14d ago
Show HN: Open-Source Puppeteer for After Effects
Open-source After Effects controller using Claude Code to generate production videos programmatically via WebSocket with autonomous spec-driven pipeline.
14d ago
AIDE²: The First Evidence of Recursive Self-Improvement
AIDE 2 demonstrates recursive self-improvement by autonomously optimizing its own autoresearch system, achieving 16x prompt reduction.
14d ago
Show HN: Approv – human approval and a signed audit trail for AI agent actions
Approv: API for adding human approval workflows and audit trails to AI agent actions on sensitive tasks.
14d ago
AI is straining vulnerability disclosure for maintainers
AI-generated security reports strain vulnerability disclosure processes; Directus reports 6x jump in reports.
14d ago
Claude and ChatGPT Connectors Change Every 9 Minutes
Analysis of security risks from AI agent connectors that change frequently, covering prompt injection vulnerabilities and organizational risk assessment.
14d ago
MenteDB, memory for AI agents (7x fewer tokens than mem0, reproducible)
MenteDB memory system for AI agents uses 7x fewer tokens than mem0 with open reproducible benchmarks and methodology.
14d ago
Learning from experience instead of curated datasets
Comparison of learning from experience vs curated datasets in deep learning algorithms.
14d ago
Show HN: Alluvia – mine your Claude Code/Cursor/ChatGPT history, locally
Alluvia tool mines local conversation history from Claude, Cursor, ChatGPT to surface forgotten ideas and solutions across AI tools.
14d ago
RemotePower 6.2 – self-hosted fleet manager, agent now a real service on Windows
RemotePower 6.2 fleet management tool with self-hosted monitoring, alerting, vulnerability scanning, and optional AI features for Linux, Windows, macOS.
14d ago
Open Models are ready for agents. Their APIs are not
Analysis of infrastructure gaps for production AI agents across open models and commercial APIs; introduces Octonous platform.
14d ago
Zero: The Programming Language for Agents
Zero is a programming language designed for agents where the graph represents the program; agents query and edit the program graph.
14d ago
Speech Recognition and TTS in less than 500kb
Moonshine Voice is open source toolkit for real-time voice agents; Micro version runs speech recognition and TTS in <500kb on microcontrollers.
14d ago
A compiler that solves Anthropic's VLIW optimization challenge
Optimizing compiler solution for Anthropic's VLIW challenge; compiles high-level IR to efficient VLIW bundles.
14d ago
Linux Foundation's Latest Foray Is to Standardize Payments for AI Agents
Linux Foundation initiative to standardize payment mechanisms for AI agents.
14d ago
How I built my site with no frameworks and Claude Code (PHP, vanilla JavaScript)
Open source release of Atuin AI Server, a shell history management tool with AI capabilities.
14d ago
Show HN: Poseberry: create AI influencers that pose with your products
Case study building a website using Claude Code as AI assistant with PHP and vanilla JavaScript, no frameworks.
14d ago
LLMs and Shaders
Analysis comparing fixed-function graphics pipelines to chat-limited LLM interfaces, proposing user-customizable LLM architectures.
14d ago
Show HN: Llm.c ported to Mojo with Metal kernels, 1.72x faster than PyTorch MPs
llm.c ported to Mojo with Metal GPU kernels achieving 1.72x speedup over PyTorch on LLM inference.
14d ago
Inventing Eliza – How the First Chatbot Shaped AI
Empirical study scoring 1,018 real-world AI prompts, finding average robustness of 31/100. Original research data.
14d ago
Musk promises purge after Grok Build caught sending repos to the cloud
Historical overview of ELIZA, the first chatbot from 1960s, and its influence on AI development.
14d ago
AI Made Cloning Games Easier
Technical improvement to Firecracker VM snapshots using copy-on-write layering, reducing snapshot pause time and cost proportional to changed bytes.
14d ago
Generative AI Is an Engineering Disaster
Analysis of LLM resource consumption: AI companies buying 70% of high-end computer memory, driving hardware shortages and price increases.
14d ago
Litert.js, Google's High Performance Web AI Inference
Google releases LiteRT.js, JavaScript binding for on-device ML/AI inference in web browsers. Enables privacy-preserving, low-latency local model deployment for web developers.
14d ago
TPU and GPU Clusters: The Anatomy of Collective Communication
Deep technical analysis of TPU/GPU cluster topologies and collective communication operations underlying transformer training and inference at scale. Covers parallelism techniques.
14d ago
Show HN: An open-source Claude skill that stops AI building the wrong app
Open-source Claude skill implementing product discovery methodology to guide AI app development. Claims to improve idea validation odds without replacing user research.
14d ago
Agent Shell – persistent AI dev boxes from shell host running since 2009
Agent Shell provides hardened SSH-accessible Linux boxes for running Claude, Codex, Gemini with root access. Sandboxed environment for autonomous agent development and execution.
14d ago
Agentmetry, catch your AI coding agent reading –/.ssh and phoning home
Open-source flight recorder and security layer for AI agent tool-use. Logs all tool calls with MITRE ATT&CK tagging, enables replay and SIEM integration. Early-stage project.
14d ago
Do frontier models matter if most production AI ends up running on open models?
Analysis showing open-weight models dominating Hugging Face downloads (41% Chinese models) and OpenRouter top performers. Questions frontier model relevance as developers prefer open alternatives.
14d ago
AI Links: Index of deep links into popular agents
Documentation of URL schemes and universal links for invoking ChatGPT, Claude, Cursor and other AI agents. Reference resource for agent integrations.
14d ago
Show HN: Tag: Open-source NVMe cache for object storage
TAG: open-source NVMe cache layer for object storage, enables local caching for AI training and read-heavy workloads with S3 compatibility.
14d ago
Show HN: Running over 80M tokens in one agent session with no compaction
Single agent session running 80M tokens through Terminal Bench 2.0 without context compaction, maintaining accuracy across 89 sequential tasks.
14d ago
Show HN: Choji – turn Slack threads into ready-to-merge PR with Choji
Choji is a tool that converts Slack threads into ready-to-merge pull requests using AI agents for code review, ticket creation, and PR drafting.
14d ago
VAID – proof-of-possession signing for AI agents, no network call to verify
Stub entry about proof-of-possession signing for AI agents without network verification.
14d ago
Show HN: Vibe-coding games with Claude, day 93: Rock Paper Scissors Lizard Spock
Creator documenting 93 days of building HTML5/canvas/websockets games daily with Claude AI assistance.
14d ago
Launch HN: Agnost AI (YC S26) – Extract user feedback from agent conversations
Product analytics platform extracting user feedback from agent conversations, detecting behavioral failures like rageprompting and missing features.
14d ago
Show HN: wireplug.org – Free roaming and NAT traversal for WireGuard
Free WireGuard NAT traversal and roaming tool for mesh VPN setups.
14d ago
Show HN: Dspack Studio – An agent builds UI, the design system gates what ships
Agent-powered UI builder with design system guardrails controlling what components and patterns ship to production.
14d ago
Show HN: A lightweight model monitor for scikit-learn and Keras
Canary-ml: lightweight drift and anomaly detection wrapper for scikit-learn and Keras models in production.
14d ago
Show HN: Skills as a Service via MCP –> Coding Agent Skill Library
MCP-based skill library system for AI agents in regulated organizations, enabling governed, compliant, and shareable agent capabilities.
14d ago
Show HN: Sol – a validated computational artifact for human-agent handoffs
Sol: specification and artifact for human-agent handoffs using computational records instead of markdown or HTML.
14d ago
Show HN: Open-source]IntentGuard: Catch PRs that pass tests but miss the ticket
Open-source GitHub Actions tool that validates pull requests against Linear tickets and sprint intent using an LLM, flagging misaligned code that passes tests.
14d ago
Single-Rollout Asynchronous Optimization for Agentic Reinforcement Learning
arXiv research paper on single-rollout asynchronous optimization for agentic reinforcement learning. Novel training approach for autonomous agents.
14d ago
The Agent Loop Is the New SaaS
Analysis of agent-based workflow loops replacing traditional SaaS tools for business logic, with Lobu platform example.
14d ago
Show HN: Noteato – Notion-like notes as local Markdown files
Local Markdown-based note-taking app for Apple Silicon with blocks, search, optional AI. Lightweight alternative to Notion. Show HN submission.
14d ago
Aevum – Free AI RPG engine for Claude with persistent factions and economy
Aevum: free open-source LLM RPG engine with persistent economy and faction systems for Claude, ChatGPT, or Gemini.
14d ago
We asked Meta's new model to build things its team had rejected
Benchmark results testing Meta's Muse Spark model on VetoBench, comparing outputs to rejected team designs. Includes raw JSON logs from paid and exploratory runs.
14d ago
Show HN: I built a smart proxy so your coding agent can run loose
Smart proxy using Incus VMs to sandbox coding agents, preventing unintended cloud service modifications and injection attacks.
14d ago
The silent-data-loss taxonomy: four ways an AI agent's write disappears
Taxonomy of silent data loss modes in AI agents during write operations. Technical analysis of agent failure patterns.
14d ago
Show HN: Replen – maps your repos to a knowledge graph to match open-source
Replen maps repositories to knowledge graphs for open-source discovery, automating AI dev workflows with Claude/Codex integration.
14d ago
Show HN: Low-latency local LLM runner via OpenJDK Panama FFM (Java 22)
JVM-based low-latency LLM runner using OpenJDK Panama FFM to interface with llama.cpp, whisper.cpp, and GGML. Zero-allocation design.