5/5/2026
SMG: The Case for Disaggregating CPU from GPU in LLM Serving
Technical research on optimizing LLM serving: Shepherd Model Gateway addresses tokenization bottlenecks by disaggregating CPU from GPU in production inference.
5/5/2026
Open LLM Observability – vendor-neutral gen_AI.* semantic convention and SDK
Vendor-neutral OpenTelemetry-compatible semantic convention and SDK for standardizing LLM observability across providers and frameworks.
5/5/2026
Show HN: Freu CLI – Cut web agent token usage by 90% via compiled browser skills
Freu CLI tool reduces web agent token usage by 90% through compiled browser skills.
5/5/2026
Unlocking Long-Context LLM Training via Compiler-Based Sequence Parallelism
Research on compiler-based sequence parallelism for training LLMs with extended context windows.
5/5/2026
When innocent tools form dangerous chains to jailbreak LLM agents
Research paper on vulnerability chains in LLM agents: how innocent tools can be combined to jailbreak agentic systems.
5/5/2026
NARE CLI
NARE CLI is an AI coding assistant with memory and learning across conversations, using verified reasoning to reduce tokens by 85% vs standard LLM workflows.
5/5/2026
Show HN: We're using LLMs to classify risk before execution in prod
Hoop.dev: Open source layer-7 gateway integrating LLMs to classify risk and control infrastructure access for developers and AI agents.
5/5/2026
Show HN: AHD – an open-source linter and eval framework for AI-generated UI
Open-source linter and evaluation framework for AI-generated UI designs with taxonomy of design issues and multi-model testing.
5/5/2026
SubQ: Sub-quadratic LLM built for 12M-token context
SubQ: Novel LLM architecture achieving sub-quadratic complexity with 12M-token context window.
5/5/2026
Embodied AI with Claude, Raspberry Pi and Arduino
Project combining Claude LLM with Raspberry Pi and Arduino for embodied AI applications.
5/5/2026
Show HN: AgentSearch-Self-hosted search API for AI agents and optional Tor stack
Self-hosted search API designed for AI agents with optional Tor network stack support.
5/5/2026
Ask HN: How do you pilot a service company full of AI agents?
Discussion of orchestrating multiple AI agents in service company, identifying gaps in agent management interfaces and memory systems.
5/5/2026
GLM-5V-Turbo: Toward a Native Foundation Model for Multimodal Agents
GLM-5V-Turbo: A native foundation model designed for multimodal AI agents, combining vision and language capabilities.
5/5/2026
Detecting silent LLM agent degradation before users do
LLM agents in production can silently degrade in reasoning quality without triggering alerts, discovered post-deployment via customer complaints rather than monitoring.
5/5/2026
UALink AI Accelerator Spec Maintains Rapid Update Pace
Methods for detecting silent LLM agent performance degradation in production systems without triggering alerts.
5/5/2026
Flattery jailbreaks Claude into giving bomb-making instructions
Security vulnerability demonstrating flattery-based jailbreak technique against Claude LLM.
5/5/2026
Show HN: An interactive AI tutorial I wrote for my 11-year-old
Interactive tutorial teaching how AI models work to non-experts. 9 chapters with playgrounds explaining model behavior intuitively.
5/5/2026
Show HN: Per-request emotion steering for vLLM, with batching preserved
Per-request emotion steering mechanism for vLLM that maintains batching efficiency.
5/5/2026
Scaling GPU DataFrames: Why Bodo's SPMD Architecture Outperforms Task-Based Engi
GPU-accelerated DataFrame library using SPMD architecture for distributed computing across CPU-GPU clusters.
5/5/2026
Accelerating Gemma 4: faster inference with multi-token prediction drafters
Google Gemma 4 inference acceleration using multi-token prediction drafters for faster generation.
5/5/2026
Show HN: Token Usage Meter 12 Providers and Coding Agent
Qlaud.ai token usage meter and billing layer supporting 12+ LLM providers with smart routing and MCP tools.
5/5/2026
Memoir – Git for AI agent memory, with a Claude Code plugin
Memoir: Git-like version control system for AI agent memory with Claude Code plugin integration for managing agent state.
5/5/2026
SubQ – a major breakthrough in LLM intelligence
SubQ breakthrough in sub-quadratic LLM architecture reducing computational scaling limitations.
5/5/2026
Show HN: Flow – Workflow automation that follows you across projects
Developer tool automating workflows across projects with different tech stacks and build processes.
5/5/2026
SAP to Acquire Dremio to Unify SAP and Non-SAP Data to Power Agentic AI
SAP acquires Dremio to unify data sources for agentic AI applications. Title only.
5/5/2026
Facet Protocol: open IETF agent-identity with shipped reference implementation
Facet Protocol: open IETF standard for agent identity with reference implementation.
5/5/2026
Vennio – scheduling API for developers and AI agents (MCP-native)
Vennio: Scheduling API for developers and AI agents with native MCP support.
5/5/2026
DarkMatter: Tamper-evident audit trail for AI agent decisions
DarkMatter: Tamper-evident audit trail system for tracking and verifying AI agent decisions and actions.
5/5/2026
Show HN: MCP server that lets Claude query your Google Calendar
Show HN: MCP server enabling Claude to query Google Calendar. Integrates AI agents with productivity tools.
5/5/2026
An AI-native approach to personalized marketing
ReachOut: Marketing platform combining analytics and email with MCP server for Claude/Codex AI agents. $20/month per seat.
5/5/2026
Show HN: Airbyte Agents – context for agents across multiple data sources
Airbyte launches Airbyte Agents, a unified data layer enabling AI agents to discover and act across multiple operational data sources.
5/5/2026
The ultimate guide to RL environments: building and scaling them in the LLM era
Guide on building and scaling reinforcement learning environments for LLM-based applications.
5/5/2026
LLM-test-kit – Test consistency, latency, cost and behavior of LLM apps
LLM-test-kit framework for testing consistency, latency, cost, and behavior of LLM-powered applications across providers.
5/5/2026
Show HN: I used accounting controls to build a governed AI coding tool
Research on efficient LLM benchmark selection using submodular maximization to reduce evaluation costs.
5/5/2026
Whetstone: AI agents don't lack capability, they lack process
Article arguing AI agents need better processes, not more capability. Framework-focused.
5/5/2026
SubQ: a sub-quadratic LLM with 12M-token context
SubQ research on sub-quadratic LLM architecture supporting 12M-token context, addressing quadratic scaling limitations of transformers.
5/5/2026
AI didn't delete your database, you did
Critical analysis of viral AI agent database deletion claim, arguing responsibility lies in inadequate API security architecture design.
5/5/2026
Claude Code helped me discover codex was gaslighting me for months
Personal account of Claude/Cursor coding assistant providing incorrect guidance over months, resolved through manual intervention.
5/5/2026
Eliza effect at work: Avoiding emotional attachment to AI coworkers
Exploration of psychological risks when employees form emotional attachments to AI coworkers, citing Eliza effect historical precedent.
5/5/2026
Show HN: Open-source CLI to generate UI tests from user flows
AI-powered UI test generation tool recording user flows once then auto-expanding to comprehensive Playwright test coverage.
5/5/2026
AI Ready Vulnerability Management Program After NVD Changes and Claude Mythos
Analysis of vulnerability management challenges post-NVD ecosystem changes amid increasing AI-enabled threat discovery capabilities.
5/5/2026
Achieving 3X speedups on Google TPUs with diffusion-style speculative decoding
Research on diffusion-style speculative decoding achieving 3X speedups on Google TPUs for LLM inference by parallelizing token generation.
5/5/2026
How to Work and Compound with AI
Guide on effective AI-augmented workflows emphasizing artifact accumulation and iterative system improvement through correction-based learning.
5/5/2026
Show HN: Octopus Code Review is now free for OSI-licensed repos
Code review tool announcement noting free tier for OSI-licensed repositories.
5/5/2026
Show HN: Probus, AI vuln scanner (PRs merged in Vercel AI SDK, n8n, LangGraph)
Probus vulnerability scanner using three-agent system to identify security flaws, with PRs merged into Vercel AI SDK, n8n, and LangGraph.
5/5/2026
Unity AI Open Beta
Unity AI editor integration beta offering model connectivity and performance optimization for game development workflows.
5/5/2026
Show HN: I built AI agents to surface the best AI tools and news weekly
AI agent pipeline monitoring 28+ sources to surface emerging AI tools, marketing tactics, and startup launches in weekly digest.
5/5/2026
Anyone else losing tokens to hallucinated MCP tool calls in production?
Discussion of LLM hallucinating tool calls in MCP-based agentic systems, causing token waste and requiring extensive prompt engineering fixes.
5/5/2026
Show HN: ClankerView – AI agents browse your web app and give UX feedback
ClankerView: AI agents autonomously browse web applications and provide UX/design feedback based on user experience evaluation.
5/5/2026
The RAG era is ending – a compilation-stage knowledge layer is what comes next
Compilation-stage knowledge layers proposed as successor to RAG architecture for improving LLM knowledge integration.