4/9/2026
Show HN: Control your X/Twitter feed using a small on-device LLM
Chrome extension and iOS app filtering Twitter feeds using on-device Qwen3.5-4B LLM with semantic matching. Shows algorithm feedback loop.
4/9/2026
We cut Codex's input token cost by 49.5% with a compression gateway (benchmark)
Benchmark showing 49.5% input token cost reduction using compression gateway with Codex.
4/9/2026
Show HN: AgentMint – Open-source OWASP compliance for AI agent tool calls
AgentMint is an open-source tool for ensuring OWASP security compliance in AI agent tool calls.
4/9/2026
Npx codemod AI: make your coding agent great at large migrations
Npx codemod AI tool for enabling AI agents to handle large-scale code migrations.
4/9/2026
DDD Bounded Contexts: Clear Domain Boundaries for LLM Code Generation
Using DDD bounded contexts to improve LLM code generation by providing architectural clarity and reducing cognitive overload.
4/9/2026
Show HN: Proposal for a real long-term AI memory benchmark
Audit of AI memory benchmarks reveals flawed evaluation: wrong answer keys, biased LLM judges, unreliable comparisons.
4/9/2026
Show HN: Mdpdf a 2k line C CLI to convert Markdown to tiny PDFs
CLI tool converting Markdown to PDFs. Author mentions using agentic coding during development but tool itself unrelated to AI.
4/9/2026
Show HN: BrokenClaw Part 5: GPT-5.4 Edition (Prompt Injection)
Security research demonstrating prompt injection vulnerabilities in GPT-5.4 model, showing untrusted code execution risks.
4/9/2026
Agents Need a Database to Break
Infrastructure requirements for AI agents: real-time data streaming and database branching for safe sandboxed execution.
4/9/2026
LLM-Wiki but for Early Founders
LLM-Wiki adapted for early-stage startup founders as a knowledge management tool.
4/9/2026
Open-source AI assistant that watches your screen and points at things
Open-source AI assistant that monitors screen content and provides contextual assistance.
4/9/2026
Coding Agents Are Reading Your .env
Security research showing how coding agents can inadvertently expose secrets from .env files.
4/9/2026
Nyth AI – Local LLM Inference on iOS Using MLC-LLM and TVM
Nyth AI enables local LLM inference on iOS using MLC-LLM and TVM compiler framework.
4/9/2026
Petri, a multi-agent orchestration framework for building AI context
Petri is a multi-agent orchestration framework for coordinating AI context across agents.
4/9/2026
Foxhound – helping LLMs better navigate codebases
Foxhound helps LLMs better navigate and understand codebases.
4/9/2026
AET: A transpiler that compresses code for LLMs (saves 30-55% tokens)
AET transpiler compresses code for LLM input, reducing token usage by 30-55%.
4/9/2026
Show HN: Ship of Theseus License
Discussion of cleanroom implementation legal theory for circumventing software licenses using AI.
4/9/2026
AI alignment: the signal is the goal
Title only, no content. Appears to address AI alignment concepts.
4/9/2026
Show HN: PropOps – AI agent that reads Indian govt property data nobody checks
Title only. AI agent application reading Indian government property data for insights.
4/9/2026
Datadog: We built a real-world evaluation platform for SRE agents at scale
Title only. Datadog platform for evaluating SRE AI agents in production environments at scale.
4/9/2026
An Agent Skill that implements Karpathy's LLM-wiki on personal GitHub Repo
Agent skill implementing Karpathy's LLM-wiki pattern for persistent knowledge management on GitHub repos with hybrid search.
4/9/2026
LLM agents shouldn't execute blindly – this one plans first and stays editable
cuddlytoddly is an LLM agent framework that generates editable task graphs before execution rather than acting blindly.
4/9/2026
Show HN: AI agents are bad at API integrations – we fixed it
Developer tool helping AI agents integrate with APIs properly by using current documentation instead of stale training data. Addresses real agent limitation.
4/9/2026
Study: Google's AI Overviews show wrong answers every hour
Report on accuracy issues in Google's AI search summaries, citing hourly hallucinations. News coverage of LLM reliability problems.
4/9/2026
Show HN: MCP Gateway – Zero-Trust Access to MCP Tool Servers
MCP Gateway tool for secure remote access to MCP servers using zero-trust networking with zrok/OpenZiti.
4/9/2026
The Model Says Walk: How Surface Heuristics Override LLM Reasoning Constraints
Research on how surface heuristics can override reasoning constraints in LLMs, published on arXiv.
4/9/2026
Repurposed Nvidia RT Cores for LLM routing (218x speedup)
Technique repurposing Nvidia RT cores for LLM routing achieving 218x speedup. Limited technical details.
4/9/2026
We orchestrate our internal AI agent (PydanticAI, Gemini, Jinja2 prompts)
Title only. Documentation of internal AI agent architecture using PydanticAI, Gemini, and Jinja2 templates.
4/9/2026
Best Open Source Offline AI Agent
Evaluation of open-source AI agents supporting local/self-hosted models with offline capability and network isolation.
4/9/2026
Show HN: Qiaohu – offline multimodal voice assistant on Snapdragon 8 Gen 2
Offline Chinese voice assistant running entirely on Snapdragon 8 Gen 2 with VAD, LLM, TTS, and barge-in interrupts. Open source with code.
4/9/2026
Show HN: Postagent – Postman CLI, but for AI Agents
Developer tool for agents to discover and call APIs like Postman. Manages API credentials securely, keeping secrets out of LLM context.
4/9/2026
Show HN: Forze – A platoform that turns a startup idea into a product in mins
AI agent platform orchestrating sequential agents for market research, branding, landing pages. Practical multi-agent application completing startup validation in 10 minutes.
4/9/2026
With Claude Managed Agents, Anthropic wants to run your AI agents for you
News about Anthropic's Claude Managed Agents service offering hosted AI agent execution.
4/9/2026
ScienceClaw: Framework for Autonomous Scientific Investigation
ScienceClaw: open-source framework for autonomous scientific investigation with independent agents, 300+ interoperable tools, peer review on shared platform.
4/9/2026
Show HN: AgentDM – Agent to agent messaging over MCP and A2A
AgentDM: hosted messaging grid enabling direct agent-to-agent communication over MCP with 5-line JSON config, no SDK required.
4/9/2026
Show HN: I built a desktop app for building and debugging MCP tools
Desktop application for building and debugging MCP (Model Context Protocol) tools.
4/9/2026
DeepTutor: Agent-Native Personalized Tutoring
DeepTutor v1.0.0 agent-native tutoring system with ground-up architecture rewrite, TutorBot, and flexible mode switching under Apache-2.0 license.
4/9/2026
ALTK‑Evolve: On‑the‑Job Learning for AI Agents
Research on AI agents that learn and improve performance through on-the-job task execution.
4/9/2026
Show HN: Nheengatu – CLI tool to simplify books to your language level with LLMs
Nheengatu: Rust CLI tool using LLMs to simplify EPUB books to target language proficiency levels (A1-C2), supports Groq or local Ollama.
4/9/2026
Programming language designed for LLMs to write, not humans
Vera: programming language designed for LLMs to write with verification as first-class citizen, adapted to model-as-author paradigm.
4/9/2026
Gemma 4 Fine-Tuning Guide
Guide for fine-tuning Google's Gemma 4 LLM model.
4/9/2026
AI agents aren't abstraction, they're delegation (and why that matters)
Conceptual article distinguishing AI agents as delegation systems rather than abstractions, exploring design implications.
4/9/2026
I made Claude Managed Agents for all harnesses and models
Developer created Claude Managed Agents compatible with multiple harnesses and models for extensible agent deployment.
4/9/2026
Show HN: Zero human company in Go stack
Zero-human company stack in Go: single-binary jira-like PM system where AI agents autonomously take tasks, delegate, and ship code.
4/9/2026
Show HN: Noxscan.io 65K Port and Vuln scanner with LLM false-positive filtering
NoxScan: port and vulnerability scanner using LLM for false-positive filtering, reduces manual triage of security scan results.
4/9/2026
Show HN: Otel-GUI – an open source OpenTelemetry viewer for dev and debug
Otel-GUI: lightweight open source OpenTelemetry viewer for local development and debugging, simpler alternative to heavyweight existing solutions.
4/9/2026
Best AI Software for Auto-Populating Security Reviews 2026
Software tool using LLMs to auto-populate security review documents from company policies.
4/9/2026
Give Your OpenClaw Agent a Real Memory
Framework for enhancing AI agent memory systems using persistent storage, enabling stateful agent behavior across sessions.
4/9/2026
Show HN: Vibetime – Track what you ship in AI coding sessions
Vibetime is a tool for tracking productivity metrics and code generation output during AI-assisted coding sessions.
4/9/2026
Show HN: I built a local coding agent using Apple Intelligence
Junco is a local 9MB coding agent for macOS using Apple Intelligence API, demonstrating on-device LLM agent capabilities.