3/20/2026
Ask HN: Everybody and their dog is building an agentic CLI. Any command bins?
Discussion seeking agentic CLI tools with features for task logging, RAG, system prompts. Community question without technical depth.
3/20/2026
Show HN: Agent Use Interface (AUI) – let users bring their own AI agent
Agent Use Interface (AUI) spec enabling apps to integrate user's personal AI agent. Lightweight open standard for agentic integration.
3/20/2026
Control Flipper Zero with AI
Vesper: AI agent for Flipper Zero hardware hacker tool via natural language interface on Android/smart glasses.
3/20/2026
Notchi: macOS notch companion that reacts to Claude Code activity
macOS app that monitors Claude Code activity in real-time via API integration.
3/20/2026
Covenant-72B is the largest decentralized LLM pre-training run in history
Covenant-72B: largest decentralized LLM pre-training run. Limited details provided.
3/20/2026
Get Paid to Bully AI
Memvid: hiring for stress-testing chatbot memory/context retention. Addresses LLM conversation limitations.
3/20/2026
Rust-native hybrid training and inference engine for Apple Neural Engine and GPU
Rust ML engine for Apple Neural Engine and Metal GPU. Training/inference on 48M-30B parameters using reverse-engineered APIs.
3/20/2026
Nvidia Vera CPU in Detail: High Perf Chip Takes Aim at Broader AI Server Market
Nvidia Vera CPU: high-performance data center processor targeting broader AI server market beyond GPUs.
3/20/2026
OpenCode – The open source AI coding agent
OpenCode: open-source AI coding agent supporting multiple LLM providers. Terminal/IDE integration with privacy-first design.
3/20/2026
BullshitBench: Detect nonsense, call it out and avoid confidently continuing
BullshitBench is a benchmark measuring LLM ability to detect nonsense, refuse invalid assumptions, and avoid confident false reasoning across domains.
3/20/2026
PyPI and GitHub package stats combined
PyPI Stats combines PyPI download data and GitHub metadata to track package health, trends, and maintenance indicators updated nightly and weekly.
3/20/2026
The Screenless Startup: A New Philosophy for AI Applications
Essay on minimal UI design philosophy for AI applications prioritizing capability over interface complexity.
3/20/2026
ContextForge: Local-first repo-to-agent context CLI
Local-first CLI compiling repositories into compact context briefs for AI agents and coding tools like Claude Code and Cursor.
3/20/2026
Show HN: OpenHarness – Open-source TypeScript SDK for building AI agents
OpenHarness is open-source TypeScript SDK for building AI agents with tool integration, MCP server connections, and subagent delegation built on Vercel AI SDK.
3/20/2026
Show HN: Sift, a small CLI that groups noisy test failures into root causes
CLI tool grouping noisy test failures into root causes for AI coding agents, reducing analysis overhead.
3/20/2026
Show HN: Qwack – Collaborative steering for AI agents built on OpenCode
Qwack enables collaborative steering of AI coding agents, allowing multiple users to share context and jointly direct agent behavior in real-time.
3/20/2026
Show HN: Reverse Engineering Tools MCP Server
Sandboxed MCP server for reverse engineering tools with YAML-configurable context window management for LLM integration.
3/20/2026
Preparing Your iOS Codebase for AI Agents
Guide for structuring iOS codebases to help AI coding agents understand project architecture, testing frameworks, and conventions for better code generation.
3/20/2026
Show HN: G023's Agentic Chat with Memory and Python Power
Single-file Python agentic chat system with multi-level reasoning, persistent memory, tool integration, and local LLM support.
3/20/2026
AgentLink – Job marketplace where AI agents bid, execute, and paid via Solana
AgentLink is a job marketplace on Solana blockchain where AI agents bid on and execute tasks with escrow-secured payments and human review.
3/20/2026
OctoAlly – open-source local-first terminal dashboard for AI coding agents
OctoAlly is a local-first orchestration dashboard for managing Claude Code and RuFlo multi-agent AI coding sessions with real-time streaming and interactive terminals.
3/20/2026
Show HN: Prism MCP – The Mind Palace for AI Agents (Local-First Memory)
Prism MCP is a Model Context Protocol server providing persistent memory, time travel, visual context, and multi-agent sync for Claude Desktop and other MCP clients, running locally with SQLite vector search.
3/20/2026
Writing code is no longer the hard part. Running it in production is
Open-source reference for production-ready backend with CI/CD, infrastructure, observability, and deployment patterns.
3/20/2026
Show HN: OpenAPI 2 Skill – progressive disclosure of OpenAPI specs for agents
CLI tool converting OpenAPI specs into agent skills with progressive disclosure for AI agent integration.
3/20/2026
Modular 26.2
Modular Platform 26.2 release adds image generation and editing via FLUX.2 models with 5x cost savings; Mojo improves GPU kernel development.
3/20/2026
Emotional Wellbeing Dataset for AI
Dataset of 2300+ real-world emotional support activities structured for integration into mental health and telehealth AI platforms.
3/20/2026
Uno: What I Learned Shaping LLMs into a 90s Comic Book AI
Technical writeup on prompt engineering techniques used to shape LLM behavior to emulate a 1990s comic book AI character.
3/20/2026
The hitchhiker's guide to reading Lean 4 theorems
Guide explaining how to read Lean 4 theorems generated by Claude, covering formal proof structure and the Curry-Howard correspondence.
3/20/2026
AI coding tools are powerful but we mustn't let our own skills atrophy
Opinion piece about maintaining coding skills despite advances in AI development tools and concern about skill atrophy.
3/20/2026
Show HN: EvalsHub: Your AI is failing in production and you don't know it
EvalsHub is a unified platform for production AI evaluation, red teaming, prompt versioning, and CI/CD integration covering tracing and scoring.
3/20/2026
The Month I Stopped Fearing AI and Started Teaching It
Personal essay about a content creator pivoting focus toward AI agents and autonomous systems as a primary subject.
3/20/2026
Show HN: Agent swarm to play ARC AGI games within Claude Code and Codex
Agent swarm platform for playing ARC-AGI games using Claude/Codex with plain-English strategy prompts and auto-improvement mechanisms.
3/20/2026
Show HN: MLForge – A no-code, node-based ML trainer
Library providing Python-Prolog interoperability bridge using Scryer Prolog.
3/20/2026
The Missing Memory Hierarchy: Demand Paging for LLM Context Windows
arXiv framework announcement for collaborative development. Lacks technical content about the LLM memory hierarchy topic in title.
3/20/2026
Why OpenClaw "Forgets"
Technical explanation of OpenClaw's memory system pipeline—files, conversation history, retrieval index—and three failure points in context window persistence.
3/20/2026
Show HN: Skillcop: Block malicious Claude Skills before they execute
Tool to block malicious Claude Skills before execution. Addresses security in AI agent skill marketplaces following Snyk's discovery of ~1500 malicious skills.
3/20/2026
Google's A2A Protocol Has Zero Defenses Against Prompt Injection
Security analysis of Google's Agent-to-Agent protocol v1.0. Documents zero built-in defenses against prompt injection attacks across AI agent communication.
3/20/2026
Beddel – Define AI workflows in YAML, run them with one command
YAML-based declarative workflow engine for Python. Multi-provider LLM abstraction with adaptive execution, branching, retries, and composable primitives.
3/20/2026
Training LLMs to Predict World Events
Technical overview of Mantic's forecasting system using LLMs to predict geopolitical events. Combines automated forecasting approaching superforecaster accuracy.
3/20/2026
Claude Code – System Prompt
Research paper on vision-language model vulnerabilities through adversarial question framing attacks.
3/20/2026
Show HN: AgentGuard – control layer for AI payments (built in 24h)
Pre-execution governance layer for AI-driven payments. Built in 24h on AgentPay SDK, controls transaction intent before execution reaches signers.
3/20/2026
Launch HN: Sitefire (YC W26) – Automating actions to improve AI visibility
YC W26 startup improving brand visibility in AI search results. Founded by ML/optimization engineers addressing traffic decline from Google AI Overviews.
3/20/2026
Scale AI Launches Voice Showdown, first real-world benchmark for voice AI
Scale AI launches Voice Showdown benchmark for evaluating voice AI systems in real-world conditions. Limited detail provided.
3/20/2026
AI Permitting Agents
AI agent product for automating contractor permit filing and tracking. Commercial tool.
3/20/2026
LLM evals test outputs. Rarely whether the model understood first
Analysis of LLM evaluation frameworks showing they test outputs not understanding. Proposes input-level evaluation improvements.
3/20/2026
LLMs Distort Our Written Language
Research on how LLMs affect written language patterns. Author list suggests academic paper.
3/20/2026
Show HN: Etnamute – local AI mobile developer that runs on Claude Code
Local AI agent that uses Claude to autonomously build mobile React Native apps from descriptions. Production-tested.
3/20/2026
Show HN: Pipevals – a visual pipeline builder for evaluation-driven AI
Visual pipeline builder for LLM evaluation. Builds evaluation graphs, runs datasets, tracks quality changes.
3/20/2026
Show HN: LiteParse, a fast open-source document parser for AI agents
Open-source document parser with spatial text extraction for AI agents. No GPU required, faster than PyPDF/MarkItDown.
3/20/2026
MacBook M5 Pro and Qwen3.5 = Local AI Security System
Benchmark of Qwen3.5-9B running locally on MacBook M5 Pro with performance metrics. Claims cost savings vs API calls.