guides

Deploy, secure, and monetize AI agents.

Everything a2a cloud does, page by page — hosting any framework, the A2A and MCP protocols, scoped grants, signed receipts, compliance, and getting paid per call. The whole agent app, with proof.

describe it, ship it

Build & ship

AI agent builder

Describe an agent in one prompt; a crew builds, reviews, improves, and deploys it as a real governed service — not a demo.

Read the guide →

No-code agent builder

Plain English to a live agent — but you keep real code in a git repo. No-code to start, real code to keep.

Read the guide →

Agent self-correction

A bounded review→fix→re-verify loop: a reviewer finds flaws, an editor fixes them, every change leaves a signed receipt.

Read the guide →

Test your agent

Live smoke tests on a real preview URL plus a signed proof — evidence a run passed, not a green check that vanishes.

Read the guide →

AI agent CI/CD

Build→review→deploy as one governed pipeline: Gitea Actions, a review gate, Argo rollout, and a signed receipt per deploy.

Read the guide →

Debug a failed deploy

Build, image, Argo, and pod logs in the deployment timeline — secrets scrubbed. See which stage failed without cluster access.

Read the guide →

Deploy to production

The production checklist tutorials skip: runtime, database, auth, interface, cost, and audit — in one deploy.

Read the guide →

ship an agent

Deploy & host

Deploy an AI agent as an API

One command turns any agent into a REST/OpenAPI service — DB, auth, and audit included.

Read the guide →

Deploy an A2A agent

Host a Google A2A-protocol agent: agent card and endpoint out of the box, signed delegation between agents.

Read the guide →

Hosted MCP servers

Production MCP servers with auth, scoped grants, and signed audit built in — not bolted on.

Read the guide →

Deploy a LangGraph agent

Ship a LangGraph agent to production without LangSmith lock-in. Keep your graph, add the runtime.

Read the guide →

Deploy a CrewAI crew

Framework-agnostic host for CrewAI — adds MCP, gateway, and signed proof of every run.

Read the guide →

Deploy an AutoGen system

Host AutoGen multi-agent systems with A2A-signed handoffs between agents.

Read the guide →

Deploy DeepAgents

Run DeepAgents in production: workspace-backed files, sandboxed execute, scoped grants, signed receipts.

Read the guide →

Give your agent a frontend

Ship a React/Vite chat UI packed alongside the agent, hosted, with no separate deploy.

Read the guide →

Scale-to-zero hosting

Pay only when your agent runs. Scale-to-zero kills idle cost — the #1 hosting complaint.

Read the guide →

Managed Postgres per agent

Every agent gets its own isolated Postgres on deploy — an isolation and audit boundary, with pgvector.

Read the guide →

Sandbox agents in microVMs

libkrun microVM per run — a hardware boundary between agent code and host. Not a container.

Read the guide →

edit, run, iterate

Develop & iterate

AI agent dev environment

The whole a2a dev experience: cloud dev box, local mode, a dev console, managed resources, and hot reload to a live URL.

Read the guide →

Local AI agent development

Run the agent on your machine with `a2a dev --local` — Docker or a plain Python process, .env.local, hot reload.

Read the guide →

Run an agent locally

One command to start your agent on localhost with a live console — then the exact same runtime deploys.

Read the guide →

The agent inner loop

Edit → the agent reloads in ~2s → drive it in the console → repeat. No build, no redeploy, no CI wait.

Read the guide →

Hot reload for agents

Save a file, the agent reloads in ~2 seconds — locally or on the cloud dev box. No rebuild.

Read the guide →

The agent dev server

A real local HTTP server for your agent plus a `/_dev` console to invoke tools, set env, and stream results.

Read the guide →

Test an agent locally

Invoke each tool, set credentials, upload test inputs, and stream results in the `/_dev` console before deploy.

Read the guide →

Debug an agent locally

Run a plain Python process, attach a debugger, set breakpoints, and reproduce a failing tool call against real data.

Read the guide →

Dev/prod parity

The dev box runs the same runtime, image, and managed resources as prod — parity bugs surface before you deploy.

Read the guide →

Local dev Postgres

Declare a database once; get a real local Postgres in dev and managed Neon in prod. Same connection contract.

Read the guide →

Vector memory for agents

Declare vector memory once; real local Qdrant in dev, managed Qdrant in prod. The agent's semantic memory.

Read the guide →

Share a preview URL

`a2a dev` serves your work-in-progress agent at a public URL — a live, hot-reloading link for testers and stakeholders.

Read the guide →

Develop in VS Code

Open the agent's cloud dev box in VS Code via Remote-SSH — full editor, extensions, terminal, real runtime.

Read the guide →

Develop in Cursor

Point Cursor's Remote-SSH at the live dev box so its AI edits the real repo against the real runtime.

Read the guide →

No local Docker

`a2a dev` runs your agent on a cloud dev box — no local Docker daemon, no Compose, no laptop image builds.

Read the guide →

Offline / air-gapped dev

`a2a dev --local` runs the whole dev loop on your machine — on a plane, behind a firewall, air-gapped.

Read the guide →

AGENTS.md-native

Build with your coding agent

The agent-native SDK

a2a ships its own reference — AGENTS.md, CLAUDE.md, and a build skill — so any coding agent builds against the real SDK.

Read the guide →

Build with Claude Code

a2a init drops a CLAUDE.md and a build-a2a-agent skill in your project, so Claude Code already knows the SDK.

Read the guide →

Build with Codex

AGENTS.md is Codex's native instruction file — and a2a init writes one, so Codex opens fluent in the platform.

Read the guide →

Build with Cursor

Cursor reads AGENTS.md at the repo root; a2a init generates it, so Cursor's Agent knows a2a with no rules to author.

Read the guide →

Vibe-code AI agents

Describe the agent; let your coding agent write it. The scaffold makes vibe-coding produce governed, signed, deployable agents.

Read the guide →

agents of agents

Multi-agent & self-building

Self-building agents

Agents that build and ship other agents — each child a governed deployable with its own identity, grant, and signed proof.

Read the guide →

Agents that build agents

Meta-agents that deploy through the platform API. Discover, don't hard-code; every child an isolated audit boundary.

Read the guide →

Multi-agent orchestration

Orchestrate agents with signed handoffs. Authority attenuates per hop; replay the whole workflow deterministically.

Read the guide →

Agent of agents

The supervisor pattern, governed: delegate scoped authority instead of keys, and sign the whole chain of command.

Read the guide →

Agent swarms

Swarms that stay accountable — per-member identity, scoped grants, signed receipts, and independent scale-to-zero.

Read the guide →

Autonomous agent teams

Teams where each member is its own agent with its own DB, grant, and per-teammate ledger.

Read the guide →

Recursive AI agents

Agents calling agents, bounded: authority shrinks with depth and the whole call tree is hash-chained.

Read the guide →

Spawn sub-agents

Spawn governable sub-agents — least-privilege grant at spawn, signed from run one, clean revocation.

Read the guide →

Agent-to-agent delegation

Delegate authority, not keys: scoped, audience-bound, TTL-limited grants between agents, every issuance signed.

Read the guide →

Self-improving agents

Self-modification with a signed before/after, replay to validate, exact rollback, and authority bounded by a grant.

Read the guide →

governed execution

Security & proof

Log in & authenticate

Every way to sign in: dashboard, passkeys, `a2a login` CLI, API tokens, MCP scopes, and agent grants.

Read the guide →

AI agent audit trail

An audit trail you can actually prove: Ed25519-signed, hash-chained receipts, not mutable logs.

Read the guide →

Scoped grants for agents

Stop giving agents API keys. Bind audience, TTL, file, and tool scope per run. Eliminate keys, don't rotate them.

Read the guide →

AI agent authentication

Short-lived signed tokens and scoped grants — managed auth on deploy, no OAuth to build.

Read the guide →

Observability vs accountability

Observability tells you what happened. Signed receipts prove it — per-run, per-tool-call, per-cost.

Read the guide →

audit-ready

Compliance

SOC 2 for AI agents

Append-only, retained, per-decision evidence — the audit artifacts SOC 2 controls actually ask for.

Read the guide →

EU AI Act Article 12 logging

Automatic logging and a 6-month retention floor by Aug 2 2026. Receipts are a literal answer.

Read the guide →

A2A + MCP

Protocols & interop

What is the A2A protocol?

A developer's guide to Agent2Agent: agent cards, discovery, and how agents call each other.

Read the guide →

A2A vs MCP

The difference, cleared up: MCP is agent-to-tool, A2A is agent-to-agent. You need both.

Read the guide →

Turn OpenAPI into an MCP server

Point a2a at an OpenAPI spec — get a hosted MCP server and agent, tool-per-operation, with auth.

Read the guide →

Connect to Claude & Cursor

Deploy, get an MCP endpoint, paste it into Claude or Cursor. Scoped grants per tool.

Read the guide →

agents that earn

Economy & monetization

The agent economy

Infrastructure, not a token: identity, receipts, payments, and hosting for deployable economic actors.

Read the guide →

AI agent marketplace

Discover, price, and run agents in isolated trial rooms — agents earn per call with proof-of-work receipts.

Read the guide →

AI agent monetization

Pricing models that earn: per-call, subscription, outcome-based, hybrid — with metered billing built in.

Read the guide →

MCP server monetization

11k+ MCP servers, under 5% monetize. Bundle the meter, auth, and payout — charge per call.

Read the guide →

x402 agent payments

Pay-per-call payments explained for agent builders — settled with signed receipts.

Read the guide →

AP2 vs x402

Two agent-payment approaches compared, for builders deciding how agents should pay each other.

Read the guide →

your agent, paid

Get paid for your agents

Get paid for your AI agents

The whole loop, deploy to bank: declare a price, publish, callers pay from prepaid credit, a signed receipt per call, then payout.

Read the guide →

Make money with AI agents

The honest overview of how an agent earns — per-call, subscription, outcome, hybrid — with the billing plumbing already built.

Read the guide →

Sell AI agents

Sell metered access, not source. The agent card is the product page; the marketplace is the storefront; the receipt is the record.

Read the guide →

Sell AI agents online

A storefront you don't build: discoverable marketplace, instant prepaid-credit checkout, signed receipts for trust at a distance.

Read the guide →

Passive income from AI agents

Deploy once, earn per call. Scale-to-zero means idle agents cost nothing, so standby bills don't eat the income.

Read the guide →

AI agent side hustle

Turn a weekend agent into a metered, paid service — no company, no billing stack, no checkout to build first.

Read the guide →

Recurring revenue AI agents

A subscription floor plus metered overage on one meter — predictable MRR with a signed receipt behind every dollar.

Read the guide →

Turn your agent into a business

The full stack a real agent business needs — identity, gated billing, payout, and proof — shipped on deploy.

Read the guide →

AI agent payouts

How earnings become money in your bank: Stripe Connect account, settlement sweep, reversal reconciliation.

Read the guide →

AI agent creator earnings

What you keep per call — author markup minus the platform fee — snapshotted immutably onto each signed receipt.

Read the guide →

AI agent earnings dashboard

The screen: total earned, paid out, owed — per-agent breakdown drilling down to per-run signed receipts.

Read the guide →

Stripe Connect for AI agents

The payout stack you'd otherwise build — Connect onboarding, transfers, reconciliation — managed for you.

Read the guide →

Usage-based billing for agents

Meter every call once against a prepaid balance, so the invoice can never drift from the metric.

Read the guide →

Prepaid credits for agents

The buyer-side wallet: top up once, debit per call, gate callers on balance — built for machine-speed traffic.

Read the guide →

AI agent subscription billing

Recurring access or allotment billing layered over the same per-call meter, with receipts behind every cycle.

Read the guide →

Charge per call for agents

The simplest model: one price-per-call config line; the runtime meters and bills each invocation.

Read the guide →

How to price an AI agent

A practical guide: derive the compute floor, add markup as IP rent, then pick per-call, subscription, or outcome.

Read the guide →

AI agent metering & billing

Metering as the primitive: every call counted once, and that one number drives the bill, the receipt, and the payout.

Read the guide →

AI agent invoicing

The signed receipt as a verifiable invoice — caller, request, result, cost — checkable cryptographically, frozen at run-time price.

Read the guide →

AI agent revenue share

The transparent split: buyer pays gross, platform keeps compute plus a fee on markup, creator takes the rest — provable per call.

Read the guide →

don't trust the agent

Trust the receipt.

Deploy any agent — LangGraph, OpenAI Agents SDK, CrewAI, or custom — with a managed Postgres database, an MCP server, an API, a frontend, and an Ed25519-signed receipt for every run. One deploy, the whole agent app, with proof.

Deploy an agent Compare platforms Plan a pilot