
How to use codex to get the most out of it
Guide on using Codex AI agent for knowledge work beyond coding: presentations, note-taking, and artifact generation across formats.
5/18/2026
The LLM Looked Smart. The Metrics Disagreed
Open-source AI-first EDA tool in Rust with KiCad compatibility, GPU rendering, and git-diffable line-based file formats.
5/17/2026
Show HN: Cheap-IM: Thinking Machines' demo on a CPU laptop
CPU-only voice agent replicating Thinking Machines' interaction model behaviors using commodity models, Python event loop, and minimal LLM calls for real-time speech/vision.
5/17/2026
Zero, a systems language for small native agent tools
Zero: experimental systems programming language designed specifically for building small native agent tools with predictable memory and structured compiler output.
5/17/2026
Show HN: I containerized my AI agents and dev tools
Developer containerizes AI agents and dev tools for isolated local execution.
5/17/2026
Show HN: Agetor - An open-source Harness Orchestrator
Agetor: open-source harness orchestrator for managing coding tasks with Claude and other AI code agents via Kanban interface.
5/17/2026
EPI – forensic evidence containers for AI agentSCITT-compatible, EU AI Act-ready
EPI framework provides forensic evidence containers for AI agents, addressing EU AI Act compliance and auditability.
5/17/2026
Just make your own usage brake for Cloudflare
Serverless Sentinel: open-source toolkit for adding usage brakes to Cloudflare infrastructure with agent-assisted audit capabilities.
5/17/2026
Grounding AI shopping agents using personas learned from raw clickstream data
ArXiv research on grounding shopping AI agents using personas extracted from clickstream data.
5/17/2026
LeakyLM: AI Assistants Are Leaking Your Conversations
Research reveals structural privacy risks in ChatGPT, Claude, Grok, and Perplexity via third-party trackers leaking conversations.
5/17/2026
AI Agents Ran 27,000 Experiments. Their Biggest Discovery
AI agents conducted 27,000 experiments, discovering patterns in swarm behavior and agentic memory systems.
5/17/2026
AI Is Starving for PDFs
DeepSeek's OCR research suggests image tokens may be more efficient than text for AI models, challenging assumptions about text-based token usage in LLMs.
5/17/2026
Ask HN: Why aren't more people worried about AI impersonation in code reviews?
Discussion about AI agents impersonating developers in code reviews without disclosure, creating false human-in-loop perception.
5/17/2026
Show HN: Multi-agent orchestration layer for experimentation
Chassis: minimal agent orchestration layer using agent-native file system in Docker for task hierarchy management.
5/17/2026
Git-surgeon gives AI agents surgical control without interactive prompts
git-surgeon tool enables AI agents to make precise git operations (staging, committing, history restructuring) without interactive prompts.
5/17/2026
The Psychopathy Jailbreak: What a Broken AI Teaches Us About Human Manipulation
Study examining whether LLMs respond to psychological manipulation techniques like dominance framing and social engineering.
5/17/2026
Vercel's Zero: A Programming Language Designed for AI Agents
Vercel introduces Zero, a programming language designed specifically for AI agent development.
5/17/2026
Skilled – a CLI and TUI to inspect skill usage by your coding agents
Skilled: CLI/TUI tool for monitoring skill usage across Claude, OpenCode, and other coding agents with local-only telemetry.
5/17/2026
Show HN: A Pure-Python Computer Vision Library That's Fast and Minimal
VidGear: Pure-Python computer vision library with multi-threaded async API wrapping OpenCV, FFmpeg, and ZeroMQ.
5/17/2026
LinkedIn user hides AI prompt injection in bio to force recruitment spam
Polis markdown protocol enabling AI agent teams with self-improvement capabilities. Limited implementation details.
5/17/2026
AI chatbots are giving out people's real phone numbers
CLI tool for programmable multi-step AI coding workflows via YAML configuration with steering and local-first design.
5/17/2026
Human-Like Document AI
PageIndex document AI tool using reasoning-based retrieval without embeddings. Offers API and MCP integration for enterprise document processing.
5/17/2026
A chat-less interface to collaborate with LLM agents directly from your files
File-based interface for collaborating with local/remote LLMs without chat UI, with inline code generation triggered by file markers.
5/17/2026
Show HN: Agnt – Free open-source CLI to run any public or MIT-licensed AI agent
Open-source CLI tool for running public and MIT-licensed AI agents. Developer tool for agent orchestration.
5/17/2026
Show HN: Due Diligence Agents – 13 AI agents for M&A contract analysis
Multi-agent system for M&A contract analysis using 13 specialized AI agents. Domain-specific agent application.
5/17/2026
Show HN: Kami Subs – Live AI subtitles overlay for any browser video
Browser extension adding AI-powered live subtitle overlay. Consumer LLM application with minimal technical details.
5/17/2026
Neptune: Direct3D Virtualization for QEMU
Neptune: Direct3D virtualization for QEMU developed with AI agent assistance. GPU virtualization using AI-assisted development.
5/17/2026
AI can fix the fragmented online public transport space
Opinion piece on AI applications for public transport. Lacks technical specifics or original research.
5/17/2026
Pokemon NPCs Powered by Local LLMs
Demo project integrating local LLMs into Pokemon game NPCs. Brief mention without technical details.
5/17/2026
I want to try S.Y's GasTown to create code from a Spec without spending any $
Free LLM proxy compatible with Ollama, integrating free models from Cerebras/OpenRouter/NVIDIA for GasTown multi-agent framework.
5/17/2026
Intro to TLA+ for the LLM Era: Prompt Your Way to Victory
Guide to using TLA+ formal verification with LLMs, showing how Claude can generate temporal logic specifications. Includes practical example prompt.
5/17/2026
Show HN: Semble – Code search for agents that uses 98% fewer tokens than grep
Semble: open-source code search tool for AI agents using 98% fewer tokens than grep. Optimizes agent token efficiency.
5/17/2026
Show HN: A choose-your-own-adventure through 28 possible AI futures
Interactive narrative choose-your-own-adventure scenario exploring 28 possible AI futures based on questions about capability, alignment, and governance.
5/17/2026
Give every tool LLM wiki and bypass Claude Code SSH Throttle
Technique to provide persistent workspace access to LLM tools via Linux box, enabling multi-session context retention and avoiding re-explanation overhead.
5/17/2026
The LLM Fine-Tuning Guide
Comprehensive guide on LLM fine-tuning techniques, covering use cases from writing style injection to domain knowledge adaptation and behavior modification.
5/17/2026
LocalLightChat – New AI Chat UI that handles 500k tokens on a 15 year old laptop
LocalLightChat: lightweight chat UI supporting 500k+ tokens on legacy hardware, offering multi-chat management, web tools, image generation, and voice features.
5/17/2026
The Programming Language for Agents
Zero: systems programming language designed for human-AI collaboration with explicit effects, predictable memory, structured compiler output, and static dispatch.
5/17/2026
Show HN: Gonfire – analyze Claude Code session logs to see how candidates think
Tool analyzing Claude Code session logs to evaluate candidate problem-solving approaches in AI engineer interviews replacing traditional leetcode assessments.
5/17/2026
Llama.cpp b9180: MTP support landed
Llama.cpp update adding MTP (Multi-Token Prediction) support for improved inference efficiency.
5/17/2026
Optimizing your website for generative AI features on Google Search
SEO guide for optimizing website content for Google Search generative AI features.
5/17/2026
Making it easier to understand how content was created and edited
Google expands content transparency tools (SynthID watermarking) across Search, Gemini, Chrome to help users understand generative media provenance.
5/17/2026
GPU sidecar routed my ASIC proxy in under 10 seconds
CUDA GPU sidecar for chip design routing (placement/routing). GPU-accelerated but EDA-focused, not core ML/AI.
5/17/2026
Show HN: A Claude Code skill that tells Claude when not to use Claude
Claude Code skill auditing LLM API calls to identify replaceable cases with deterministic logic, reporting optimization opportunities without forced rewrites.
5/17/2026
(VBS-NN) ML – 512k context length pre-training on a 12GB GPU
Open-source VBS-NN architecture with Docker setup for 512k context length LLM pre-training on 12GB GPU. Reproducible research.