6/20/2026
Open source alternative to brain by perplexity
Rudi: LLM memory system using causal graphs instead of full transcripts. Reduces API costs and context limits.
6/20/2026
Show HN: Autonomy – Self-Harness/Self-Directed AI Agent Core Under Development
Autonomy is self-directed AI agent framework with autonomous execution, risk-aware action gating, and outcome evaluation via approval policies.
6/20/2026
I created an open source app to check if a photo is from an AI or a Camera
C2PA Verify is open-source Android app using Content Credentials standard to verify photo authenticity and detect AI-generated images.
6/20/2026
Show HN: AI Voice Generator for Creators
Speakora is AI voice generator for creators to produce natural narration for YouTube, product demos, and brand content.
6/20/2026
Release 4.0.0 · HuggingFace/Transformers.js
Transformers.js v4.0.0 released with new WebGPU runtime in C++. JavaScript/WebGPU LLM inference library.
6/20/2026
Show HN: Local OCR and Image Analysis for LLMs via Apple Vision (macOS)
MCP server wrapping macOS Vision framework for on-device OCR, PDF text extraction, barcode detection, and image classification without cloud APIs or token costs.
6/20/2026
Compress tool outputs, logs, files, RAG chunks before LLM for 60-95% less tokens
Compression library reducing LLM token usage by 60-95% for tool outputs, logs, and RAG chunks.
6/20/2026
AI and the Great CMS Unbundling
Analysis of how AI is unbundling content creation from control in CMS systems, changing their role rather than replacing them. Discusses AI agents as alternative.
6/20/2026
Moebius: 0.2B Lightweight Image Inpainting Framework with 10B-Level Performance
Lightweight 0.2B image inpainting model achieving 10B-level performance through systematic diffusion backbone optimization.
6/20/2026
Show HN: AgentArk – open-source self-hosted AI agent OS
AgentArk is open-source self-hosted AI agent runtime. Build agents from prompts/tools, deploy as apps/automations, monitor actions, manage context. Full lifecycle management.
6/20/2026
Show HN: Muninn - 8 Security scanners in one GitHub Action
Open-source GitHub Action combining eight security scanners with unified output format.
6/20/2026
LLMs Are Complicated Now
Analysis of increasing architectural complexity in modern LLMs compared to earlier Transformer-based designs.
6/20/2026
I am dreading our LLM-written incident report future
Critical perspective on LLM-generated incident reports, discussing risks and limitations while acknowledging toil reduction benefits.
6/20/2026
Build a full knowledge graph locally with any video
GoodQ4All: offline video processing system with transcription, diarization, semantic search using local Whisper, Qdrant, and SQLite.
6/20/2026
Out of Stealth (Kinda)
LodeDB: embedded vector database for local RAG with GPU acceleration, achieving 24k-50k queries/sec with sub-millisecond persistence.
6/20/2026
Show HN: Agentcard – virtual cards for AI agents, now with DoorDash checkout
Agentcard: virtual card infrastructure enabling AI agents to interact with real-world services like DoorDash with payment processing.
6/20/2026
Show HN: P34 – tabular profit prediction without the optimism bias
P34 framework addressing optimism bias in ML models trained on historical business decisions, preventing over-optimistic production deployments.
6/19/2026
Automating model design for edge AI
DeepGate's automated model search produces edge AI models 45× faster and using 11× less RAM than reference implementations.
6/19/2026
Our edge AI compiler outperforms Google and vendor toolchains
DeepGate compiler achieves 3× RAM reduction and 2× speedup vs Google TFLM on microcontrollers.
6/19/2026
GPU Puzzles (2021)
Interactive notebook teaching GPU programming fundamentals using NUMBA for Python developers to build GPU kernels without prior GPU experience.
6/19/2026
Knowledge workers don't need frontier models
Analysis arguing smaller LLMs with intelligent routing outperform frontier models for knowledge workers, optimizing cost and performance tradeoffs.
6/19/2026
Show HN: Evaluating Local LLMs as language translators for my app
Evaluation of local open-source LLMs for language translation tasks, comparing against Google Translate with methodology discussion.
6/19/2026
Show HN: CWC scans your Claude Code history and auto-builds agent workflows
Visual desktop app for composing multi-agent Claude Code workflows with drag-and-drop canvas interface.
6/19/2026
Show HN: PostgreSQL MCP Server with 135 tools for various purpose
PostgreSQL MCP server with 135 tools for Claude Desktop and MCP-compatible AI systems. Lock-free connection pooling, sub-10ms latency.
6/19/2026
Show HN: Surface skill for HTML pages a coding agent watches and reacts to
Web interface framework for AI agents to interact with rich UIs beyond text. Enables agents to manipulate tables, images, and complex layouts.
6/19/2026
Show HN: NanoEuler – GPT-2 scale model in pure C/CUDA from scratch
GPT-2-scale language model built from scratch in pure C/CUDA without ML frameworks. Includes hand-written tokenizer, training pipeline, and CUDA engine.
6/19/2026
Reverse engineering the Qualcomm NPU compiler
Reverse engineering Qualcomm NPU compiler to enable faster edge deployment of ML models on NPU hardware.
6/19/2026
Uptime.com's New MCP
Model Context Protocol server for Uptime.com monitoring, enabling AI assistants to manage infrastructure checks and alerts.
6/19/2026
querychat
Natural language interface for querying tabular data using LLMs and SQL, available for R and Python.
6/19/2026
How to get product feedback from agents
Sanity MCP server handles 3M+ agent tool calls; article describes gathering feedback from AI agents using a Model Context Protocol integration.
6/19/2026
Teams using AI for outbound email are finding it harder, Folderly research finds
Folderly survey shows 70% of B2B sales teams use AI for outbound email but face 79% delivery failure rates as recipients detect AI content.
6/19/2026
Beast – governed output gateway for AI coding agents
Beast is a governance layer for AI coding agents that enforces output contracts, repairs patches, and optimizes tool calls before filesystem writes.
6/19/2026
Estonia to give digital identities to AI agents
Estonia issuing digital identities to AI agents for legal/regulatory recognition.
6/19/2026
Show HN: I built an 11-LLM consensus engine to detect AI hallucination
11-LLM consensus engine to detect hallucinations in AI outputs. Developer tool for improving LLM reliability.
6/19/2026
PageToMD – A CLI tool to turn web pages into clean Markdown for AI agents
CLI tool converting webpages to clean Markdown for AI agents. Open source utility with piping support and crawling capability.
6/19/2026
Switching AI Tools Mid-Sprint Cost Us a Day (and What We Learned) – Week 6 R
Case study on switching AI tools mid-project causing productivity loss. Brief anecdote without technical depth.
6/19/2026
Why local AI – and why it matters
Argument for local, on-premise AI systems vs cloud AI for privacy and reliability. Conceptual analysis of deployment approaches.
6/19/2026
How to Drive an LLM
Guide on LLM prompting and interaction techniques. Incomplete content.
6/19/2026
What 'Getting Your Hands Dirty' Means at LLM-Era
Article about practical LLM engineering and hands-on development practices. Incomplete content.
6/19/2026
High-performance code intelligence MCP server
High-performance MCP server provides fast code intelligence for AI agents, indexing large repositories in seconds with tree-sitter AST parsing across 158 languages.
6/19/2026
Show HN: Redteam:If you are using more than 2 coding agents
Redteam is an adversarial agent-pair harness for AI-assisted coding that uses two independent models to review and challenge code outputs before merge.
6/19/2026
I tested Gemma4 12B on my 8GB GPU, now I don't want to go back to smaller models
User tests Google's Gemma4 12B local LLM on 8GB GPU, finding recent model improvements focus on consumer hardware efficiency over scale.
6/19/2026
Do Vision-Language Models Understand 3D Scenes or Just Catalogue Objects?
Benchmark evaluating whether vision-language models understand 3D spatial layout beyond object recognition via occlusion and geometry tasks.
6/19/2026
Pseudo-Formalization for Automatic Proof Verification
Pseudo-Formalization method automatically converts AI-generated mathematical proofs into verifiable formal language representations.
6/19/2026
Superhuman Safe and Agile Racing through Multi-Agent Reinforcement Learning
Multi-agent reinforcement learning approach for safe, superhuman autonomous racing coordination in dynamic shared spaces.
6/19/2026
Light Interaction: Training-Free Inference Acceleration for Interactive Video World Models
Light Interaction provides training-free inference acceleration for interactive video world models via optimized attention and caching.
6/19/2026
UltraEP: Unleash MoE Training and Inference on Rack-Scale Nodes with Near-Optimal Load Balancing
UltraEP system optimizes expert parallelism training and inference for mixture-of-experts models with dynamic load balancing.
6/19/2026
The ACUTE Protocol: Operationalizing Language Model Activations for Better Calibration, Utility, and Trust
ACUTE Protocol uses language model internal activations to improve confidence calibration, reducing overconfidence and enhancing trustworthiness.
6/19/2026
Streaming Interventions: Can Video Large Language Models Correct Mistakes as They Occur?
Evaluates video large language models' ability to provide real-time mistake correction guidance during instructional task execution.
6/19/2026
An adaptive framework for the axisymmetric pulsar magnetosphere using physics-informed Kolmogorov-Arnold networks
Physics-informed Kolmogorov-Arnold networks with domain-specific adaptation for modeling axisymmetric pulsar magnetosphere dynamics.