Architecture notes, launch updates, and hard-won details from building runtime infrastructure for agents that ship, earn, and prove their work.
`a2a use <agent>` turns any published agent into typed terminal commands — real flags from the card's JSON Schema, @file and stdin support, and a setup handshake for keys and config.
`a2a dev` keeps the local inner loop but moves the runtime to the cloud: edit locally, run your agent on its scale-to-zero dev box, hot-reload on save, and share it on a public URL.
EU AI Act enforcement lands 2 August 2026. Today we ship Article 12 decision records, floored retention, risk classification, and one-click evidence packs - as a view over receipts your agents already sign, not a second store.
A2A Cloud Academy is live: four free, in-depth courses — your first deployed agent, the skill runtime, production operations, and frontends/OpenAPI/MCP — 53 lessons grounded in the shipping platform.
You built a CrewAI crew. Now it needs a database, tools, an API, and proof it ran safely. On a2a cloud all four are provisioned with the agent on the first deploy. Here's the whole path.
A trace log is something you're asked to trust. A signed receipt is something you can verify. For AI agents acting on their own, that difference decides whether your audit trail survives a real dispute — or an auditor.
a2a ssh <agent> drops you into a throwaway cloud box with Node, Python, a2a-pack, and your agent's repo already loaded — over your existing ingress, VS Code-ready, and scaled to zero when idle.
We don't run our go-to-market with a stack of SaaS tools. We run it with an agent deployed on a2a cloud — the same platform we sell. Every move it makes leaves a signed receipt. Here's one.
You wrote a LangGraph graph. Now it needs a database, tools, an API, and proof it ran safely. On a2a cloud all four are provisioned with the agent on the first deploy. Here's the whole path.
Most stacks only offer agents one DB access pattern: a single connection string with full access to prod. Here's the safer model — scoped per-agent grants, human approval on risky steps, and a signed receipt for every action.
A high-level look at how a2a cloud combines signed grants, bounded workspaces, microVM sandboxes, and scoped source access so agents can act without inheriting unlimited authority.
A technical look at a schema-aware semantic memory system where Pydantic models, tool calls, agent traces, vector search, graph edges, and feedback loops become trainable memory.
Modern cloud platforms were designed for deterministic microservices. Autonomous AI agents need stronger isolation, agent-native networking, and auditable decision trails.
A look inside the A2A OpenAPI generator: how it reads a spec, turns routes into tools, wires auth and egress, and deploys an editable agent.
Every agent is software. Giving each agent its own repo makes it easier to understand, edit, review, deploy, and improve without stepping on unrelated work.
Use the A2A Pack template to scaffold a typed agent, run it locally, generate OpenAPI and TypeScript clients, and deploy it to a2a cloud.
Knative Services are controller-owned resources, so Argo CD needs health checks, diff ignores, and careful sync settings to avoid fighting the platform.
a2a cloud is making Postgres a first-class agent resource: declared by agents, governed by the platform, durable at rest, and able to suspend compute when idle.
A2A Cloud's Kernel turns agents from isolated tools into participants in bounded, replayable simulations: markets, tournaments, cohorts, arenas, resource competitions, and future agent economies.
If Part 1 was the big idea, Part 2 is the backstage tour. A meta-agent is not one giant chatbot. It is a smart coordinator that keeps a goal in view, brings in the right specialist agents, remembers what matters, and keeps moving until the job is done.
Building an agent is only step one. Agent Studio is A2A Cloud's loop for creating an agent, checking whether it works, improving it, and handing back a clear report instead of a mystery result.
Describe an agent. Get a live URL. Agent-Builder is the meta-agent inside A2A Cloud that scaffolds, writes, sandboxes, deploys, and verifies new agents — powered by DeepAgents and the platform's own runtime.
Real agents do work that takes minutes. They ask questions. They request more scope. They write files mid-task. A2A Cloud's streaming protocol turns every one of those into a live UI event — so users stay in the loop, not in the dark.
Multi-agent systems break when one agent has to trust another with everything. A2A Cloud ships Ed25519-signed grants that scope workspace access, budget, and runtime per handoff — so trust is bounded, every time.
A Python class, a YAML file, one CLI command. That's the path from "I have an idea" to a live, callable agent on A2A Cloud. Here's exactly how it works.
Containers aren't enough. A2A Cloud runs every agent workload inside libkrun microVMs with FUSE-backed workspaces, signed grants, and locked-down egress — so agents can do real work without ever touching the host.
How a2a cloud uses Argo-driven deployments to turn custom agents into reliable, observable, production-ready services teams can scale with absolute confidence.
A2A Cloud now ships the foundation for elastic HTTP agents, async KEDA workers, GitOps rollouts, scale-to-zero, and fleet-level growth. The platform shape just changed.
Describe an agent. Get a live URL. A2A Cloud fuses LiteLLM routing, scoped workspaces, microsandbox execution, and GitOps deploys into one ridiculously fast build-to-ship loop.
