llms.txt
Compare with other sites' llms.txt files →# Agentic AI Foundation (AAIF)
> Agentic AI Foundation (AAIF), a Linux Foundation project, is the neutral and open foundation built on transparency, collaboration, and standardization to advance the public interest in agentic AI innovation.
Markdown for agents: any page is also available as Markdown — prefix its path with /md/ (e.g. https://aaif.io/md/blog/<slug>), or request the canonical URL with the header `Accept: text/markdown`.
## Pages
- [About](https://aaif.io/about): Learn about the Agentic AI Foundation (AAIF) — our mission, governance, and role advancing open, standardized agentic AI in the public interest.
- [About Ambassador Program](https://aaif.io/learn-more-about-the-program): How the AAIF Ambassador Program works, what ambassadors do, and how to get involved.
- [Ambassador Program](https://aaif.io/ambassadors): An elite cohort of practitioners helping the world understand, use, and contribute to AAIF-hosted projects.
- [Awards](https://aaif.io/awards): Recognizing the individuals, teams, and organizations building the open agentic AI ecosystem.
- [Blog](https://aaif.io/blog): Guides, release notes and field reports on MCP, A2A, goose, AGENTS.md and agentgateway from the builders and maintainers at the Agentic AI Foundation.
- [Brand](https://aaif.io/brand): We welcome our community to use AAIF brand assets.
- [Case study](https://aaif.io/case-study): Case studies from projects hosted by the Agentic AI Foundation (AAIF).
- [Community Chapters](https://aaif.io/community-chapters): AAIF Community Chapters bring members together through local events and community-driven initiatives that connect, educate, and inspire collaboration.
- [Contact Us](https://aaif.io/contact): Contact us with questions related to Agentic AI Foundation (AAIF).
- [Events](https://aaif.io/events): Join Agentic AI Foundation events worldwide: AGNTCon + MCPCon, MCP Dev Summits and community meetups where builders shape open standards for agentic AI.
- [Governing Board](https://aaif.io/board): The governing board members of the AAIF.
- [Home](https://aaif.io/): Agentic AI Foundation (AAIF) is the neutral and open foundation built on transparency, collaboration, and standardization to advance the public interest in agentic AI innovation.
- [Maintainer Benefits](https://aaif.io/projects/maintainer-benefits): AAIF takes on the non-technical work of running an open source project, so maintainers can focus on the code and the community.
- [Members](https://aaif.io/members): Agentic AI Foundation members work together to further innovate, protect and standardize agentic AI.
- [News](https://aaif.io/news): The latest news and press from the Agentic AI Foundation (AAIF).
- [Projects](https://aaif.io/projects): AAIF is host to top open source
- [Staff](https://aaif.io/staff): The team supporting AAIF.
- [Submit a Project](https://aaif.io/submit-a-project): Propose an open source project to be hosted by the Agentic AI Foundation (AAIF).
- [Technical Committee](https://aaif.io/tc): The members of the AAIF Technical Steering Committee.
- [Videos](https://aaif.io/video): Watch exclusive interviews, event highlights, and deep dives into AAIF and agentic AI.
- [Working Groups](https://aaif.io/working-groups): AAIF.io Working Groups bring members together to collaborate on focused initiatives, share expertise, and drive practical outcomes across the AI ecosystem.
- [Working Groups calendar](https://aaif.io/working-groups/calendar): AAIF.io Working Groups bring members together to collaborate on focused initiatives, share expertise, and drive practical outcomes across the AI ecosystem.
## Blog
- [Give goose sandbox powers with OpenSandbox and MCP](https://aaif.io/blog/give-goose-sandbox-powers-with-opensandbox-and-mcp): Connect goose to OpenSandbox over MCP to create and use local sandboxes, run code safely, and understand the setup’s current security limitations.
- [Let the model suggest, let code decide: notes from AAIF Chennai Connect #2](https://aaif.io/blog/let-the-model-suggest-let-code-decide-notes-from-aaif-chennai-connect-2): A recap of AAIF Chennai Connect #2, covering practical approaches to multi-agent consistency, deterministic workflows, tool discovery, CI failure triage, and stateless MCP.
- [Most agentic work will become workflows](https://aaif.io/blog/most-agentic-work-will-become-workflows): Repeated agent interactions reveal useful workflows. Capturing those patterns as reusable skills and code helps improve consistency and preserve knowledge across tools.
- [Total cost of AI over value realized](https://aaif.io/blog/total-cost-of-ai-over-value-realized): J.R. Storment is Executive Director of the Tokenomics Foundation, a Linux Foundation project.
- [Four cool things from AGNTCon + MCPCon](https://aaif.io/blog/four-cool-things-from-agntcon-mcpcon): Four cool things from AGNTCon + MCPCon North America in San Jose: topographic MCP slides, Kids’ Day, a coding interface on a Playdate, and a phone booth full of questions.
- [We added TOON to our MCP server, then removed it](https://aaif.io/blog/we-added-toon-to-our-mcp-server-then-removed-it): After a 30-day trial of TOON, Apify switched its MCP server’s text output back to JSON, sharing lessons about token costs, traffic patterns, and testing real workloads.
- [Where voice agents fail in production: Part 2 - tools, recovery, and observability](https://aaif.io/blog/where-voice-agents-fail-in-production-part-2-tools-recovery-and-observability): Tool failures, uncertain retries, and handoffs in production voice agents, with a focus on recovery, evaluation, and observability.
- [Agents are applying for jobs: building an open protocol for hiring](https://aaif.io/blog/agents-are-applying-for-jobs-building-an-open-protocol-for-hiring): OJCP proposes a structured way for AI agents to discover jobs, apply, verify identity and consent, and track application status using MCP and WebMCP.
- [Isolating agent-generated code with MCP and Microsandbox](https://aaif.io/blog/isolating-agent-generated-code-with-mcp-and-microsandbox): How MCP and Microsandbox work together to isolate agent-generated code, with a practical example covering microVM execution, tool interfaces, resource limits, and network access.
- [Where voice agents fail in production: Part 1 - latency, turns, and interruptions](https://aaif.io/blog/where-voice-agents-fail-in-production-part-1-latency-turns-and-interruptions): Why do voice agents fail even when the transcript looks correct? Explore the challenges of latency, turn detection, interruptions, and conversation state in production.
- [The infrastructure factory: a foundation for agentic AI](https://aaif.io/blog/the-infrastructure-factory-a-foundation-for-agentic-ai): Agentic AI needs an infrastructure factory alongside software and AI factories, with capacity, policy, and costs managed through repeatable services.
- [Introducing Dapr Durable Execution for goose: AI agents that pick up where they left off](https://aaif.io/blog/introducing-dapr-durable-execution-for-goose): goose can execute complex tasks. Now, with Dapr Durable Execution, it can survive crashes and resume its work without starting over.
- [Announcing the winners of the first AAIF Annual Technical Excellence Awards](https://aaif.io/blog/winners-of-the-first-aaif-annual-technical-excellence-awards): Meet the first AAIF Annual Technical Excellence Awards winners, recognized for technical contributions, community building, and open agentic AI innovation, including the inaugural Keystone Award.
- [Training agents for real-world tasks with OpenEnv](https://aaif.io/blog/training-agents-for-real-world-tasks-with-openenv): Learn how OpenEnv connects task environments to agent training through a shared interface, with practical guidance on rewards, MCP tools, and deployment.
- [The Sandbox as an Agent Primitive](https://aaif.io/blog/the-sandbox-as-an-agent-primitive): Inside OpenSandbox’s approach to separating how agents execute work from where that work runs.
- [What Open Responses changes for AI developers](https://aaif.io/blog/what-open-responses-changes-for-ai-developers): AI developers should be able to support models from multiple providers without writing separate code to send requests and handle responses for each one.
- [Agent Skills: Sharing Team Expertise Across AI Agents](https://aaif.io/blog/agent-skills-sharing-team-expertise-across-ai-agents): Agent Skills is an open standard for packaging reusable instructions and supporting resources that teach AI agents how to perform specific tasks. Originally developed by Anthropic, it gives people a w
- [AI agents can't just walk in](https://aaif.io/blog/ai-agents-can-t-just-walk-in): AI agents can browse and shop, but websites decide what they can access. Shopify's MCP tiers show how permissions work and what happens when access fails.
- [James Ward thinks your MCP server could use some work](https://aaif.io/blog/james-ward-thinks-your-mcp-server-could-use-some-work): James Ward has a useful response to complaints that MCP wastes tokens: Show me the tools. If your agent needs 100 customer records and you gave it a tool that fetches exactly one, the ensuing parade o
- [Why "agent failed" doesn't tell you enough](https://aaif.io/blog/why-agent-failed-doesn-t-tell-you-enough): An agent asked to cancel an order and issue a refund can fail in several different ways. The order service might be temporarily unavailable, the customer might not have permission to cancel, or the or
- [How do you prove your AI governance works? Lessons from Toronto](https://aaif.io/blog/how-do-you-prove-your-ai-governance-works-lessons-from-toronto): Lessons from MCP Dev Summit Toronto on how enterprises are handling agent identity, access controls, enforcement, and audit evidence, with examples from NVIDIA, Accenture, and RBC.
- [The security architecture customer-facing agentic apps actually need](https://aaif.io/blog/the-security-architecture-customer-facing-agentic-apps-actually-need): A practical look at securing customer-facing agentic applications with scoped tokens, human approval gates, intent logging, and continuous access checks built on open standards.
- [Maintainer roundtable at AGNTCon + MCPCon EU: Composability, identity, and the cost of agents](https://aaif.io/blog/maintainer-roundtable-at-agntcon-mcpdev-summit-eu-composability-identity-and-the-cost-of-agents): At a maintainer roundtable during AGNTCon + MCPDev Summit EU in Amsterdam, moderator Nitish Tyagi of Gartner sat down with four people building the infrastructure underneath the agentic AI stack: Davi
- [I saw something in MCP. Then I spent 18 months building it.](https://aaif.io/blog/i-saw-something-in-mcp-then-i-spent-18-months-building-it): TL;DR: In March 2025, my first working MCP connection made me see a different future for network operations. Eighteen months of open-source experiments became NetClaw and a talk at MCP Dev Summit Toro
- [Welcome to Glass Town: building a software factory with a paper trail](https://aaif.io/blog/welcome-to-glass-town-building-a-software-factory-with-a-paper-trail): A hands-on look at building Glass Town, a transparent software factory using agent workers, persistent remote environments, Git-linked session history, and practical token controls.
- [Can you reproduce an agent run?](https://aaif.io/blog/can-you-reproduce-an-agent-run): Reproducing an agent run takes more than rerunning code. What execution history needs to capture, from model responses and tool results to state, permissions, and human decisions.
- [Notation for the human-agent symphony: how CHAP records human judgment alongside MCP](https://aaif.io/blog/notation-for-the-human-agent-symphony-how-chap-records-human-judgment-alongside-mcp): How the Collaborative Human-Agent Protocol gives approvals, overrides and handoffs a written form that anyone can replay, verify and improve.
- [The Agent Gets a Body](https://aaif.io/blog/the-agent-gets-a-body): Giving agents control of physical devices makes state, identity, observability and confirmed outcomes much more important, from hardware fleets to factories and autonomous vending machines.
- [Stateless MCP changes your security model: how to harden your MCP servers](https://aaif.io/blog/stateless-mcp-changes-your-security-model-how-to-harden-your-mcp-servers): A practical guide to hardening stateless MCP servers, covering state handles, requestState protection, routing headers, cache scope, retries, and authorization changes.
- [There Is No One Agentic Commerce Protocol](https://aaif.io/blog/there-is-no-one-agentic-commerce-protocol): Agentic commerce spans discovery, authorization, payment and fulfillment. UCP, AP2, x402 and A2A each address different parts of that journey rather than replacing one another.
- [Trace and analyze Goose sessions with OpenTelemetry, Jaeger and ClickHouse](https://aaif.io/blog/trace-and-analyze-goose-sessions-with-opentelemetry-jaeger-and-clickhouse): Use OpenTelemetry, Jaeger, and ClickHouse to trace Goose sessions, identify where latency comes from, inspect agent and tool activity, and persist telemetry for later analysis.
- [Five themes from AGNTCon + MCPCon Europe](https://aaif.io/blog/five-themes-from-agntcon-mcpcon-europe): TL;DR: The clearest message from Amsterdam was how much of agent engineering now depends on what sits around the model: protocols, identity, state, evaluation and workflows that keep human review mana
- [Agent identity and delegated access in MCP systems](https://aaif.io/blog/agent-identity-and-delegated-access-in-mcp-systems): A GitHub issue about a failing deployment has been assigned to you. You ask an internal agent to check the issue against the deployment runbook and post a status update. The agent works through tools
- [Before Agents Can Work Together, They Have to Find Each Other](https://aaif.io/blog/before-agents-can-work-together-they-have-to-find-each-other): Agent interoperability depends on discovery as well as communication. AI Catalog, AGNTCY, ARD, DNS and A2A are exploring ways for agents to find and evaluate capabilities dynamically.
- [AGENTS.md speaks UNIX, and you should too](https://aaif.io/blog/agents-md-speaks-unix-and-you-should-too): Unix gives coding agents the tools. AGENTS.md explains where to edit, how to check, and what to preserve. Lessons from my dotfiles and human review.
- [What agents look like when the data can't leave the building](https://aaif.io/blog/what-agents-look-like-when-the-data-cant-leave-the-building): Explore how agentic AI architectures must adapt when production data cannot leave secure environments like mainframes in banking, insurance, and regulated industries.
- [Docker's Sandbox Kit Spec puts Agent Permissions inside the container image](https://aaif.io/blog/dockers-sandbox-kit-spec-puts-an-agents-permissions-inside-the-container-image): Docker released V3 of the Sandbox Kit Spec, moving agent permissions directly inside OCI images to bridge the agentic container gap and standardize agent governance.
- [The Web Is Growing a Second Interface](https://aaif.io/blog/the-web-is-growing-a-second-interface): WebMCP, MCP Apps and A2UI are creating new ways for agents to interact with websites and services without relying entirely on browser automation built for humans.
- [Securing requestState with OpenBao](https://aaif.io/blog/securing-requeststate-with-openbao): A follow-up to our article on Designing RequestState for Multi-Round Trip Requests. All code in this post is from the companion repository `openbao-mcp-requeststate`.
- [AI governance in a regulated industry: How MoonPay brought MCP sprawl under control](https://aaif.io/blog/ai-governance-in-a-regulated-industry-how-moonpay-brought-mcp-sprawl-under-control): When AI agents can act inside internal systems, governance becomes an infrastructure problem: knowing which connections exist, what they can touch, and what they did.
- [Before an agent uses a model: verifying model custody on customer infrastructure](https://aaif.io/blog/before-an-agent-uses-a-model-verifying-model-custody-on-customer-infrastructure): An enterprise wants an agent to answer questions over sensitive internal records. It can keep the records on its own infrastructure, run inference locally, and give the agent access to approved tools.
- [MCP's big, open-world vision](https://aaif.io/blog/mcp-big-open-world-vision): Understanding why the protocol is for strangers and the critical role that extensions play
- [Your Agent Will Wake Up Eventually. Plan Accordingly.](https://aaif.io/blog/your-agent-will-wake-up-eventually-plan-accordingly): Long-running agents need more than persistence. They need durable state, recovery, idempotency and checkpoints that let work resume safely after failures.
- [402 payment required: what enterprise MCP servers owe the agents that pay them](https://aaif.io/blog/402-payment-required-what-enterprise-mcp-servers-owe-the-agents-that-pay-them): Two protocol releases seven months apart made a similar move, for related reasons, and the two are usually discussed as if unrelated.
- [Safety, messaging and open source agents in Shanghai](https://aaif.io/blog/safety-messaging-and-open-source-agents-in-shanghai): On September 5, members of the AAIF community gathered in Shanghai ahead of AGNTCon + MCPCon China to share work across open source agents, safety, messaging, edge computing, and agent infrastructure.
- [Introducing the First Open Source Vending Machine Benchmark](https://aaif.io/blog/introducing-the-first-open-source-vending-machine-benchmark): TLDR: We just launched Prosus Vending Bench, an open-source simulated vending business for testing tool-using agents. We packaged six machines, three locations and 22 MCP tools with Harbor, the framew
- [From supervised shells to purpose-built MCP tools](https://aaif.io/blog/from-supervised-shells-to-purpose-built-mcp-tools): Testing a narrow GPU metrics server on a live NVIDIA GPU.
- [Dex Still Thinks Long-Running Coding Agents Stink](https://aaif.io/blog/dex-still-thinks-long-running-coding-agents-stink): The models are better. The benchmarks are higher. Dex Horthy has fresh evidence that unattended coding agents still turn healthy codebases into radioactive spaghetti. He wants you to hear about it.
- [Implementing enterprise-managed MCP authorization in Keycloak](https://aaif.io/blog/implementing-enterprise-managed-mcp-authorization-in-keycloak): Enterprise MCP authorization gets complex at scale. This post introduces ID-JAG, an enterprise-managed authorization approach implemented in Keycloak, and previews upcoming EMA ecosystem support.
- [Envoy AI Gateway becomes Agent Router and joins AAIF](https://aaif.io/blog/agent-router-joins-aaif): The open source project, powered by Envoy, that gives developers a single integration for models and tools has a new name and a new home.
- [Agent Router: Powerful traffic handling for agent builders](https://aaif.io/blog/agent-router-powerful-traffic-handling-for-agent-builders): Envoy AI Gateway is now Agent Router and has joined the Agentic AI Foundation. This post is about what Agent Router is and what it does for the agent you are building.
- [Code Is Cheap. Proof Is the Bottleneck.](https://aaif.io/blog/code-is-cheap-proof-is-the-bottleneck): Coding agents can generate software faster than teams can verify it. This piece looks at the emerging bottleneck: proving code is correct, safe, reliable, and maintainable.
- [Agent memory: patterns, tradeoffs, open problems](https://aaif.io/blog/agent-memory-patterns-tradeoffs-open-problems): A practical guide to agent memory patterns, from conversation state and retrieval to shared memory, covering tradeoffs in freshness, provenance, access, deletion, and security.
- [MCP 2026 changes the rules](https://aaif.io/blog/mcp-2026-changes-the-rules): The latest MCP version marks a break with the standard we’ve known until now, particularly for deployments that need to scale across multiple serv