7/5/2026
Ask HN: How Do You "Not Write Any Code by Hand" with a Token Budget?
Discussion on balancing AI-first development mandates with strict token budgets when building agentic systems.
7/5/2026
I built a weather station that runs machine learning to forecast weather
Weather forecasting system using ML ensembles with KNN and neural networks trained on historical data; sensor logging and hourly predictions.
7/5/2026
Show HN: Aletheia – The Uncertainty Loop Agent for Claude Code and Codex
Aletheia is an open-source loop engineering agent that maintains uncertainty views and uses evidence to adjust confidence, compatible with Claude and Codex.
7/5/2026
Engram – persistent memory for AI agents, in-process, no cloud
Engram provides persistent in-process memory for AI agents without cloud infrastructure or setup overhead.
7/5/2026
Istota – a multi-user AI agent and personal OS
Istota is a self-hosted personal OS integrating AI agents with Nextcloud, featuring multi-room chat, RSS, health dashboards, and persistent memory.
7/5/2026
Show HN: Goldseam – heal broken Cypress selectors with a local LLM
Goldseam: open-source tool that uses local LLMs to automatically repair broken Cypress test selectors, with human review before applying fixes.
7/5/2026
Domain experts: All human experts into AI agents
Open-source library of role definitions for AI agents to act as domain experts (CFO, engineer, etc), crowdsourced from practitioners.
7/5/2026
Building real agents from real context graphs [video]
Video on building AI agents using context graphs. Title and format only, no transcript or detailed description.
7/5/2026
Learning to Replicate Expert Judgment in Financial Tasks
Research on training LLMs to replicate expert financial judgment and investment decisions from limited public information.
7/5/2026
Show HN: Glyph a blockchain where the proof-of-work is neural network inference
Glyph: blockchain using neural network inference as proof-of-work instead of SHA-256, with open-source implementation on GitHub.
7/5/2026
NHS to use AI on its app to direct patients to appropriate services
NHS England implementing AI triage tool on patient app to direct users to appropriate healthcare services over next year.
7/5/2026
Harbor: An MCP gateway that connects AI clients to back end APIs via tools
Harbor: production-grade Model Context Protocol gateway connecting AI clients to backend APIs without modification, handling authentication and discovery.
7/5/2026
GCP offers agentic perimeter guardrails
Google Cloud VPC Service Controls for establishing network security boundaries around AI agents in production environments.
7/5/2026
Show HN: microide, a 100% vibecoded IDE that LLM agents can drive
100% LLM-generated IDE focused on performance and privacy. No telemetry or networking. Designed for AI agents to drive with code generation supervision.
7/5/2026
Self-hosted email client with an AI agent, running on Cloudflare Workers
Self-hosted email client with AI agent running on Cloudflare Workers. Uses Durable Objects, SQLite, R2 storage. AI agent reads inbox and searches conversations.
7/5/2026
Show HN: A Mac switchboard for Claude Cowork skills
Open-source Mac app providing a switchboard UI to discover and enable Claude skills for non-technical users in Claude Cowork.
7/5/2026
Show HN: Agent Torrent, a BitTorrent inspired mesh for idle coding agents
AgentTorrent is a peer-to-peer mesh network inspired by BitTorrent that lets coding agents share idle capacity, supporting Claude and local LLMs via Python protocol.
7/5/2026
Show HN: TraceTree – Sandboxed behavioral analysis for NPM and PyPI packages
TraceTree is a sandboxed behavioral analysis tool for NPM and PyPI packages using signature-based detection with syscall monitoring for security in agentic workflows.
7/4/2026
Show HN: A free, no-ad, no-account, no-server iOS calorie budgeting/diet app
VSCode extension for Common Lisp development with REPL support and configuration for multiple Lisp implementations.
7/4/2026
Show HN: Visual Knowledge Canvas for Learners
LLM-powered visual knowledge management tool for learning. Aggregates sources, answers questions with context, branches conversations, analyzes videos and research papers.
7/4/2026
Scientists decry conference's use of hidden prompts to snare AI peer reviews
Conference adds hidden prompts to catch peer reviewers using AI for reviews. Raises trust concerns about detection methods.
7/4/2026
Show HN: Qpilot – AI agent runs plain-text manual test cases in a real browser
Qpilot: AI agent that executes plain-text manual test cases in real browser. Supports Claude, OpenAI, Qwen, Ollama, vLLM.
7/4/2026
Show HN: An MCP server that gives your AI assistant write access to /etc./hosts
LockIn: MCP server enabling AI agents to block distracting websites via system hosts file. Works with Claude, ChatGPT, Cursor.
7/4/2026
Intent-addressable code for AI coding agents
Causari: Tool recording every AI agent action on codebases—prompts, model used, files read, reasoning. Intent-addressable code.
7/4/2026
Show HN: Autosynth – generating synthetic data with strong/weak model filtering
LLM-based synthetic data generation tool using weak/strong model filtering, inspired by Meta FAIR's work, with domain-agnostic plugin architecture.
7/4/2026
Show HN: I built LibreYOLO, an MIT-licensed YOLO library
Open-source MIT-licensed YOLO computer vision library built to provide an alternative to proprietary real-time vision solutions.
7/4/2026
Godot cracks down on "vibe-coded" pull requests
Godot game engine adopts policy restricting AI-generated pull requests, citing quality concerns and vague understanding from AI-assisted contributors.
7/4/2026
How AI models would vote in Sweden
Analysis of how major AI chatbots voted in Swedish election test, noting system prompts and safety layers affect outputs beyond base model.
7/4/2026
Varela: Neuromancer-inspired 3D self-mutating coding harness
Varela is a self-mutating AI agent harness in JavaScript with WASM sandbox, multi-agent orchestration, and model routing to Anthropic/OpenAI/OpenRouter. Rewrites own code while running, learns from wins/failures.
7/4/2026
Show HN: High performance command guard and policy enforcement for Agents in Zig
Orca: Safety layer for autonomous AI agents in Zig, enforcing policies to prevent dangerous operations like data deletion and credential leaks.
7/4/2026
Show HN: Crew – Let Claude Code agents talk to each other
Crew: Developer tool enabling multiple Claude Code agents to collaborate on same codebase by sharing context and messaging without branching.
7/4/2026
I am dreading our LLM-written incident report future
Opinion on risks of LLM-generated incident reports. Discusses potential and concerns without technical depth.
7/4/2026
Agentic coding notes from Galapogos Island
Personal experience using AI agents for coding tasks, discussing agent behavior and debugging approaches with GPT-5. Anecdotal insights.
7/4/2026
FEOM – Windows GUI automation at 8ms, no GPU needed
FEOM: Open-source Windows GUI automation framework for AI agents with 8ms latency, eliminating vision token costs for deterministic OS operations.
7/4/2026
Show HN: SmolSignal – signal copilot for Flipper Zero files
AI copilot for Flipper Zero radio capture files. Translates captures to plain English, classifies safety risks, supports local/cloud models.
7/4/2026
Show HN: Earshot, a homebrew Claude Tag
Slack bot using Claude that handles incident management. Reads context, checks trackers, and decides escalation without prompting.
7/4/2026
Show HN: Local privacy-first Microsoft Recall alternative with Gemma 4
Local privacy-first screen recording tool using Gemma 4 vision model. Searchable screenshot timeline with conversational chat interface. Data stays on-device.
7/4/2026
BrickSlider – TypeScript-first, Tailwind-first carousel
TypeScript carousel component with zero runtime dependencies. Web Animations API with plugin system for extensibility.
7/4/2026
Show HN: Fortress – open-source Chromium that keeps browser agents unblocked
Open-source Chromium variant that prevents browser fingerprint detection for scraper and agent automation. Modifies C++ fingerprint handling.
7/4/2026
Context graphs: how AI agents remember why decisions were made
Explains context graphs as agent memory for storing decisions. Shows how agents track policy overrides and decision rationale for accountability.
7/4/2026
Agent Empire: A interactive game to get started on what agent pattern to chose
Agent Empire: Interactive RTS game teaching six agentic patterns through incident response scenarios. Players dispatch pattern-agents, manage tokens, and learn real-world agent system design.
7/4/2026
China's ByteDance discovers new scaling law that could sustain AI boom
ByteDance researchers discover AI agents can double learning speed every three months through real-world task interaction, revealing new scaling laws for model improvement.
7/4/2026
Show HN: Gavio: open-source interceptor pipeline for production LLM applications
Gavio: Open-source AI gateway with composable interceptor pipelines for LLM apps. Handles PII protection, cost tracking, audit logging, retries across Python, Java, JavaScript.
7/4/2026
Show HN: Papa – open-source Hemingway-style readability linting for Markdown
Papa: open-source CLI and Python library for Markdown readability linting with Hemingway-style rules, JSON output, and HTML reports.
7/4/2026
How HN: Billy – a self-hosted AI assistant that lives in your Telegram
Billy: Self-hosted Telegram AI assistant orchestrating multiple models (Ollama, Claude) for email, CRM, web research, document generation, and skill learning.
7/3/2026
Give your AI agent its own email inbox – MailKite
MailKite provides AI agents with real email inboxes to receive and reply to messages autonomously, with working code and security guidance.
7/3/2026
Save Claude Code Tokens with Smart Routing
Brick is a mixture-of-models routing gateway that optimizes LLM costs by routing prompts to appropriate models, supporting Claude and other backends.
7/3/2026
Dispersion loss counteracts embedding condensation in small language models
Research paper analyzing dispersion loss in small language models during pre-training, showing it emerges at initialization and is not resolved by distillation.
7/3/2026
Leanstral 1.5: Proof Abundance for All
Leanstral 1.5 is a 6B-parameter open-source model achieving state-of-the-art formal verification through reinforcement learning, designed for agentic proof engineering.
7/3/2026
Show HN: Scout.kids – Agents that find activies for families
Scout uses AI agents to crawl the web and intelligently recommend family activities based on child interests and location, replacing manual blog aggregation.