Tag
Agent prompts, skills, and MCP configurations are increasingly treated as serious infrastructure, requiring a single source of truth, version locking, and drift detection—spawning a new wave of configuration management tooling for AI agents.
A developer promotes Decoy (decoys.me), a product that gives AI agents scoped, revocable task-specific identities instead of direct access to primary accounts, featuring client-side encryption and a user-owned knowledge graph for cross-agent context sharing.
Adhyyan Sekhsaria open-sourced a specification for fully agent-operated git repositories, arguing that every cloud agent should support this primitive for handling merge conflicts, observability, and scale beyond local files.
Nous Research lays out a manifesto for user-owned AI agents: support for smarter, cheaper, and local models, model switching mid-conversation, offline execution, full transparency into prompts, tools, and reasoning, and the ability to read, modify, and export the agent's code. The message centers on sovereignty, control, and portability over locked-down proprietary agents.
Garry Tan responds to a quote from a16z and OpenRouter co-founder Alex Atallah arguing that as everyone builds the same agent primitives (loops, memory, connectors, sandboxes), the industry will eventually converge on a single correct agent operating system.
The article describes the MCP Gateway architecture Uber built to unify MCP tools used independently by hundreds of teams: a control-plane Registry plus a data-plane Proxy gateway, combined with mechanisms like AutoCrawler for automatic discovery, Omni MCP as a unified entry point, Response Projection, and Code Mode. Together they support 800+ MCP servers and 5,000+ tools under one umbrella.
This post reflects on how agent development has mainly been an iterative process of patching weaknesses—adding retrieval, state management, retries, permissions, and evaluation—while noting the ongoing tension between offloading capabilities to supporting systems versus absorbing them back into the models themselves.
CanadaHonk proposes a JavaScript-focused sandbox product that accepts untrusted code in and returns output out, targeting agent/user code execution with isolation better than Cloudflare Workers at roughly 100x lower cost than dynamic workers.
Omnara, a Y Combinator-backed open-source alternative to Claude Managed Agents, launches with one API for deploying and running agents in production across any model, machine, or cloud, handling sessions, sandboxes, secrets, RBAC, and observability.
OpenHands highlights Build.One as an example of agent infrastructure, where users describe a business app and agents assemble it from IT-approved components with existing permissions and controls, powered by the OpenHands agent layer.
Cua (trycua) announced the free, source-available launch of Cua Spaces on macOS — agent-ready virtual desktops built on Cua Driver and the Lume virtualization stack, featuring Teleport session bundling, the RCDP streaming protocol with per-agent cursors, and direct-agent connectivity via the Cua MCP.
The author compares AgentWS workspace infrastructure against protected Git worktrees for multi-agent file updates, showing identical outcomes but far less custom code with AgentWS. The post highlights ACLs, recovery after worker failures, and conflict handling as requirements beyond plain Git.
Aweb 是一个面向 AI 智能体的开源通信基础设施,为智能体提供稳定身份、持久化邮件/聊天以及跨会话、跨运行时、跨机器的唤醒事件,并支持服务器联邦与自托管。
Google is donating its Agent Substrate project to CNCF, marking a potential turning point in the standardization of agent infrastructure such as sandboxes, tool calling, and context lifecycle management across agentic systems.
E2B introduces E2B Embed, an open-source (Apache-2.0) way to package the E2B Runtime and run Firecracker-based sandboxes inside a customer's own environment, aimed at regulated industries and government clients requiring data residency. It offers Docker Compose, Terraform (AWS/GCP), and Kubernetes deployment options with the same Python/JavaScript SDKs, a dashboard, and OpenTelemetry export.
Orthogonal launched a service that helps always-on AI agents avoid wasting paid tool calls by checking usage after ambiguous timeouts or truncated responses before retrying, letting agents autonomously find, use, and pay for data services from many companies. The launch was highlighted alongside OpenAI's Dots integration for autonomous lead-generation pipelines.
Kent C. Dodds promotes Kody, a software platform that gives AI agents shared primitives for memory, secrets, packages, triggers, and connections, so you don't have to re-configure your agent setup across different hosts like Cursor, Claude Code, and ChatGPT.
CoreSpeed launches as an MCP-based agent operating system that connects apps, accounts, and shared or private memory to Claude Code, Codex, Cursor and other MCP agents, with built-in web search, social research, media generation, and agent governance tools like budgets, activity logs, and Smart Approval.
Warp introduces Warp Factories, an out-of-the-box system designed to simplify building and operating AI software factories by providing an infrastructure layer for deploying AI agents, targeting smaller companies.
Agent Substrate is a Google open-source runtime that enables high-density lifecycle management for large-scale AI agent deployments, multiplexing many stateful agents onto fewer physical workers via Kubernetes and microVM/gVisor sandboxes with sub-second suspend/resume.