Articles, releases and code from Hacker News, Reddit, GitHub and the people building RPA, workflow automation and AI agents — plus what the community pushed to the top today.
AIRivet has launched an open-source orchestrator designed to scale, manage, and run high-density agentic workloads and state on top of S3.
AIGauge released an open-source CLI that runs coding agents in Docker to automatically test, detect friction in, and optimize developer documentation, skills, and tools.
AIShotoku released a local-first authorization layer that lets developers enforce YAML-defined budgets, approvals, and audit trails on AI agent spending and tool calls.
AIA contractor connected a personal Claude coding agent to Pentagon systems, highlighting urgent risks of automated administrative credentials leaking network maps to unauthorized external sites.
AIWebsiteBench introduces a benchmark to evaluate how effectively AI agents can reverse-engineer and rebuild full-stack websites purely through autonomous browser exploration.
AIPrax launched an MCP server that allows AI agents to directly screen cryptocurrency wallets for sanctions, AML risks, and exploits within automated workflows.
AIEtherPK launched an MCP server and Kanban interface that lets coding agents read markdown tasks, access attached assets, and update progress directly within user notes.
AIOpenRing launched an open-source, self-hosted voice AI agent platform compatible with Vapi, letting developers run local or cloud-based phone automations without per-minute platform fees.
AIAmigent introduces an MCP server and remote-desktop tool enabling AI agents to control, automate tasks, and navigate interfaces on retro Commodore Amiga computers over slow networks.
AIA benchmark of five decision models on Korean comprehension tests provides speed, cost, and accuracy metrics to help developers select models for Korean-language automated workflows.
AIQuanterLab launched a quantitative research platform with MCP connectors, allowing AI agents to construct visual block-diagram trading strategies and automate paper trading workflows.
AIAgentFleet open-sourced an orchestration layer that coordinates existing coding agents across multiple devices, featuring sub-second failover, zero task loss, and shared tamper-evident memory.
AISynaps released an open-source Rust agent runtime that enables developers to run autonomous background daemons with parallel orchestration directly from the terminal.
AICall4agents is an open-source Python library that prevents AI agents from re-executing approved tool calls after crashes across frameworks like LangGraph and Pydantic AI.
AINetflix developed autonomous AI agents to analyze raw profiler data and patch code, enabling teams to automate complex performance tuning and bottleneck detection at scale.
AIA QCon presentation outlines zero-trust security controls for production AI agents, shifting defenses to scoped tool permissions and audit-ready observability to prevent unintended database queries and data leaks.
AICamunda automated internal feedback triage using Camunda 8 and AI agents, demonstrating how to deduplicate support tasks in Asana and accelerate issue resolution.
AIAirlock provides runtime containment for multi-agent workflows, isolating downstream tainted actions upon fault detection while preserving unaffected executions instead of halting the entire system.
AIA new algorithm uses conformal prediction to accurately pre-allocate compute and bandwidth for unpredictable agentic workflows, cutting over-provisioned infrastructure waste by half while preventing mid-execution throttling.
AIResearchers introduced evidence-gated authority for LLM reviewers, allowing automation builders to prevent reviewer agents from corrupting valid decisions while still fixing actual errors.
AIOpenAI used an autonomous agent pipeline to patch 948 Linux vulnerabilities, while new open-weight models from Moonshot and Ricursive expand options for automated coding workflows.
AIResearchers released an open-source library of 33 modular agent skills, enabling developers to automate complex CAD, simulation, and manufacturing workflows using natural language interfaces.
AIGoogle Cloud API Gateway now supports real-time streaming, allowing builders to route live LLM outputs and multi-step agent interactions through standard authentication and security policies.
AIPylon outlined a framework for building reusable AI agent skills with human-in-the-loop approvals, letting teams automate repetitive support investigations and ticket routing workflows.
AIA new research framework helps developers make AI agents less brittle to prompt variations by systematically enriching context to ensure consistent, decision-relevant behavior.
AIDropbox acquired Tigris Data to launch DBX Agent Cloud, providing S3-compatible globally distributed storage designed to help developers run AI agents across persistent file sessions.
AINorthmeadow released its public alpha, offering an encrypted shared memory layer to help developers coordinate context securely between local AI agents and human users.
AIServices4Agents launched a pay-per-call marketplace offering utility APIs and infrastructure services that autonomous agents can consume and pay for using USDC micropayments.
AIBansat is an open-source framework and Claude skill that forces AI agents to plan architecture, security, and UI states before generating production code.
AIAlgoVesta released a hosted Model Context Protocol server that allows AI agents to execute trades across MetaTrader 5 and 16 crypto exchanges under server-side risk controls.
AIVetto released ScienceBench, a 50-task evaluation measuring autonomous data-analysis agents across scientific disciplines, showing top models fail over half the trials when handling real-world research anomalies.
AIMailbox.bot launched physical mailing addresses and print-and-mail services with API and MCP support, letting AI agents send, receive, and process physical postal mail and packages.
AIApplying Pareto frontier analysis to AI models reveals that only a few options optimize cost, speed, and intelligence, helping developers eliminate inefficient models from automated workflows.
AIRicursive Intelligence released Ripple-65B Preview, an open-weight model with a 1M context window, giving developers a compute-efficient option for running coding agents on local hardware.