3/6/2026
Show HN: Kaeso – infrastructure for connecting AI agents to real services
Kaeso infrastructure for connecting AI agents to real-world services, solving fragmented authentication, token management, and API integration challenges.
3/6/2026
Show HN: An OTLP observability plugin for OpenClaw AI agents in Grafana
OTLP observability plugin for OpenClaw AI agents in Grafana. Minimal content provided.
3/6/2026
Poet-X Enables Billion-Parameter LLM Training on Single H100 GPU
POET-X enables billion-parameter LLM pretraining on single H100 GPU through memory-efficient techniques. Builds on POET framework with significant computational resource reduction.
3/6/2026
Show HN: Kaeso: an OAuth hub for AI agents
Kaeso: OAuth hub for AI agents providing centralized secure integrations with external services. Addresses agent infrastructure challenge by standardizing Google, Slack, GitHub connections.
3/6/2026
Show HN: WebBridge turns any website into MCP tools by recording browser traffic
WebBridge converts websites into MCP tools by recording and analyzing browser traffic patterns. Enables AI agents to interact with web services without manual tool configuration.
3/6/2026
Show HN: MultiPowerAI – Trust and accountability infrastructure for AI agents
MultiPowerAI: trust and accountability layer for autonomous agents. Cryptographic identity, behavioral circuit breakers, audit trails, human approvals, skills marketplace.
3/6/2026
Show HN: Single-header C++ libraries for LLM APIs – zero deps beyond libcurl
Single-header C++ libraries for LLM APIs including streaming, semantic caching, cost estimation, retry logic, and structured output.
3/6/2026
Show HN: Sheila, an AI agent that replaced our accounting flow
Sheila: AI agent handling complete accounting workflow (invoicing, payments, bookkeeping) built with OpenCode at Soapbox. Production agent application.
3/6/2026
Show HN: Natural language queries for Prometheus Kafka metrics (StreamLens)
StreamLens: open-source tool adding natural language queries to Prometheus Kafka metrics via built-in AI chat panel for troubleshooting.
3/6/2026
45% of AI-Generated Code Has Security Vulnerabilities
Veracode study: 45% of LLM-generated code across Java, Python, JavaScript, C# contains security vulnerabilities. No improvement with model size or recency.
3/6/2026
Capability-Tiered AI Governance Architecture (CEGP)
CEGP v0.7: research architecture for capability-tiered AI governance with compute-linked enforcement protocols and distributed verification.
3/6/2026
Show HN: key-carousel - Key rotation for LLM agents
Key-carousel library for multi-key LLM agent key rotation and rate-limit management as in-process state machine.
3/6/2026
Show HN: OculOS – Give AI agents control of your desktop via MCP
OculOS: Rust daemon enabling AI agents (Claude, Cursor) to control desktop via semantic JSON API and MCP. Reads OS accessibility tree instead of OCR.
3/6/2026
Show HN: Speclint – OS spec linter for AI coding agents
Speclint: linter scoring GitHub issues (0-100) before AI agents work on them. Prevents agents from building poorly-specified features.
3/6/2026
Qwen3.5-35B – 16GB GPU – 100T/s with 120K context AND vision enabled
Configuration and benchmarks for running Qwen3.5-35B GGUF models on 16GB GPU with llama.cpp, 120K context.
3/6/2026
Ask HN: Has anyone built an autonomous AI operator for their side projects?
User's autonomous AI agent managing side projects end-to-end: blog publishing, Stripe monitoring, uptime checks, escalation logic. 30-minute heartbeat loop.
3/6/2026
Autonomous AI Newsroom
SimpleNews.ai: autonomous AI newsroom publishing daily AI news summaries covering models, companies, research, policy, tools, funding.
3/6/2026
Airtable: Rewriting Our Database in Rust
Airtable rewriting database backend in Rust to handle Fortune 100 scale usage and AI-era demands on infrastructure.
3/6/2026
Show HN: PlateSpinner – A Kanban board that orchestrates AI coding agents
PlateSpinner is a local web app that orchestrates AI coding agents via kanban board. Routes Claude Code/Codex/Gemini through propose-plan-execute workflow with branch management.
3/6/2026
Show HN: Auto-Co – 14 AI agents that run a startup autonomously (open source)
Auto-Co open-source system runs 14 AI agents with expert personas to operate a startup autonomously. Uses Claude Code CLI, markdown consensus files, Telegram escalation. 12 cycles demonstrated.
3/6/2026
Research Shows Models Know Answers Before Finishing Chain-of-Thought Reasoning
Research paper reveals reasoning models generate unnecessary chain-of-thought tokens after forming answers. Activation probing can decode answers earlier, enabling up to 80% token reduction.
3/6/2026
Open-source LLM router with Thompson Sampling and energy-aware routing
Open-source LLM router using Thompson Sampling and energy-aware routing, MIT-licensed alternative to closed-source competitors.
3/6/2026
Aigate – An OS-level sandbox for AI coding agents (Claude, Cursor)
Aigate: OS-level sandbox (kernel-enforced) for AI coding agents (Claude, Cursor, Copilot). File, command, network isolation for regulated environments.
3/6/2026
Show HN: How to Catch Documentation Drift with Claude Code and GitHub Actions
GitHub Actions workflow using Claude Code to detect and flag documentation drift during code iteration.
3/6/2026
Runtime observability and policy enforcement for AI coding agents
Oculi intercepts and enforces security policies on AI agent tool calls in IDEs. Integrates with Claude Code, Cursor, and Windsurf via YAML rule configuration.
3/6/2026
OpenPawz Conductor Protocol
Open-source protocol for optimizing AI workflow execution through parallelization and LLM-driven graph compilation for multi-agent systems.
3/6/2026
Show HN: Verified n8n community node for Anytype
Verified n8n community node providing full CRUD operations and API access for Anytype local-first encrypted note-taking app.
3/6/2026
Show HN: GPT-5.4 is interesting for one boring reason: fewer retries
Observation that GPT-5.4 reduces retry rates on real coding tasks and tool-heavy workflows compared to earlier models.
3/6/2026
Show HN: Graph-Oriented Generation – Beating RAG for Codebases by 89%
Deterministic graph engine that offloads reasoning from LLMs, reducing token usage by 89%. Uses 0.8B model with white paper and reproducible benchmark for codebase analysis.
3/6/2026
3W for In-Browser AI: WebLLM and WASM and WebWorkers
In-browser AI agents using WebLLM, WASM, and WebWorkers for local model inference and agent logic without API calls.
3/6/2026
Show HN: Open-source multi-model code review council (BYOK, free tier)
Multi-model code review system using multiple AI models in parallel. Lead model synthesizes independent reviews into structured categories highlighting consensus and disagreement.
3/6/2026
Show HN: Contexa – Git-inspired context management for LLM agents
Open-source Git-inspired context management system for LLM agents. Provides snapshots, branching, and history tracking for reproducible agent behavior and experimentation.
3/6/2026
Show HN: Hydra – Real-time ops dashboard for developers running AI agents
Hydra macOS dashboard for monitoring AI agents, LLMs, and dev servers simultaneously. Shows process status, resource usage, and port mappings.
3/6/2026
Agor: Figma but for AI Coding Assistants
Multiplayer spatial canvas for orchestrating multiple AI coding assistants with git worktrees, conversation tracking, and real-time collaboration visualization.
3/6/2026
Teaching LLMs to reason like Bayesians
Google Research paper on training LLMs to perform Bayesian reasoning by mimicking optimal Bayesian models. Enables better probabilistic reasoning for LLM-based agents interacting with environments.
3/6/2026
Google Workspace CLI can connect AI Agents to your cloud
Google Workspace CLI bundles cloud APIs to enable AI agents to connect to Workspace services. Not yet official product.
3/6/2026
Netflix Acquires Ben Affleck's AI Filmmaker Tools Startup InterPositive
Netflix acquires InterPositive, Ben Affleck's startup making AI-powered filmmaker tools. 16-person engineering team joins Netflix.
3/6/2026
Show HN: Sandclaw – A Safety First Agent
Human-in-the-loop safety framework for AI agents with three-part architecture: Muteworker, Gatekeeper, and Confidante for isolation and control.
3/6/2026
ATK: A Git-backed CLI for managing AI dev tools
ATK CLI tool manages MCP servers and AI services across multiple coding agents. Git-backed plugin manager for Claude, Codex, Gemini integration.
3/6/2026
New Research Reassesses the Value of Agents.md Files for AI Coding
ETH Zurich research paper evaluates AGENTS.md files for AI coding. Finds documentation often hinders rather than helps agents; recommends selective use.
3/6/2026
Cursor Goes to War for AI Coding Dominance
News coverage of Cursor IDE expanding AI coding capabilities. Market/competitive analysis rather than technical research.
3/6/2026
Show HN: Aporium AI – a study app that refuses to give you answers
Aporium AI tutoring platform using Socratic method for student learning. LLM-based educational application with question-guided approach.
3/6/2026
Show HN: Rental Property Deal Analyzer – 20 metrics, deal scoring, AI analysis
Open-source rental property investment calculator using FastAPI and vanilla JS. AI analysis feature but domain is real estate, not AI/ML focus.
3/6/2026
Nested Training for Mutual Adaptation in Human-AI Teaming
Research on nested training approach for human-AI collaboration and mutual adaptation.
3/6/2026
Show HN: Tri·TFM Lens – 5-axis quality evaluation for ChatGPT/Gemini responses
Chrome extension evaluating ChatGPT and Gemini responses across 5 dimensions: emotion, fact, narrative, depth, and bias with quality scoring.
3/6/2026
Show HN: Claudine – A Kanban board for your Claude Code and Codex conversations
VS Code extension that creates a live kanban board from Claude Code and Codex agent session files with auto-detection and real-time sync.
3/6/2026
Show HN: I built the first scripting language for multiplayer game dev
Custom scripting language for multiplayer game development that abstracts networking details, designed to work well with AI code generation tools.
3/6/2026
Show HN: CloakPipe – Rust privacy proxy for LLM APIs with pseudonymization
Rust proxy for LLM APIs that detects sensitive entities, replaces them with consistent pseudonyms, and rehydrates responses. Preserves semantic structure for embeddings while protecting data privacy.
3/6/2026
Show HN: Mog, a programming language for AI agents
Mog: statically-typed programming language for AI agent plugins with capability system for enforcing permissions and I/O control.
3/6/2026
From vibes to data: measuring how LLMs attend to your prompt, layer by layer
Open-source mechanistic interpretability toolkit for analyzing LLM token attention and logit projections across layers with visualizations. Model-agnostic analysis.