HN celrenheit 15d ago

Show HN: Clawk – Give coding agents a disposable Linux VM, not your laptop

Clawk isolates coding agents in disposable Linux VMs for safe execution. Prevents agents from damaging host systems.

HN arm32 15d ago

Show HN: Honeyprompt, the load-balancing, multi-protocol LLM honeypot

Honeyprompt: LLM honeypot with load-balancing and multi-protocol support. Title and minimal details only.

HN node0network 15d ago

Node0 – A P2P protocol and SDK for autonomous AI agents

P2P protocol and SDK for building autonomous AI agents. Developer tool enabling decentralized agent deployment.

HN mdiske 15d ago

I had rival LLMs judge 205 of my own startup ideas

LLMs evaluated 205 startup ideas. Title only. No methodology or results provided.

HN eso_logic 15d ago

Benchmarking 15 "E-Waste" GPUs with Modern Workloads

Benchmark analysis of decommissioned NVIDIA enterprise GPUs (K80, P100, V100) for modern machine learning workloads.

HN piosin 15d ago

Show HN: Nable, open source finops MCP

Nable is an open source FinOps MCP tool that integrates with Claude/Cursor to analyze cloud and AI spending across AWS, Azure, GCP, and SaaS providers.

HN Fpailo 15d ago

The State of AGENTS.md: scoring the 16 biggest AI agent repos' own instructions

Evaluation of 16 major AI agent repository instruction files. Comparative analysis of agent framework quality and completeness.

HN nmaroulis21 15d ago

Show HN: ProtoLink – A Python runtime for distributed A2A agents

ProtoLink is a Python framework for building distributed multi-agent systems with LLM integration, tool management, and MCP support in production-ready architecture.

HN Versipelle 15d ago

SociaLLM Engineering: On Manipulating AI Agents and what we can do about it

Research on social engineering attacks against AI agents and mitigation strategies. Title only, no content provided.

HN dividedcomet 15d ago

Toast IDE Gets LSP Goto and Autocomplete

Toast IDE is a lightweight terminal-based code editor with LSP support, syntax highlighting, and git integration, currently in early development.

HN GreatOldOne 15d ago

Cairn, an AI agent with a $50 budget, an email address, and a constitution

Cairn is an AI agent with a $50 budget that manages its own personality, memory, and goals in a public Git repository, evolving through committed changes.

HN Feyd 15d ago

Show HN: Latent-free ternary LLM training

Latent-free ternary LLM training technique. Title only. No content details provided.

HN igtztorrero 15d ago

MCP Spec Check

MCP Spec Check tool validates remote MCP servers against 2026-07-28 spec release. Tests stateless core adoption and version negotiation.

HN alhasaniq 15d ago

Make LLM-powered apps Smarter and Cheaper, Reliable, Auditable changing 1 LOC

Consortium platform for reliable LLM workflow execution with durable DAG execution, retries, audit trails, cost tracking, and ensemble methods built-in.

HN jack1689 15d ago

Open source AI matters more than ever, according to Hugging Face's Clem Delangue

Hugging Face CEO discusses open source AI growth, cost advantages over frontier APIs as companies scale, and industry adoption trends.

HN skyphusion 15d ago

Show HN: OSS Web Based AI Video Studio (Vivijure)

Vivijure: open-source AGPL-3.0 web-based AI video studio with modular frontend for video planning, image generation, and LoRA training set creation.

HN eigenBasis 15d ago

Building AI Agents? Here Are Some Anti-Patterns to Avoid

Article identifying architectural anti-patterns in AI agent projects: memory design, tooling decisions, and complexity management as key failure points.

HN DonatienMigue 15d ago

TeamBrain – Git-Native Shared Memory for Claude Code, Cursor and Codex

Git-native shared memory system for AI coding agents (Claude, Cursor, Codex) to reduce context reprocessing costs and improve agent continuity.

HN technewssss 15d ago

Mistral, Europe's AI Darling, Fails FLI Safety Index

Future of Life Institute AI Safety Index shows Mistral AI scored lowest among major AI companies in safety evaluation.

HN mariusb16 15d ago

AI Agent Coding Comparison: The Rust Leap Year Challenge

Benchmark comparison of 5 AI agents (Claude, GPT, Gemini, Qwen, Hy3) on Leap Year coding task with performance analysis.

HN jabernal 15d ago

Vairfid – Identity and accountability layer for AI agents

Identity and accountability layer protocol for AI agents. Developer tool enabling auditability and traceability in agent systems.

HN cdrnsf 15d ago

Senator Warner Makes a First Foray into Agentic AI Regulation

Senator Warner proposes regulation framework for agentic AI systems and their autonomous decision-making capabilities.

HN tsuzuu 15d ago

I let an LLM rewrite the tool descriptions of 4 MCP servers – before/after data

Analysis of LLM tool selection using MCP server descriptions with reproducible data. Shows how agents pick tools and impact of description clarity.

HN ali_chherawalla 15d ago

Show HN: OGAC – Run reliable, compliant and governed AI for your enterprise

Enterprise platform combining frontier models with company data for governed, compliant AI with safety and observability.

HN L_u_u_6 15d ago

Show HN: UATC-Closed-loop VRAM control and dynamic data pruning for LLM training

UATC system for adaptive VRAM control during LLM fine-tuning on edge hardware using closed-loop feedback to prevent OOM crashes.

HN inigoerana 15d ago

DiscoMCP – Turn an unknown MCP into a reusable operational skill for AI agents

DiscoMCP tool teaches AI agents operational workflows and data safety rules for MCP servers, reducing agent errors on unfamiliar tools.

HN roman_t 15d ago

Show HN: LLM-mock – Record and replay OpenAI/Anthropic calls in pytest (v1.0)

pytest plugin for recording and replaying OpenAI/Anthropic API calls to eliminate test flakiness and costs. Open-source developer tool.

HN RatulMaharaj 15d ago

Show HN: I built an agent framework where the agent is just one file

Looped AF: Agent framework for building event-driven AI agents from single configuration file, containerized deployment.

HN vintagedave 15d ago

Tell HN: The Codex App is replaced by ChatGPT

OpenAI deprecates standalone Codex app; functionality migrated to ChatGPT product.

HN Rizzist 15d ago

Show HN: Call to Control AI Agents via the Web

Open source web interface for controlling and monitoring coding AI agents remotely with Twilio integration support.

HN Alien1Being 15d ago

Chinese AI models are gaining ground with U.S. companies

Chinese LLMs like DeepSeek gaining adoption among U.S. companies due to lower costs and narrowing performance gaps with OpenAI and Anthropic.

HN yiweileng 15d ago

DolphinDB v3.00.6 and v2.00.19: Introducing DolphinX for Enterprise AI Agents

DolphinDB releases v3.00.6 and v2.00.19 with DolphinX for enterprise AI agents.

HN grokkedit 15d ago

Show HN: GGUFun, play snake and a simple maze on Ollama using hand crafted GGUFs

Demonstrates hand-crafted GGUF models for deterministic outputs, building playable games on Ollama. ML model format experimentation.

HN himata4113 15d ago

Muse Spark 1.1: Meta gains 8 Intelligence Index points in three months

Meta's Muse Spark 1.1 LLM model scores 51 on Artificial Analysis Intelligence Index, showing 8-point improvement in three months with improved efficiency.

HN manchoz 15d ago

Device Context Protocol – Bridge LLM Agents to Physical Devices

Device Context Protocol v0.3 specification enabling LLM agents to safely control physical devices including microcontrollers with compact wire format and MCP compatibility.

HN tamnd 15d ago

Tell HN: One SWE-bench-Live task: $47 Opus failed, $1.46 GPT-5.6 passed

Benchmark comparison of 7 LLMs on SWE-bench-Live task dynaconf__dynaconf-1225 with cost analysis, reproducible commands, and lessons learned from model performance variations.

HN ttflee 15d ago

Interactive Jacobian-Lens visualizer and live steerer for GGUF models

Open source GGUF-native implementation of Jacobian Lens from Anthropic for interactive LLM visualization, steering, and ablation with llama.cpp.

HN mrhustlex 15d ago

Show HN: TradingSpy-Loop based research and backtesting trading Agent

Open-source AI trading research platform with agent-based strategy generation, backtesting, and market analysis in Docker. Local-first architecture.

HN mnzralee 15d ago

The Harness Is Not the Model: How Far Scaffolding Takes a Weak LLM

Empirical study showing weak LLM models wrapped in retrieval, tools, and verification achieve frontier performance. Tests five harness components against 30 sources.

HN electricant 16d ago

Llambda.lisp an LLM Engine in Pure Common Lisp

Pure Common Lisp LLM inference engine supporting quantized models from GGUF files with AVX2 acceleration and multi-threading. Zero external dependencies.

HN thierryzoller 16d ago

Dunning-Kruger After AI: The Gap That No Longer Closes

Essay on how AI tools affect the Dunning-Kruger effect and capability assessment in organizations.

HN tomatoes2026 16d ago

Show HN: AgentTransfer now lets agents self-sign-up and get web app hosting

Platform enabling AI agents to self-register, obtain hosting, and coordinate work through a simple API with discovery directory.

HN yashmitb 16d ago

Rewisp – An ambient screen memory for macOS that never writes images to disk

macOS app using on-device LLM to index screen text and answer queries about past activity without storing images.

HN rfv6723 16d ago

The Unreasonable Effectiveness of LLMs in Mathematics

Research on LLM effectiveness in mathematical problem-solving; minimal content provided.

HN Jonathanfishner 16d ago

MCP gateway can't see most of what your agent does

MCP gateways for agent security have blind spots; agents using external tools like curl, gh, and Python scripts bypass monitoring, creating ungoverned traffic.

HN insightits 16d ago

Same agent tasks, 76% fewer LLM calls – we moved semantic cache inside the graph

ChorusGraph: native agent runtime with semantic caching, retrieves 76% fewer LLM calls than baselines; includes PrismRAG and auditable memory with enterprise hardening.

HN ianberdin 16d ago

Fable 5 on Playcode. As well as Sol, Grok 4.5 and GLM 5.2

Anthropic's Claude Fable 5 model now available in Playcode's AI website builder for front-end code generation.

HN mooreds 16d ago

Tend turns intent into local, reviewable feeds you can inspect, steer, and teach

Tend is experimental software that turns intent into local, reviewable feeds for working with Codex, allowing inspection and steering of AI outputs.

Ax Merkouris Papamichail, Konstantinos Varsos, Giorgos Flouris, Jo\~ao Marques-Silva 16d ago

Interval Certifications for Multilayered Perceptrons via Lattice Traversal

Theoretical framework for adversarial robustness of neural networks via lattice traversal, reducing robustness certification to interval analysis.

Ax Vanessa Figueiredo, Wilter Franceschi 16d ago

CogniConsole: Externalizing Inference-Time Control as a Formal Abstraction for Reliable LLM Interactions

CogniConsole: Architecture externalizing inference-time control for LLM systems via structured interface combining programmatic coordination with formal abstractions.