Skip to content
Beam Docs

Deploy Studio on AWS

Launch a dedicated Beam Studio in your own AWS account with one click: the EC2 instance, its disks and network are yours, and Studio installs itself.

One click creates a dedicated EC2 instance in your AWS account and installs Beam Studio on it with the standard installer. You own and pay for everything it creates; Beam hosts nothing for it. Your transfers keep using the Beam network directly: nothing in the deployment proxies or stores your data.

Deploy Beam Studio onAmazon Web ServicesOne click · runs in your AWS accountDeploy

Deploy to AWS

The button opens CloudFormation's Quick Create page with the template loaded, so you can review every resource before you create it. Before creating the stack, tick the acknowledgments CloudFormation shows: the template creates an IAM role and uses a CloudFormation transform.


What gets deployed

ResourceDetails
InstanceEC2, Ubuntu 24.04, IMDSv2 only
Studio dataIts own encrypted EBS volume
NetworkA dedicated VPC, or a VPC and subnet you choose
ManagementAWS Systems Manager Session Manager (no SSH)
Cloud permissionsAn instance role that only allows Session Manager
Default sizet3.large (2 vCPU, 8 GiB)
Approximate costabout $72 a month

The cost is the on-demand list price in us-east-1, including the data volume and a public IPv4 address. Studio's own data transfer is small. The minimum size is t3.medium (2 vCPU, 4 GiB).

Studio's internal services are never reachable from the internet:

  • its web port (3004);
  • its API (8787);
  • its MCP server (8766);
  • PostgreSQL, NATS and Docker.

Choose how people reach Studio

  • Private (default): leave Studio hostname empty. No inbound port is open, and you open Studio through two Session Manager tunnels. You need the AWS CLI with the Session Manager plugin on your computer.
  • HTTPS: enter a DNS name you control as Studio hostname, such as studio.example.com. Point it at the Elastic IP shown in the stack outputs, or give a Route 53 hosted zone and the record is created for you. Studio is served at https://<name> with an automatic Let's Encrypt certificate. Only port 443 is open.

In private mode, features that need Studio to be reachable from the internet are not available:

  • webhook triggers;
  • Room transfers between an agent and a bucket member;
  • Box sign-in beyond localhost.

The stack finishes when Studio is installed and healthy. It never waits for DNS or the certificate; sudo beam-studio-cloud status on the instance shows the HTTPS state separately.

Open and claim Studio

A new Studio serves nobody until your organization claims it:

  1. Open Studio. The URL is in the stack outputs. In private mode, start the two tunnels from the outputs first, then open http://localhost:3004.
  2. Choose Continue with Beam Auth and approve the code at auth.b1m.ai/device.
  3. Read the claim code on the instance: open SessionManagerUrl from the outputs and run sudo beam-updater claim-code.
  4. In Studio, go to Settings → Access, choose your organization, enter the code, and approve Studio's instance key.

Only someone who can run commands on the instance can read the code.

Connect an AI agent

Studio's MCP server lets agents such as Claude Code, Cursor or VS Code list and run your workflows. It is not exposed publicly unless you enable it for one address range.

  1. Create a token in Studio at MCP → Tokens → New token. Choose the smallest preset that works and an expiry.

  2. Reach the endpoint:

    • Any deployment: a Session Manager tunnel to port 8766. The command is in the McpEndpoint output. Then use http://localhost:8766/mcp.
    • HTTPS deployments: you can instead enable remote MCP for your agents' address range (Remote MCP over HTTPS and Address range allowed to reach MCP), and use https://<studio-host>:8443/mcp.
  3. Add it to your agent, for example in Claude Code:

    claude mcp add --transport http beam-studio <endpoint> \
      --header "Authorization: Bearer ${BEAM_STUDIO_MCP_TOKEN}"

    On the instance, sudo beam-studio-cloud mcp-config prints ready-to-use configuration for Claude Code, Cursor and VS Code. sudo beam-studio-cloud mcp-check verifies a token with a real read-only tool call.

Clients that only support OAuth-based connectors, such as custom connectors in claude.ai or Claude Desktop, are not supported.

Update, stop and resize

  • Studio updates itself from signed releases when an admin applies them in Settings. Each update backs up the database first and rolls back if Studio does not come back healthy.
  • Ubuntu installs security updates automatically.
  • Stop or start the instance in the EC2 console. Studio starts with the instance.
  • Resize in the EC2 console: stop the instance, use Change instance type, then start it.

Back up and restore

The data volume holds everything Studio needs: the database, its secrets and configuration, and every container volume.

  • Back up by taking an EBS snapshot of the data volume (DataVolumeId in the outputs).
  • Restore by creating a new stack with the snapshot as Restore from EBS snapshot ID. The restored Studio keeps its claim, instance key, MCP tokens, credentials, workflows and schedules.

Never run the original and a restored Studio at the same time: they share one instance key and would both run your schedules. Stop or delete the original first.

Delete the deployment

  1. In Session Manager, run sudo beam-updater uninstall to revoke Studio's Beam instance key. If the instance is already gone, revoke Studio: ‹host› under API Keys in the Console.
  2. Delete the stack in the CloudFormation console. What happens to Studio's data depends on the Keep a snapshot when the stack is deleted setting:
    • Snapshot (the default): a final snapshot of the data volume is kept, so you can restore Studio later. Delete it under EC2 → Snapshots when you no longer need it.
    • Delete: the data is deleted with the stack. To uninstall this way, first update the stack with the setting changed to Delete, then delete it.

Snapshots you keep and DNS records you created at your own provider remain until you delete them.


Next steps

Decentralized distributed bandwidth infrastructure.

On this page