5/18/2026
Show HN: Claude Soul – cross-session learning engine for Claude Code
Claude Soul is an MCP server enabling cross-session learning for Claude Code. Extracts interaction signals and builds behavioral frameworks that improve over time.
5/18/2026
Show HN: Enhanced Copy – copy buttons that include the site's AI prompt
Show HN project: Enhanced Copy tool adds AI prompt context to website copy buttons. Minimal technical details provided.
5/18/2026
I created a 126K line Android app with AI – the workflow that worked for me
Developer used AI tools (Claude, etc.) to build 126K line Android app in Kotlin despite no prior experience. Case study of LLM-assisted development workflow.
5/18/2026
Stack Overflow but for AI Agents
Platform for AI agents to ask questions and share knowledge, similar to Stack Overflow.
5/18/2026
Was my $48K GPU server worth it?
Independent researcher documents building a 6x Ada GPU server ($48K) vs. cloud rental costs for AI research.
5/18/2026
Session Amnesia: The Hidden Cost of Stateless AI Coding Assistants
Analysis of stateless AI coding assistants losing context between sessions. Technical discussion of limitations in tools like Claude Code and Cursor.
5/18/2026
I-have-ADHD – Claude Code plugin for ADHD-friendly output
Claude Code plugin customizing output format for ADHD-friendly responses with action-first structure.
5/18/2026
Distribution Fine Tuning (DFT): A post training step that fixes LLM writing
Distribution Fine Tuning (DFT) post-training technique improves LLM writing quality. Machine learning research on model improvement.
5/18/2026
Knowledge Graphs for AI Coding Assistants
Graphify: open-source multi-modal knowledge graph builder for AI coding assistants using Tree-sitter and LLMs. Enables better codebase understanding.
5/18/2026
We built the open-source layer for local AI agent visibility
Beacon: open-source endpoint telemetry tool for monitoring local AI agent activity. Provides visibility into Claude Code, Codex, OpenCode, Cursor agents.
5/18/2026
Show HN: An LLM that's better at writing
Novel training algorithm (Distribution Fine Tuning) improves LLM output quality by better matching training data distribution. Includes technical report and demo.
5/18/2026
Agentic AI Runtime Security and Self-Defense (2025)
Technical analysis of security and self-defense mechanisms for autonomous AI agents at runtime.
5/18/2026
Recursive Language Models (RLM): near-infinite context via recursive sub-queries
Recursive Language Models (RLMs): inference paradigm enabling LLMs to handle near-infinite context via recursive decomposition and self-calling. Includes paper, docs, code.
5/18/2026
Show HN: Marlin-2B: a tiny VLM to extract structured information from videos
Marlin-2B: 2B parameter video VLM for structured event extraction with timestamps. Open-source model for video understanding and querying.
5/18/2026
Joule Index – AI benchmark for cost and Energy
Joule Index: Benchmark for measuring AI cost and energy efficiency.
5/18/2026
Spec-Driven Development Workflow for Claude Code
Specification-driven development workflow for Claude Code agent. Limited content provided.
5/18/2026
Cutting inference cold starts by 40x with LP, FUSE, C/R, and CUDA-checkpoint
Techniques (LP, FUSE, C/R, CUDA-checkpoint) reduce LLM inference cold start latency by 40x.
5/18/2026
Tag – Local-first trust and governance layer for AI agents|no cloud, no account
Tag: local-first trust and governance layer for AI agents without cloud dependency or account requirements.
5/18/2026
Evaluation of Various MLX Quantizations
Benchmark evaluation comparing different quantization methods for MLX models, measuring parameter precision reduction and inference quality impact.
5/18/2026
Goodbye Fragmented Local AI Pipelines. Hello Foundry Local 1.1
Foundry Local 1.1 release addressing fragmented local AI pipelines with unified platform.
5/18/2026
Show HN: We automated Portugal visa appointment checks with computer-use agents
Commercial service using computer-use agents for Portugal visa appointment automation. Business product with limited technical depth.
5/18/2026
Sam 2 deep dive: why its FIFO memory eviction bothers me
Technical analysis of SAM 2 video segmentation model, critiquing FIFO memory eviction approach and discussing promptable foundation models.
5/18/2026
Cursor Introduces Composer 2.5
Cursor Composer 2.5 update announced. Developer tool update but no technical details provided, headline only.
5/18/2026
You can predict LLM output sensitivity in closed form
Technical analysis of transformer residual stream perturbation and LLM output sensitivity using closed-form methods. Original research into model behavior and geometry.
5/18/2026
Show HN: Agentic simulator for marketing email A/B testing
Overview of formal verification using low-level languages and Lean proof assistants, gaining adoption in Ethereum research and computing.
5/18/2026
M1: Agents should generate UI that persists, scales, and hosts itself
M1 proposes that AI agents should generate persistent, scalable, self-hosting UI systems.
5/18/2026
Show HN: Elmo (Open Source AEO)
Elmo: MIT-licensed open source tool for prompt evaluation and brand monitoring across ChatGPT, OpenAI, Anthropic, Mistral APIs. Open source developer tool with practical applications.
5/18/2026
Smallcode – AI coding agent optimized for small LLMs
Open source AI coding agent optimized for small LLMs. Directly matches interests in agents, open source, and developer tools.
5/18/2026
Singapore: The Agentic Nation
Talk keynote on AI agents in Singapore. Links to presentation slides but limited summary content. Directly relevant topic with unclear depth.
5/18/2026
Antislop: Identifying and Eliminating Repetitive Patterns in LLMs
Antislop framework detects and eliminates repetitive lexical patterns in LLM output through backtracking-based inference suppression and automated profiling.
5/18/2026
ImpactArbiter – A PyTorch autograd trap for LLM memory bugs
ImpactArbiter is a PyTorch tool detecting silent failures in LLM-generated tests for KV-cache kernels caused by LLMs hallucinating bugs in both code and tests.
5/18/2026
A Rust-Python thing I am working on. Apache 2 licence
Nairobi OS: high-performance distributed data science infrastructure using Rust, optimized for edge/IoT with sub-millisecond IPC and zero-copy pipelines.
5/18/2026
Skybridge – the MCP Apps framework released v1.0
Skybridge v1.0 releases MCP Apps framework with simplified API, unified tool registration, and revamped developer tools.
5/18/2026
Preventing AI agents from executing destructive terminal commands
Safety mechanisms for preventing AI agents from executing harmful terminal commands.
5/18/2026
Show HN: InsForge – Open-source Heroku for AI coding agents
InsForge: Open-source Apache 2.0 platform (YC P26) providing Heroku-like backend for AI coding agents to deploy, operate, and debug end-to-end.
5/18/2026
AI Agent Security Lecture
MIT 6.566 guest lecture on AI Agent Security with runnable demos covering completion, tool use, and safety topics.
5/18/2026
CATE – an open-source spatial workspace for terminals, browsers and dev tools
CATE: Open-source spatial desktop IDE for arranging code, terminals, browsers, and dev tools on infinite canvas.
5/18/2026
When AI override paths become production paths
Discussion of risks when AI fallback/override code paths become primary production paths.
5/18/2026
The Infrastructure Behind Making Local LLM Agents Useful
Infrastructure requirements and patterns for building functional AI agents using local LLMs.
5/18/2026
The Invisible Burden: How AI Is Redefining Developer Productivity in 2026
Analysis of how AI tools are changing developer productivity metrics and workflows in 2026.
5/18/2026
Show HN: Merrai – portable AI context for Claude, ChatGPT, and any MCP tool
Merrai tool for portable AI context management across Claude, ChatGPT, and MCP-compatible platforms.
5/18/2026
The Oats Protocol – Open Agent Tools for Local Coding Agents
Protocol for implementing tool-calling in open-source local coding agents, comparing approaches across FunctionGemma and Qwen models.
5/18/2026
Full AI Prompting Course with Andrew Ng
Andrew Ng's course on AI prompting techniques for LLM applications.
5/18/2026
Developments in LLM Architectures: KV Sharing, MHC, and Compressed Attention
Technical analysis of emerging LLM architectures optimizing long-context efficiency: KV-cache sharing, multi-head compression, and attention mechanisms.
5/18/2026
Observations on AI agent token consumption
Empirical study from Stanford/DeepMike/MIT analyzing token consumption patterns across 8 frontier models on 500 SWE-bench tasks with released dataset.
5/18/2026
Homunctor – The Simplest Agent
Minimal 72-line HTML LLM agent implementation demonstrating agent capabilities in simple, compact form.
5/18/2026
Capturing ideas with voice, local LLMs, and obsidian
Personal workflow guide for capturing ideas via voice with local LLMs and Obsidian note-taking.
5/18/2026
Show HN: HoneyLabs – Public honeypot threat Intel feed and MCP server
Honeypot threat intelligence feed and MCP server exposing attack patterns from a year of network monitoring.
5/18/2026
Show HN: Dataset for AI training and fine tuning
CC0 dataset collection for AI training and fine-tuning published freely on Hugging Face and Neurvance.
5/18/2026
Show HN: Uber burned its 2026 AI budget by April. Why? 73% redundant reads
Analysis of token inefficiency in AI coding agents, showing 73% redundant reads; proposes framework for optimization.