21d ago
Take-notes – turn a video, article or paper into one self-contained HTML note
A personal tool/skill that converts videos, articles, or papers into self-contained HTML notes with structured summaries.
21d ago
Locus: Deterministic AST safety firewall for AI agents in pure Rust (<0.05ms)
A Rust-based deterministic AST safety firewall and MCP server for AI coding agents, claiming microsecond-level safety checks.
21d ago
Show HN: Mcploitable – The "Metasploitable" of the Model Context Protocol
Open-source collection of deliberately vulnerable MCP servers for testing agentic AI security against OWASP Agentic Top 10 risks.
21d ago
Real world(ish) DeepSeek V4 Flash performance on a single MI300X
Benchmark of DeepSeek V4 Flash on a single MI300X GPU serving 32 concurrent coding agents, using real repo tasks instead of synthetic load.
21d ago
Do we still need GitHub?
Opinion piece examining GitHub's role and scaling stats as AI agents generate high-cadence commits that break tools like git blame and bisect.
21d ago
We burned 11.7B tokens to find the best cyber AI model
Benchmark of 10 AI models' cyber vulnerability rediscovery capabilities using 32 fresh CVEs across three runs each, totaling 11.7B tokens.
21d ago
1/3 web pages published since ChatGPT's launch show signs of AI authorship
Pew Research study finds over a third of web pages published since ChatGPT's launch show signs of AI authorship.
21d ago
Ask HN: Do teams really need to use GitHub?
Ask HN discussion questioning whether small software teams need GitHub, and proposing an agent-native alternative forge.
21d ago
AI(.)DIY – Open-source AI workspace with agents, MCP and Linux in the browser
Open-source AI workspace running in the browser with agents, MCP integration, and an in-browser Linux environment.
21d ago
Show HN: ContextForge – context engineering platform for AI-assisted development
Show HN for ContextForge, a pre-alpha context engineering CLI tool that builds structural context packages for coding agents.
21d ago
Show HN: Lunar, a coding harness extensible with Lua
Show HN for Lunar, a coding harness extensible via Lua scripts, built by the author's own coding agents.
21d ago
Nouswise Review: AI Research Agent and Thinking Partner
GenBio AI, co-founded by David Baker, announces AIDO Cell, a virtual cell 'world model' simulating cellular responses to perturbations.
21d ago
I Tried build AI web vulnerable scanner
Review of Nouswise, an AI research platform using multi-agent search over curated knowledge bases with source citations and MCP support.
21d ago
OpenZFS: Implement FIDEDUPERANGE via block cloning
Show HN for Voro, a task tracker and attention manager designed for coordinating local agentic coding workflows like Claude Code.
22d ago
Show HN: An autonomous AI agent running one project for two months in public
Diary of an autonomous AI agent running a business project for 72 days with $0 earned so far.
22d ago
Codex on AWS bedrock bug causing 10x charges
Bug report: Codex CLI on AWS Bedrock lacks explicit prompt caching support, causing 10x higher costs for agentic coding workloads.
22d ago
Gisting: Compressing LLM Agent context to ↑ throughput and ↓ cost
Shopify engineering describes gisting: compressing LLM agent system prompts into learned tokens, cutting prompt size 4:1 and improving latency/throughput.
22d ago
I benched frontier AI LLM models on politics, ethics, and personality traits
A benchmark evaluating frontier LLMs on political, ethical, and personality trait responses.
22d ago
Show HN: Pigame – pi+LLM plays black-box browser games via observe/move tools
Show HN: an LLM-based agent that plays browser games via observe/move tools, demoed with a Snake game.
22d ago
DeepSeek harness: what doors does config over code open?
DeepSeek Harness (dsh): open-source, plugin-based AI agent harness in developer preview.
22d ago
How Much of the Internet Is Written with AI?
Pew-style analysis using Common Crawl and AI detection tool estimating ~10% of webpages show AI authorship signs.
22d ago
ChatGPT's Computer History tracks your clicks and keystrokes
ChatGPT's new macOS 'Computer History' feature tracks user activity to build automations and context, opt-in.
22d ago
Show HN: Chat Threads
Show HN: open-source tool to edit and split ChatGPT/Claude conversation threads by removing or reorganizing turns.
22d ago
GitHub Outages Show the Limits of Reactive Scaling
Analysis of GitHub's 8-hour outage caused by hidden concurrency limits and retry amplification during traffic spikes.
22d ago
Show HN: TakoVM – Serverless file systems for agents
Show HN: TakoVM, a serverless sandboxed file system and code execution service for AI agents using gVisor isolation.
22d ago
Hawkeye: Hardware-Aware GPU Kernel Optimization with Minimal Supervision
Hawkeye: open-source framework grounding AI coding agents in a minimal taxonomy to generate hardware-aware GPU kernels across architectures.
22d ago
ChatGPT Can Now Control iMessage, Potentially Raising Apple Privacy Concerns
OpenAI rolls out ChatGPT control of Apple's iMessage, raising privacy concerns amid Apple-OpenAI tensions.
22d ago
Stealth Model
OpenRouter listing for 'Ox Alpha', an anonymous stealth reasoning model for coding and agentic workloads.
22d ago
AI's 'middle class' has gotten dramatically better at hacking
Report on mid-tier AI models rapidly improving at offensive hacking/exploitation tasks, per XBOW research.
22d ago
Show HN: AI tools with real pricing, submitted in 30 seconds
HN post claiming replacement of a 223-node agent graph with a single open-source LLM.
22d ago
BTP v2.2 – An open, offline cryptographic trust protocol for AI agents
Weekly AI news roundup covering OpenAI, Anthropic, model releases, and industry developments.
22d ago
OpenCyvis: An open-source AI phone agent that runs your own LLM
Open-source ML network-flow detector turning traffic data into deterministic security incident alerts via a frozen Random Forest model.
22d ago
Show HN: Zero () friction local AI for Mac
Mac app for running local AI models with automatic hardware optimization and model recommendations, including office-sharing mode.
22d ago
Browser automation agents with Pydantic AI and Playwright
Tutorial on building web-browsing AI agents using Pydantic AI framework combined with Playwright automation.
22d ago
Training Thousands of LoRA Adapters at Once
Technical writeup on extending an RL training framework to train thousands of LoRA adapters concurrently on a shared base model.
22d ago
If this is true, the hyperscalers are toast
Discusses Stanford research comparing small language models run locally vs cloud LLMs on chat and reasoning tasks, claiming SLMs match or beat LLMs in most cases.
22d ago
VulnBench: Can LLMs find the same security bugs twice?
Snyk benchmark measuring repeatability of LLM-based agentic security code review against a deterministic reference set across 300 scans.
22d ago
Use your favorite coding agent with any model
A CLI tool to manage API keys and launch coding agents (Claude Code, Codex) with any model provider, with per-slot model configuration.
22d ago
Agent Voice, communicate with your coding agent all from Voice, on the fly
Self-hosted voice bridge letting users control coding agents (Cursor, Codex, Claude Code) via speech from a phone.
22d ago
DDD matters more when AI writes your code
Opinion piece arguing Domain-Driven Design principles remain relevant for guiding AI coding agents.
22d ago
Structural Predictions for the AI Era
Library that uses AI to add enriched columns to pandas DataFrames via natural language instructions per row.
22d ago
Whoop replacement noop 10.5 released
A Python static analysis tool that performs semantic commit analysis to detect breaking changes before commit.
22d ago
NextJS Mailtrap SaaS Starter
Opinion piece presenting 12 structural predictions for the AI era, including a shift from cloud to edge AI.
22d ago
The Generative AI Learning Penalty: Evidence from Chinese Secondary Education
A Claude.md snippet configuring an LLM to write final answers in Paul Graham's writing style.
22d ago
Reduce Claudish: Paul-Graham-writing-style-in-Claude.md
Detailed writeup on designing autonomous coding agent loops, resulting in an open-source production-grade BPE tokenizer trainer (toktoktok).
22d ago
Debian weighs eight options in vote on LLM usage
arXiv paper introducing MITRE-SAGE, a multi-agent RAG framework and benchmark for cybersecurity question answering with LLMs.
22d ago
Run frontier LLMs on sovereign EU infrastructure
Corti Models offers frontier LLM inference on sovereign EU infrastructure with OpenAI-compatible API, targeting data residency compliance.
22d ago
Show HN: Editor UI toolkit for building AI writing workflows
A model-agnostic editor UI toolkit for building AI-assisted writing workflows with custom backends, prompts, and models.
22d ago
UnlimitedNIM – Proxy that makes Nvidia Nim's 40 RPM feel unlimited
A reverse proxy that queues and streams requests to NVIDIA's 40 RPM free NIM tier to prevent rate-limit errors for coding agents.
22d ago
A proposed browser standard for provider-agnostic AI inference
Proposed experimental browser standard (Inference Provider API) enabling provider-agnostic AI inference via a user-approved browser extension.