Chains

MAIN CHAINS

BNB Smart Chain

Fast. Affordable. EVM-Compatible

BNB Beacon Chain

Sunset Complete

BNB ecosystem’s staking & governance layer

SHAPE THE CHAIN

Staking

Earn rewards by securing the network

DocumentationFaucetBscScanBSCTraceDocumentationFaucetBridgeopBNBScanDocumentationFaucetBridgeGreenfieldScanDCellarLearn more about FusionToken Recovery ToolBeacon Chain ExplorerNative StakingLiquid Staking

Build

GET STARTED

Submit dApps

Explore

Accelerate

See All Programs

Connect

Join us

Careers🔥

Explore Opportunities on BNB Chain

BNB Chain CareersEcosystem Jobs

BNB Agent Studio is live on BNB Chain: AI agents from one prompt

2026.7.1  •  6 min read
Blog post image.

TL;DR

  • BNB Agent Studio is live on BNB Smart Chain. Builders describe an AI agent in one prompt inside platforms like Cursor, Claude Code and other platforms, and it ships to the chain.
  • Agents pay their own LLM bills from a wallet you fund. They top up automatically via the x402 protocol, so the agent keeps running between your wallet refills.
  • Each agent gets an onchain identity via ERC-8004 and a task interface via ERC-8183. Other agents can find it and call it.
  • Open standards across the stack: x402, ERC-8004, ERC-8183. Nothing about the agent is locked into a single vendor.

If you want to ship a useful AI agent on chain today, you have to assemble at least five things separately: 

  1. A wallet
  2. An identity
  3. A payment rail
  4. A hosting environment
  5. Access to a Large Language Model. 

Each one is a different vendor, a different SDK, a different login. Every seam in that stack is a place something can break, and every dependency is a future migration headache waiting to happen.

Even when that stack is wired up correctly, the agent often doesn't last long. The moment its language model credits run out, it stops. It has no way to refill its own balance. The result is an agent that looks autonomous until it isn't, and you end up checking on it the way someone checks on a houseplant.

BNB Agent Studio is a developer product that takes those pieces and puts them inside one workflow. You describe the agent in one prompt inside Cursor or your favourite vibecoding platform, and the product handles the rest. Your agent comes out of deployment already paying its own bills and already addressable on BNB Smart Chain.

Inside BNB Agent Studio

The whole flow lives inside an AI IDE. Install the bnb CLI, describe what you want the agent to do, and BNB Agent Studio scaffolds the code, sets up the wallet, registers the agent's onchain identity, and deploys it to a managed runtime.

The pieces underneath:

  • bnb CLI - A single-line install. Detects supported AI IDEs (Cursor, Claude Code) and registers Studio's MCP server with them automatically.
  • Studio MCP server - Exposes Studio's tools to the IDE, so the AI assistant can scaffold and deploy agents on your behalf.
  • BNB Chain SDK - Runtime layer with primitives for identity, payments, and language model calls. Available in Python today, with additional languages coming.
  • AWS Bedrock AgentCore - Where your deployed agent runs. Agents live on production-grade infrastructure rather than your laptop.
  • x402 payment protocol - The payment rail an agent uses to top up its own balance.
  • ERC-8004 - Agent identity standard. Each deployed agent gets its own onchain identity.
  • ERC-8183 - Agent task interface standard. Other agents can discover and call the one you've built.

From One Prompt to a Live Agent

The developer experience is meant to look like a normal coding session.

Local testing happens in the same environment. Once you're satisfied, deployment is a single instruction. Behind that, Studio compiles the agent, pushes it to AWS Bedrock AgentCore, registers an ERC-8004 identity for it, binds its wallet to that identity, registers the ERC-8183 task interface, and turns on the self-funding loop.

Self-funding is the part that does the heavy lifting once your agent is live. Your agent watches its own language model balance. When the balance falls below a threshold, the SDK initiates a payment over x402, drawing from the wallet you funded at deployment and settled in $U on BNB Smart Chain. Your agent stays online during the top-up. When the underlying wallet runs low, you refill it.

Less Plumbing, More Agent

What changes for you is mostly what you no longer have to build. Wallets, identity, payments, and hosting are part of the product. The SDK ships in Python today, with more languages coming. The CLI works in any MCP-compatible AI IDE, which means your existing Cursor and Claude Code workflows pick up Studio without any extra setup.

Three things to know about how this actually works:

  1. Your agent handles its own credit top-ups, drawing from the wallet you fund. That changes the operational model. An agent you deploy today keeps running without you intervening between tasks. You only step in when the underlying wallet runs low.
  2. Open standards are the default. ERC-8004, ERC-8183, and x402 are all open. Nothing about your agent's identity, payments, or task interface is locked to BNB Chain or to Studio.
  3. Studio doesn't get in the way of how you build. The user flow above is a reference shape, not a contract. The SDK and CLI work fine if you want to wire things up differently.

To get started, install the CLI from studio.bnbchain.org/install and read the quickstart in the docs.

Live Today on BNB Chain

BNB Agent Studio is live on BNB Smart Chain mainnet. The CLI is publicly available, the SDK installs via pip, and the supported AI IDEs at launch are Cursor and Claude Code, with additional MCP-compatible environments coming through MCP itself.

AWS Free Tier: A Limited-Time Free Experience

To support developers, the BNB Agent Studio AWS Free Tier offers a no-cost way to try the full cloud agent deployment pipeline using only a GitHub login with no AWS account or credit card required.

From there you get access to agent creation, cloud deployment, onchain registration via ERC-8004, and ERC-8183 / x402 onchain interactions, all without setting up your own infrastructure first.

This is a limited-time campaign with a capped budget, so access closes once it's gone. Here's everything you need to know about how it works.

Eligibility

Dimension

Rule

GitHub account

Must be registered for at least 30 days.

Participation limit

Each GitHub account may participate once only. No re-enrollment.

Network

bsc-testnet only. Mainnet is not supported.

Total campaign budget

Limited on a first come, first served basis. Campaign ends automatically once the budget is exhausted.

Timing

Dimension

Rule

Timer start

First successful bag deploy (not at login).

Trial duration

48 hours.

Expiry reminders

CLI reminders at 12 hours and 1 hour before expiry.

On expiry

All cloud resources automatically and permanently deleted. Cannot be recovered.

Login without deploying

Does not count toward the 48 hours. No resources consumed.

Usage limits

Limit

Default Value

Invocation rate

60 requests / minute

Concurrent invocations

Max 2

Max single session duration

5 minutes (forcibly terminated on timeout)

Idle auto-reclaim

Reclaimed after 1 minute of inactivity

Max agents per user

10

Deployments per hour

Max 10

Concurrent in-flight deployments

Max 3

Zip package size

≤250 MB

Container image size

≤2 GB

Single request size

Max 10 MB

Campaign End Mechanism

When total campaign spend reaches the budget threshold ($3,000 × 80% ≈ $2,400), campaign end is automatically triggered:

  • New user entry is immediately closed: Deploy and invoke requests return "Campaign has ended."
  • Users already inside their 48-hour window are unaffected and continue running until their individual expiry.

Note: AWS billing has a delay of roughly 24 hours, so the actual trigger point is based on estimated spend reaching the threshold (subject to adjustment). Final spend is determined by actual AWS billing.

Data After Expiry

When the 48-hour trial ends, all cloud resources are immediately and permanently deleted and cannot be recovered. Onchain identity and transaction records are permanently retained and unaffected.

Content

Status After 48 Hours

Local code files (main.py, etc.)

Fully retained (stored locally on user's machine)

Onchain identity (ERC-8004)

Permanently retained (on BSC)

Onchain transaction records

Permanently retained (on BSC)

Local wallet files

Retained (stored locally on user's machine)

Cloud AgentCore runtime instance

Permanently deleted

CloudWatch runtime logs

Permanently deleted

ECR images (runtime + user-uploaded)

Permanently deleted

Secrets (wallet keys, etc.)

Immediately and permanently deleted. No recovery window.

S3 code bundle

Permanently deleted

Important:

  • Always use a testnet-dedicated wallet. Never use a mainnet wallet holding real assets.
  • The system sends CLI reminders before expiry. Back up your local code in time.

What’s Next?

The current shape of the product is the developer surface for shipping individual agents. Further work covers ecosystem features and additional wallet and language model integrations. None of that is required for you to start building today.

Here’s what’s to come:

Late June 2026 - more options at every step:

  • TWAK wallet integration as an additional wallet option.
  • BinancePay B402 merchants integration, so agents can purchase CMC data through x402.
  • Microsoft Azure as a cloud runtime option alongside AWS.
  • Free AWS runtime for up to 48 hours.

Mid July 2026 - more choice and control:

  • Developer dashboard to view, pause, and restart agents without touching the CLI.
  • Enterprise-grade security model for agent wallet private keys.
  • Additional wallets to pick from and more data services your agent can pay to use.

The roadmap updates regularly as we ship. For the latest, see the BNB Agent Studio page:

bnbchain.org/en/bnb-agent-studio 

BNB Chain has been pointing toward a chain where autonomous software does real work for some time. BNB Agent Studio is the developer surface that makes that workable. The product is built for you if you want to ship agents that actually do the work, not chatbots that need supervision.

To start, install the CLI:

pip install bnbagent-studio

Or read the quickstart in the docs.

Share